Professional Documents
Culture Documents
2020 Book ComputationalGeometryTopologyA
2020 Book ComputationalGeometryTopologyA
James F. Peters
Computational
Geometry, Topology
and Physics of
Digital Images with
Applications
Shape Complexes, Optical Vortex Nerves
and Proximities
Intelligent Systems Reference Library
Volume 162
Series Editors
Janusz Kacprzyk, Polish Academy of Sciences, Warsaw, Poland
Computational Geometry,
Topology and Physics
of Digital Images
with Applications
Shape Complexes, Optical Vortex Nerves
and Proximities
123
James F. Peters
Department of Electrical and Computer
Engineering, Engineering and Information
Technology Complex
University of Manitoba
Winnipeg, MB, Canada
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
This book is dedicated to
Somashekhar (Som) Naimpally, 1931–2014
and Anna Di Concilio, Amma, and sweet P
for offering many glimpses of the proximities
of surface shapes.
Preface
This book introduces the computational geometry, topology and physics of digital
images and video frame sequences. It is the geometry of mesh generation by
Edelsbrunner [1] and the geometry of polytopes by Ziegler [2] that provide a solid
basis for the computational geometry approach to the study of geometric struc-
tures that infuse triangulated visual scenes explored in this monograph. A planar
polytope is a filled polygon defined by the intersection of closed half-planes
covering the interior of the polygon. In addition, an introduction to computational
geometry in this monograph can be found on the geometric foundations of com-
puter vision by Peters [3]. It is the topology of cellular complexes introduced by
Alexandroff [4, 5], (beautifully extended and elaborated by Cooke and Finney [6]),
Borsuk [7, 8, 9], a recent formulation of this topology by Edelsbrunner and Harer
[10] and the work on persistence homology by Munch [11] that provide a solid
basis for an introductory study of the computational topology of visual scenes.
This form of topology explores the fabric, shapes and structures typically found in
visual scenes. Coupled with the inherent geometry and topology of visual scenes,
there is the computational physics arising from the structures and events recorded in
videos and the concomitant sensitivity concerning the fine structure of light to
consider. The fine structure of light and light caustics that we have in mind are
introduced by Nye [12]. A consideration of light caustics brings into play catas-
trophe theory and the appearance of light caustic folds and cusps, which leads to the
introduction of optical vortex nerves in triangulated digital images. In this context,
computational physics is synonymous with the study of the structure of light
choreographed in video frames. This choreography of the structure of light appears
as a sequence of snapshots of light reflected and refracted from surface shapes that
provides a solid basis for the study of the structures and shapes that appear in visual
scenes.
The study of the persistence of image object shapes in sequences of video frames
as well as sequences of photographs that record surface shape changes in a visual
scene, is important. Surface shapes tend to appear, undergo a change in the varying
ix
x Preface
light and surface conditions, and eventually disappear. The familiar tendency to
look for unusual appearances objects (both natural and artificial) in visual scenes is
a tacit recognition of continuous change and the momentary persistence of observed
components in visual scenes. In other words, it is important to take into account the
spacetime character of visual scene shapes. By this, I mean an understanding of
visual scenes includes not only a study of the geometry and topology of visual
scenes but also a consideration of the physics of light, the character and energy
of the photons colliding with curved surfaces in a visual scene.
Physics enters into the picture in cases where we take into account the
description of surface shapes and the light reflected from surface shapes recorded in
photos and, especially, in video frames. Computer engineering also enters in the
picture here with the study of photonics and reflected light-capturing devices. In
terms of the physics of digital images, the shape-shifting character of energy is
important. For more about this view of energy, see Susskind [13, x7, p. 126].
Computational geometry facilitates the capture of fine-grained structures
embedded in image object shapes. And computational topology enables the capture
and analyses of the proximities found in cellular complexes (collections of vertices,
line segments, filled triangles, cycles, vortexes, nerves) embedded in the geometry of
triangulated visual scenes (see, e.g., Peters [14, 15]). It is the homology of cell
complexes (an offspring of Alexandroff’s approach to topology [4]) that is an
important component, here. Homology is a mathematical framework that focuses on
how space is connected, utilizing algebraic structures such as groups and maps that
relate topologically meaningful subsets of a space to each other [10, xIV.1, p. 79].
A group G is a nonempty set equipped with a binary operation that is asso-
ciative and in which there is an identity element e and every member a in G has an
inverse b, i.e., a b ¼ e . A cyclic group H is a group in which every member of
G can be written as a positive integral power of a single element called a generator.
A cyclic group is Abelian, provided a b ¼ b a, for every pair elements in G. A
free abelian group is an AbelianP group with multiple generators, i.e., every element
of the group can be written as gi a for generators gi in G. For a good introduction
i
to cyclic groups from a homology perspective, see Giblin [16, A.1, p. 216].
In practical terms, homology is a source of insights into how the pieces of a visual
scene can be connected to each other. Cyclic groups are useful in representing in a
concise way how the pieces of a visual scene that are attached to each other and
connected together. Cyclic groups with multiple generators are also a source of an
important feature in earmarking surface shapes of interest, namely, Betti numbers
(counts of the number of generators in a free Abelian group). H. Poincaré named
such numbers in honour of Enrico Betti, based on the paper [17]. The focus in
computational geometry, topology and physics of digital images is on finite spaces.
Betti observes that a finite space has properties independent of the size of its
dimensions, and from the shape of its elements. These properties refer to it only to
the way of connection of its parts… [17, x3, p. 143]. The properties of a finite
bounded spatial region tend to be revealed by the path-connected vertices in a cell
complex covering the space. (see, for example, the path-connected vortices
Preface xi
covering bounded region occupied by the surface shape in Fig. 1). For more about
this, see Tucker and Bailey [18], Salepci and Welshinger [19] and Pranav and
Edelsbrunner and van de Weygaert and Vegter [20].
A thorough study of a computational approach to homology is given by
Kaczynski, Mischaikov and Mrozek [21]. The focus here is discerning and tracking,
analyzing and representing, and approximating the closeness of shifting surface
shapes. To grapple with continual shape shifting from one visual scene (one video
frame to another one), feature vectors on descriptive proximity spaces, provide us
with a means of representing shape changes that are either close or sometimes far
apart. For more about this, see Di Concilio, Guadagni, Peters and Ramanna [22].
Fig. 1 Nesting,
non-overlapping vortexes
covering a shape
In the Euclidean plane, these geometric structures are vertices, line segments,
and filled triangles (3-sided polytopes). A polytope is an intersection of closed half
planes [2]. An individual polytope is a spatial region with a filled interior bounded
on all sides. In a topological setting, the focus is on the decomposition of regions of
visual scenes into very simple polytopes such as filled triangles that are easily
measured and analyzed. The basic ingredients of this topology are simplicial
complexes, shape theory and persistent homology.
The secret underlying this work is the decomposition of closed digital image
regions into sets of shape complexes that provide a basis for shape analysis. A shape
complex is a covering of a shape with a collection of nesting, usually
non-overlapping vortexes (see, for example, Fig. 1). A sample decomposition as a
partial triangulation of a Napoli flower display1 in Fig. 2a is shown in Fig. 2b. The
1
Many thanks to Arturo Tozzi for this photo.
xii Preface
end result is a covering of image scene shapes such as flower petals with collections
of filled triangles (simplicial complexes). Collections of triangles forming nerve
structures in each complex have a common vertex. For example, the complexes
covering the white flowers in Fig. 2b give us a means of measuring, comparing,
describing and classifying scene segments occupied by the flower petals. In general,
the objectives of shape analysis are to classify, compare, quantify similarities and
differences, and measure the distances between shapes [23]. In this book, the focus is
on the application of computational topology in the shape analysis of visual scenes.
With a cell complex on a triangulated visual scene, the scene shapes are covered
with clusters of filled triangles called nerve complexes. Let K be a finite collection
of sets of points. A nerve of the collection of sets K (denoted by Nrv K) consists of
all nonempty subcollections of K that have nonvoid intersection [10, xIII.2, p. 59].
Each nerve has its own distinctive shape. For example, the nerve covering part
of the flowers in Fig. 2b gets its shape from the filled triangles satelliting round a
single vertex. On a triangulated surface, an Alexandroff nerve complex A (denoted
by Nrv A) is a collection of triangles with a common vertex [4, x33, p. 39] (see, for
example, Fig. 4)
This book not only introduces the basics of a computational geometry, topology
and physics of digital images, but also gives a number of practical applications. The
applications include
Ap.1 Cellular division trails: Sect. 3.13, Application 3.13.
Ap.2 Maximal barycentric star nerves: Sect. 4.3, Application 4.3.
Preface xiii
References
1. Edelsbrunner, H.: Geometry and topology for mesh generation. In: Cambridge Monographs
on Applied and Computational Mathematics, vol. 7, pp. xii+177. Cambridge University Press,
Cambridge, UK (2001), Zbl 1039.55001
2. Ziegler, G.: Lectures on polytopes, In: Graduate Texts in Mathematics, vol. 152. pp. x+370.
Springer, New York (1995). ISBN: 0-387-94365-X, MR1311028
3. Peters, J.: Foundations of computer vision. In: Computational Geometry, Visual Image
Structures and Object Shape Detection, Intelligent Systems Reference Library, vol. 124,
pp. i-xvii, 432. Springer International Publishing, Switzerland (2017). https://doi.org/10.1007/
978-3-319-52483-2, Zbl 06882588 and MR3768717
4. Alexandroff, P.: Elementary concepts of topology, 63pp., Dover Publications, Inc., New York
(1965). In: Translation of Einfachste Grundbegriffe der Topologie, Springer, Berlin (1932),
translated by Alan E. Farley, Preface by D. Hilbert, MR0149463
5. Alexandroff, P.: Über den algemeinen dimensionsbegriff und seine beziehungen zur ele-
mentaren geometrischen anschauung. Math. Ann. 98, 634 (1928)
6. Cooke, G., Finney, R.: Homology of cell complexes. In: N.E. Steenrod, (ed.) Based on
lectures by Princeton University Press and University of Tokyo Press, Princeton, N.J., USA;
Tokyo, Japan, pp. xv+256 (1967), MR0219059
7. Borsuk, K.: On the imbedding of systems of compacta in simplicial complexes. Fund. Math.
35, 217–234 (1948), MR0028019
8. Borsuk, K.: Theory of shape. Monografie matematyczne, Tom 59. In: Mathematical
Monographs, vol. 59, PWN—Polish Scientific Publishers (1975), MR0418088. Based on K.
Borsuk, Theory of shape, Lecture Notes Series, No. 28, Matematisk Institut, Aarhus
Universitet, Aarhus (1971), MR0293602
9. Borsuk, K., Dydak, J.: What is the theory of shape? Bull. Austral. Math. Soc. 22(2), 161–198
(1980), MR0598690
10. Edelsbrunner, H., Harer, J.: Computational topology. An introduction. Amer. Math. Soc.,
Providence, RI pp. xii+241 (2010). ISBN: 978-0-8218-4925-5, MR2572029
11. Munch, E.: Applications of persistent homology to time varying systems. Ph.D. thesis, Duke
University, Department of Mathematics (2013). Supervisor: J. Harer, MR3153181
12. Nye, J.: Natural focusing and fine structure of light. Caustics and dislocations, pp. xii+328.
Institute of Physics Publishing, Bristol (1999), MR1684422
13. Susskind, L.: The Black Hole War, pp. 470. Back Bay Books, New York, NY, USA (2008)
14. Peters, J.: Proximal planar shape signatures. Homology nerves and descriptive proximity.
Advan. Math: Sci. J. 6(2), 71–85 (2017), Zbl 06855051
15. Peters, J.: Proximal planar shapes. Correspondence between triangulated shapes and nerve
complexes. Bull. Allahabad Math. Soc. 33, 113–137 (2018), MR3793556, Zbl 06937935,
Review by D. Leseberg (Berlin)
16. Giblin, P.: Graphs, Surfaces and Homology, pp. Xx+251, 3rd edn. Cambridge University
Press, Cambridge, GB (2016). ISBN: 978-0-521-15405-5, MR2722281, first edition in 1981,
MR0643363
17. Betti, E.: Sopra gli spazi di un numero qualunque di dimensioni [italian]: above the spaces of
any number of dimensions. Annali di Matematica Pura ed Applicata 4(1), 140–158 (1870)
18. Tucker, W., Bailey, H.: Topol. Sci. Am. 182(1), 18–25 (1950). http://www.jstor.org/stable/
24967355
19. Salepci, N., Welshinger, J.Y.: Tilings, packings and expected betti numbers in simplicial
complexes. arXiv 1806(05084v1), 1–28 (2018)
20. Pranav, P., Edelsbrunner, H., van de Weygaert, R., Vegter, G.: The topology of the cosmic
web in terms of persistent betti numbers. Mon. Not. R. Astron. Soc. 1–31 (2016). https://
www.researchgate.net
xvi Preface
21. Kaczynski, T., Mischaikov, K., Mrozek, M.: Computational homology, Appl. Math. Sci. 157,
pp. xvii+480. Springer, New York, NY (2004). ISBN 0-387-40853-3/hbk, Zbl 1039.55001
22. Concilio, A.D., Guadagni, C., Peters, J., Ramanna, S.: Descriptive proximities. Properties and
interplay between classical proximities and overlap. Math. Comput. Sci. 12(1), 91–106
(2018), MR3767897, Zbl 06972895
23. Zeng, W., Gu, X.: Ricci flow for shape analysis and surface registration. theories, algorithms
and applications, pp. Xii+139. Springer, Heidelberg (2013). ISBN: 978-1-4614-8780-7,
MR3136003
24. Beer, G., Di Concilio, A., Di Maio, G., Naimpally, S., Pareek, C., Peters, J.: Somashekhar
naimpally, 1931–2014. Topol. Appl. 188, 97–109 (2015). http://doi.org/10.1016/j.topol.2015.
03.010, MR3339114
25. Peters, J., Naimpally, S.: Applications of near sets. Notices Am. Math. Soc. 59(4), 536–542
(2012). http://doi.org/10.1090/noti817, MR2951956
26. Concilio, A.D., Guadagni, C.: Bornological convergences and local proximity spaces. Topol.
Appl. 173, 294–307 (2014), MR3227224
27. Concilio, A.D., Guadagni, C., Peters, J., Ramanna, S.: Descriptive proximities i: properties
and interplay between classical proximities and overlap. Math. Comput. Sci. 12(1), 91–106
(2018), ArXiv 1609.06246v1, MR3767897
28. Peters, J., Guadagni, C.: Strongly proximal continuity & strong connectedness. Topol. Appl.
204, 41–50 (2016), ArXiv 1504.02740, MR3482701
29. Guadagni, C.: Bornological convergences on local proximity spaces and xl- metric spaces.
Ph.D. thesis, pp. 79. Università degli Studi di Salerno, Salerno, Italy (2015). Supervisor: A. Di
Concilio
30. Concilio, A.D., Gerla, G.: Quasi-metric spaces and point-free geometry. Math. Struct.
Comput. Sci. 16(1), 115–137 (2006), MR2220893
31. Peters, J., Tozzi, A.: Quantum entanglement on a hypersphere. Int. J. Theor. Phys. 55(8),
3689–3696 (2016), Zbl 1361.81025, MR3518899
32. Worsley, A.: Harmonic quintessence and the derivation of the charge and mass of the electron
and the proton and quark masses. Phys. Essays 24(2), 240–253 (2011). https://doi.org/10.
4006/1.3567418
33. Worsley, A.: The formulation of harmonic quintessence and a fundamental energy equiva-
lence equation. Phys. Essays 23(2), 311–319 (2010). https://doi.org/10.4006/1.3392799
34. Worsley, A., Peters, J.: Enhanced derivation of the electron magnetic moment anomaly from
the electron charge from geometric principles. Appl. Phys. Res. 10(6), 24–28 (2018). https://
doi.org/10.5539/apr.v10n6p24
35. Yurkin, Peters, J., Tozzi, A.: A novel belt model of the atom, compatible with quantum
dynamics. J. Sci. Eng. Res. 5(7), 413–419 (2018)
36. İnan, E.: Approximately groups in proximal relator spaces. Commun. Fac. Sci. Univ.
Ank. Ser. A1. Math. Stat. 68(1), 572–582 (2019), MR3827537
37. Peters, J., İnan, E.: Strongly proximal edelsbrunner-harer nerves. Proc. Jangjeon Math.
Soc. 19(3), 563–582 (2016), MR3618825
38. İnan, E., Öztürk, M.: Near groups on nearness approximation spaces. Hacettepe J. Math. Stat.
41(4), 545–558 (2012), MR3060371, MR3241196
39. Peters, J., İnan M.A. Öztürk, E.: Monoids in proximal banach spaces. Int. J. Algebra 8(18),
869–872 (2014)
40. Peters, J., İnan M.A. Öztürk, E.: Spatial and descriptive isometries in proximity spaces.
Gen. Math. Notes 21(2), 125–134 (2014)
41. Öztürk, M., Uçkun, M., İnan, E.: Near groups of weak cosets on nearness approximation
spaces. Fund. Inform. 133(4), 443–448 (2014), MR3285076
42. Peters, J., Öztürk, M.A., Uçkun M.: Exactness of homomorphisms on proximal groupoids.
Fen Bilimleri Dergisi X(X), 1–14 (2014)
43. Peters, J., Öztürk, M., Uçkun, M.: Klee-phelps convex groupoids. arXiv 1411(0934),
1–5 (2014). Published in Mathematica Slovaca 67 (2017), no. 2.397–400
Preface xvii
44. Öztürk, M., İnan, E., Tekin, O., Peters, J.: Fuzzy proximal relator spaces. Neural Comput.
Appl. (2018). https://doi.org/10.1007/s00521-017-3268-1
45. Zaka, O., Peters, J.: Isomorphic-dilations of the skew-fields constructed over parallel lines in
the desargues affine plane. arXiv 1904(01469), 1–15 (2019)
46. Pawlak, Z.: Classification of objects by means of attributes. Pol. Acad. Sci. PAS 429 (1981)
47. Orłowska, E., Peters, J., Rozenberg, G., Skowron, A.: In memory of professor zdzisław
pawlak. Fund. Inform. 75(1–4), vii–viii (2007), MR2293685
48. Peters, J.: How near are Zdzisław Pawlak’s paintings? Study of merotopic distances between
digital picture regions-of-interest. In: A. Skowron, Z. Suraj (eds.) Rough Sets and Intelligent
Systems, pp. 89–114. Springer (2012)
49. Peters, J., Skowron, A., Stepaniuk, J.: Nearness of visual objects. Application of rough sets in
proximity spaces. Fundam. Inf. 128(1–2), 159–176 (2013), MR3154898
50. Peters, J., Skowron, A., Stepaniuk, J.: Nearness of objects: Extension of approximation space
model. Fundam. Inf. 79(3–4), 497–512 (2007), MR2346263
51. Wolski, M.: Toward foundations of near sets: (pre-)sheaf theoretic approach. Math. Comput.
Sci. 7(1), 125–136 (2013), MR3043923
52. Dochviri, I.: On submaximality of bitopological saces. Kochi J. Math. 5, 121–128 (2010),
MR2656713, Zbl 1354.54027
53. Dochviri, I., Peters, J.: Topological sorting of finitely many near sets. Math. Comput. Sci. 10
(2), 273–277 (2016), Zbl 1345.54020, MR3507604
54. Peters, J., Dutta, H.: Equivalence of planar ˘Cech nerves and complexes. Natl. Acad. Sci. Lett.
(2019). https://doi.org/10.1007/s40009-019-0790-y, ISSN 2250-1754
55. Henry, C.: Near sets: Theory and applications. Ph.D. thesis, Univ. of Manitoba, Dept. Elec.
Comp. Engg. (2010). http://130.179.231.200/cilab/. Supervisor: J.F. Peters
56. Henry, C., Ramanna, S.: Signature-based perceptual nearness: application of near sets to
image retrieval. Math. Comput. Sci. 7(1), 71–85 (2013), MR3043919
57. Henry, C., Ramanna, S.: Quantifying nearness in visual spaces. Cybern. Syst. 44(1), 38–56
(2013)
Contents
xix
xx Contents
Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
Subject Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
List of Tables
xxv
Chapter 1
Computational Geometry, Topology
and Physics of Visual Scenes
Abstract This chapter introduces the computational geometry, topology and physics
of visual scenes. In all three cases, the mathematics is supplemented with algorithms
to provided a basis for the analysis, comparison and classification of physical shapes
found in visual scene snapshots. Computation geometry is a geometry equipped
with a hefty set of step-by-step methods that lifts classical forms of geometry to
a level that is practical in extracting useful information from physical shapes tiled
with polygons on visual scenes. The handmaiden of computational geometry is an
algorithmic form topology. Computational topology combines step-by-step meth-
ods (algorithms) useful in establishing the nearness or apartness of nonempty sets
of cell complexes. A natural approach [1] is to construct a cell complex connected
to seed points (vertices), which serves as a proxy for the underlying shapes in a
digital image. Such a cell complex (connected vertices, edges and filled triangles)
is constructed so that it covers an image and makes it possible to detect, compare,
analyze and classify the proximities of image shapes in terms of the cell complexes
covering the shapes. The basic approach is to extract the inherent structures of image
shapes that would otherwise be hidden or, at least, escape our attention in a casual
visual scan of digital images. Computational physics is an algorithmic approach
to physics. This form of physics enters into the picture in considering methods of
determining the wavelengths of picture elements (pixels) in the visible portion of the
electromagnetic spectrum, particle characteristics of pixels such as energy and hue
angles, saturation and value, which lurk in sequences of video frames.
1.1 Introduction
The story starts with cell complexes drawn on visual scene images to give us means of
approximating surface shapes. In the plane, a cell complex is a collection of vertices
and edges attached to each other. What we call a triangle is a natural outcome of
edges attached to each other. The topology part of this study focuses on a planar
view of the nearness of cell complexes, which are sets of path-connected vertices.
The whole idea is to find ways to attach edges to vertices in such a way that they
Fig. 1.1 Yellow vortex on a green nucleus polygon on a tessellated visual scene
(a) (b)
form simple closed curves that cover and overlap the boundaries of surface shapes
(see, for example, the cell complexes in Figs. 1.1 and 1.2).
A closed curve is a curve with no endpoints and completely encloses a planar
surface region [2]. A simple closed curve is a closed curve with no loops (self-
intersections). A filled simple closed curve is a simple closed curve with nonempty
interior that may or may not have holes in it. Two examples of example of cell
complexes that are filled simple closed curves covering familiar shapes are shown in
Fig. 1.2. Every simple closed curve satisfies the Jordan curve theorem.
Theorem 1.1 (Jordan Curve Theorem [3]) A simple closed curve lying on the plane
divides the plane into two regions and forms their common boundary.
Lemma 1.2 ([4, p. 2]) A finite planar shape contour separates the plane into two
distinct regions.
The path-connected vertices form a simple closed curve in a collection of such
curves surrounding any planar shape contour form what is known as a planar vortex
complex. From Lemma 1.2, we obtain the following result.
1.1 Introduction 3
Theorem 1.3 ([4, p. 2]) A finite planar vortex complex is a collection of non-
concentric, nesting shapes on a surface shape.
Links to Geometry, Topology and Physics. The physics that we have in mind
can be found traditional approaches to electromagnetic systems such as that
found in Baldomir and Hammond [5], Zangwill [6] and in optical caustics found
in Nye [7]. The study of electromagnetism focuses on the electromagnetic
field and its interaction with matter [6, Sect. 2.5.1, p. 46f]. An electromagnetic
field is a physical field produced by electrically charged objects. Geometry,
topology and physics have long and illustrious histories that give evidence
of the importance of considering the geometry and topology of the interplay
between physical structures illuminated by crowds of photons with different
wavelengths bombarding physical surfaces and overlapping surface-covering
cell complexes (see, e.g., Nakahara [8], who considers the interplay between
quantum physics and homology groups (algebraic topology), including finitely
generated cyclic groups and free Abelian groups derived from cell complexes).
Evidence of the importance of geometry in physics can be found, for example,
in Worsley and Peters [9]. “
A planar finite bounded region is tessellated by covering the region with non-
overlapping polygons. The collection of polygons in a tessellation is example of
a cell complex.
Tessellation Polygons are Filled Polygons. And each polygon edge on a tes-
sellated region is part of a half plane that covers the interior of the polygon. In
effect, each tessellation polygon is the intersection of a collection of closed
half planes. A closed half plane includes its bounding edge. This means that
each tessellation polygon is a filled polygon such as a filled octagon ‘. For
more about this, see Sect. 1.3. “
A very simple cell complex is a collection of connected cells such as vertices with
line segments attached between them. Cell complexes are used to cover unknown
physical shapes with irregular shapes with geometric structures that have shapes that
can be measured. This approach solves the problem of measuring elusive physical
shapes with irregular contours and very complicated interiors that occupy typical
visual scene surfaces. In effect, with the mixture of geometry and topology combined
with algorithms, we obtain a tractable view of recorded physical shapes that we can
analyze, approximate, measure, compare, cluster together, and classify.
That is, geometry and topology are infused with computational methods, resulting
in the possibility of repeatable experiments with various forms of captured visual
scenes. A selection of initial symbols used in this introduction to the geometry, topol-
ogy and physics of mesh overlays on visual scene images is given in Table 1.1. These
symbols provide a form of shorthand and means of highlighting something unusual
about what are commonly known as surface shapes recorded in single snapshots
and in video frames. For example, we write sh A instead of writing shape A to call
attention to the unusual character of shapes found in subregions of images covered
by mesh overlays.
Sample tiling of X
X⊆ tilei .
i≥1
6 1 Computational Geometry, Topology and Physics of Visual Scenes
filled
Fig. 1.3 Tiling with
polygons: X ⊆ tile i
i≥1
1 Many thanks to Enze Cui for the tessellated video that is the source of the video frame in Fig. 1.1.
1.2 Tessellated Planar Finite Bounded Regions 7
e.g., Boyer [13]). Descartes’s geometric view of visible space marks the birth of
computational geometry.
Example 1.7 (Cartesian Tiling of the Nighttime Sky) A sample tiling of the visi-
ble nighttime sky by Descartes is shown in Fig. 1.4. The sides of the polygons in
this tiling are formed relative to the positions of the planets surrounding the sun
labelled S in Descartes’ drawing. Descartes viewed each heavenly body as a vortex,
pulling neighboring bodies towards themselves. The circular, dotted lines represent
gravitational waves [14], which transport radiant energy as gravitational radiation
in spacetime, first proposed by H. Poincaré in 1905 and predicted by A. Einstein in
1916. “
(a) (b)
reach) that classical general topology (see, e.g., Willard [16] and its applications in
Naimpally and Peters [17]) that depends on the intersection as well as the union
of open sets. CW topology is the study of the nearness (overlap) of finite, closed
cell complexes. Classical general topology is the study of the nearness of points to
sets. The focus here is on Computational CW topology which is the study of the
proximities of cell complexes enriched with algorithms (stepwise problem-solving
methods).
Analyzing Surface Tilings: There are a number of ways to analyze tilings. For
example, a proximity function is defined to solve what is known as the split fea-
sibility problem in Sect. 1.24. This is a distinguished point approach to analyzing
surface tilings. The basic idea is to identify pairs of distinguished points of a
surface region such as the nucleus of a triangulation nerve and barycenters of
the nerve triangles. Then define a proximity function in terms of the distance
between region distinguished points. Carrying this approach further, we arrive at
a straightforward way to compare pairs of regions in different surface tilings.
Another means of analyzing surface tilings is to consider the persistence of surface
regions over time. By observing the inception of surface region and the eventual
disappearance of a surface region over time, we can track the behaviour of a sur-
face region (i.e., changing surface region feature values) with what is known as a
Ghrist barcode introduced in Sect. 1.14.
Example 1.8 A sample tessellated drone video frame2 is shown in Fig. 1.1. Unlike the
tessellation of the walls of ancient buildings, this form of tessellation is a collection of
non-uniform filled polygons called polytopes. The portion of the video frame covered
by a filled polygon defines the interior of the polygon. This is an example of a Voronoï
diagram. There is of interest (and an advantage) in identifying a tessellation polygon
with the maximum number of adjacent polygons. The green polygon in Fig. 1.1 is
an example. “
2 Many thanks to Enze Cui for supplying the tessellated drone video, which is the source of the
video frame in Fig. 1.1.
10 1 Computational Geometry, Topology and Physics of Visual Scenes
This section introduces basic structures associated with planar filled polygons (poly-
topes), namely, boundary, hole, interior and path. These structures are useful in
delineating one polytope from another one in a tessellation of a planar region.
A planar polytope with n sides is defined by the intersection of n closed half
planes. A half plane is a planar, two-dimensional region that contains all points on
one side of an infinite straight line and no points on the other side of the line [19].
A half plane is closed, provided it includes of the line on its edge. Otherwise, the
half plane is open. A closed half plane is a half plane that includes its edge. In other
words, each planar polytope is a filled polygon in the tessellation of a plane region.
For more about polytopes, see Ziegler [20].
Example 1.9 A sample 5-sided polytope is shown in Fig. 1.5. The gray interior region
of the polytope represents the intersection of 5 closed half planes. Try drawing a
polytope using the method in Algorithm 1. “
Example 1.10 (Construction of a Visual Scene Polytope) In keeping with the steps
in Algorithm 1, the construction of a visual scene polytope is shown in Fig. 1.8. Let
π1 , π2 , . . . , π6 denote the closed half planes relative to the sides of the polygon in
Fig. 1.7. Further, let > pq be a polygon edge with endpoints p, q. Select the closed half
plane πi with edge that includes > pq and that covers the interior of polygon (see, e.g.,
half plane π1 in Fig. 1.8a). Moving counter clockwise, the selection of the remaining
half planes is represented in Fig. 1.8b–f. The intersection of these half planes is
represented in Fig. 1.8g, which is the constructed visual scene polytope. “
from its physical analogue, namely that part of a physical space that absorbs (does
not reflect) light. The darkness of a surface hole in a visual scene is an indication of
its depth.
In a visual scene bombarded by light, a hole is analogous to a funnel (surface
puncture) that photons fall into. From a computational perspective, surface holes
(surfaces with punctures) are important, since the profile of each visual scene shape
is delineated by one or more holes in its interior. A typical urban example is the
shape of the front of a house delineated by its doors and windows. From a physical
geometry perspective, each surface is characterized by the shapes of the boundaries of
its surface punctures. For example, house fronts are delineated by the positioning and
boundaries of its holes defined by frontage windows and entries. See, for example,
the house fronts in the video frame in Fig. 1.1. Another typical example of a surface
12 1 Computational Geometry, Topology and Physics of Visual Scenes
shape characterized by its surface hole, is a terrain catchment that becomes either a
puddle, river, lake or reservoir during the rainy season.
Physical geometry. In the context of a computational geometry of visual scenes,
we consider the physical geometry of surfaces. A physical geometry deals with
shapes and cell complexes on punctured (holed) and non-punctured physical
surfaces. A surface hole is a dark, light-absorbing region of a surface in a visual
scene. For example, a polytope Pg constructed on a visual scene typically has
a puncture surface hole in its interior. For a recent study of surface holes in
visual scenes, see Tozzi and Peters [21]. For more about physical geometry, see
Peters [22].
defined by the intersection of the closed half planes with edges on the boundary of
Pg and filled with the curved surfaces of the visual scene captured by a drone video
camera. “
Example 1.12 (Visual Scene Surface Holes) Sample surface holes in a visual scene
are the dark regions inside the green polygon in Fig. 1.1. The garage door opening
and the body of the black automobile in front of the garage are examples of holes in
this video frame. “
>
P= >pa0 , a>
0 , a1 , . . . , ak , q (Path between p and q).
pq
Example 1.13 (Tessellation Path) A sample tessellation path is shown in Fig. 1.9,
where
>
P= >pa0 , a>
0 , a1 , a1 , q (Sample Path between p and q) “.
pq
14 1 Computational Geometry, Topology and Physics of Visual Scenes
Problem 1.14 ®
Give a method for finding the shortest path between a pair tessellation vertices
on a finite, bounded plane region. For suggestions on how to do this, see, e.g.,
Morris [26]. “
This section briefly introduces Voronoï regions. Let S be a nonempty finite set of
seed points. For seed points p, q ∈ S in the Euclidean plane R2 and for all points r
in the Euclidean plane R2 , a half plane π pq is defined by
r − p is less than r − q
r − p ≤ r − q .
Also notice that if r is a point on the blue line on the edge of the half plane π pq in
Fig. 1.10a, we have
(a) (b)
(c) (d)
16 1 Computational Geometry, Topology and Physics of Visual Scenes
Taken together, the half plane π pq contains all of those planar points r that are either
on or on the lefthand side of the half plane edge. “
Let S be a set of seed points used to tessellate a finite bounded planar region. A
Voronoï region V ( p) contains a distinguished interior point p in S, used to define
the region as follows:
Voronoï region of s
V ( p) = r ∈ R2 : r − p ≤ r − q , ∀q ∈ S .
Midway property. Each edge on the boundary of Voronoï region V (s) is the
perpendicular bisector of the line joining neighbouring region boundary seed
point vertices. This line joining Voronoï region seed points is an edge on a half
plane. Recall that a half plane E (denoted by π E) is a planar region containing
all points on one side of an infinite line and no points on the other side [19]. A
closed half plane is a half plane that includes its edge (see, e.g., Fig. 1.10a).
Cover property. Let S be a set of seed points and let V ( p) be a Voronoï region
of seed point p in S. In addition, let seed point q ∈ S be one of the seed points
nearest p.
Then there is a closed half plane π with edge half way between p and q and
π covers the region of the plane containing the seed point p. See, for example,
Fig. 1.10a. Further, there is a half plane π halfway between each of the seed points
q nearest p. Each of these half planes has an edge that lies midway between p
and q and each of these intersecting half planes cover Voronoï region V (s). Let
be the set of points in the intersection of the closed half planes on V (s). In that
case, V (s) ⊆ and we say that covers the Voronoï region V (s). In general,
one nonempty set B covers another nonempty set A, provided A ⊆ B.
1.7 Voronoï Region Properties 17
That is, the Edelsbrunner–Harer nerve of the collection F equals the collection of
all nonempty subsets X of F whose intersection is nonempty. “
Nerve property. From Definition 1.17, a nerve is a collection of sets that have
nonempty intersection [27, Sect. III.2, p. 59]. A Voronoï region V (s) is a nerve,
since it is a collection of half planes fragments (subsets of half planes) that have
nonempty intersection; see, e.g., Fig. 1.12.
Closed half plane property. A half plane π is a region of the plane bounded by
a line (the bounding edge of the half plane). A closed half plane π (sometimes
denoted by > π for the sake of clarity) includes its bounding edge. Half plane > π
covering the region of the plane containing the seed point s is closed, by virtue of
the less than or equal ≤ requirement in the definition of a Voronoï region V (s),
s ∈ S (set of seed points), namely,
p − s ≤ p − q , ∀q ∈ S.
See, e.g., Fig. 1.10b in which the edges of the half planes covering V (s) are
included in the Voronoï region.
Intersection property. V (s) is a set of points in the intersection of k half planes,
provided the seed point s has k neighbouring seed points that are nearer to s than
any of the other seed points in S (see, e.g., Fig. 1.10c).
Boundedness property. V (s) is a finite, bounded region of the plane.
Polytope property. A planar Voronoï region V (s) is a polytope, since V (s) is
defined by the intersection of finitely many closed half planes.
Convex hull property. A nonempty set A in the Euclidean plane is a convex set,
provided every line segment joining any pair of points in A lies entirely on A. The
convex hull of A is the smallest convex set containing A. In our case, a Voronoï
region V (s) is the smallest convex set containing the points on the boundary of
V (s) and in the interior of V (s) (see, e.g., Fig. 1.10d).
1.7 Voronoï Region Homotopy Type Property 19
This section introduces what is known as the Voronoï region homotopy type property.
Voronoï regions V (s) have what is known as the homotopy type property. To see
this, we first consider what are known as homotopic maps. Let f and g be a pair of
20 1 Computational Geometry, Topology and Physics of Visual Scenes
h(x, 0) = f (x),
h(x, 1) = g(x),
for all x in X . That is, h agrees with f for t ∈ X such that t = 0 and h agrees with
g for t = 1. Edelsbrunner and Harer [27, Sect. III.2, p. 58] observe that t in [0, 1]
can be thought of as time and a homotopy as a time series function f t : X −→ Y
defined by f t (x) = h(x, t). In that case, this time series function starts at f 0 = f
and ends at f 1 = g.
A subset Y ⊆ X is a retract of X , provided there is a continuous map r : X −→ Y
with r (y) = y for all y in Y , an early view of retracts by Munkres [30, Sect. 19, p.
108].
Example 1.19 (Retract Over Time) We have already observed in Example 1.18 a
distinguished point in a Voronoï region, namely, the centroid of the region. Here,
we take a second look at retracts that occur over time. Observe that { p} ⊆ V (s)
(subset { p} containing distinguished point p in V(s)) is a retract of the Voronoï
region, where f t : V (s) −→ V (s), t ∈ [0, 1] is a family of continuous maps defined
by f 0 (x) = idx (x) = x (identity map) and f 1 (V (s)) = p (distinguished point
p ∈ V (s) that is the centroid of the Voronoï region V (s)). In that case, each map
f i (x) = p. The map f is called a retraction. In other words, at time t = 0 (starting
time), f 0 maps each member of the Voronoï region to itself and at time t = 1 (ending
time), f 1 maps each member of the Voronoï region to the centroid of V (x), which is
a distinguished point in the region. The identity map idx (x) = x is an example of a
fixed point map. “
Recall that an identity map of a nonempty set X onto itself (denoted by id X ) is
defined by
In that case, the sets X and Y are homotopy equivalent. The maps f and g are
called homotopy equivalences. If X has the homotopy type of a single point, then
the set X is contractible.
1.8 Voronoï Region Homotopy Type Property 21
Theorem 1.22 ([27, Sect. III.2, p. 59]) Let F be a nonempty collection of closed,
convex sets in the Euclidean plane. Then the nerve NrvF and the union of the sets in
F have the same homotopy type. “
This theorem from H. Edelsbrunner and J. L. Harer is known as the Nerve Theo-
rem, which has many alternate forms. From Theorem 1.22, we arrive at an important
property of Voronoï regions.
Voronoï Region Homotopy Type Property.
Let V (s) be Voronoï region of a seed point s. A Voronoï nerve (denoted by
NrvV (s)) is defined by
NrvV (s) = half planes π ∈ R2 : π
= ∅ (Voronoï nerve).
22 1 Computational Geometry, Topology and Physics of Visual Scenes
Each half plane π in NrvV (s) is a closed convex set, since the line segment joining
any pair of points in π is also in π. Hence, from Theorem 1.22, nerve NrvV (s) and
the union of the half planes in the nerve have the same homotopy type.
The locations of the seed points in a finite bounded surface region determines the
shapes of the resulting Voronoï regions. For example, if the seed points are all pairwise
equidistant from each other and occur at the intersections of uniformly spaced grid
lines, then the resulting Voronoï regions will be rectangular.
Problem 1.24 ®
Write either a Matlab or a Mathematica script that uses grid line intersections so that
none of the resulting Voronoï regions are square shaped. “
(a) (b)
t = x + iy
that serves as a local parameter. In that case, the surface becomes a Riemann surface.
For more about this, Weyl [32]. Of particular interest in Weyl’s work is the intersection
of closed paths on a smooth oriented surface [32, Sect. 11, p. 79ff]. “
Problem 1.26 K
Define a Voronoï region V (s) in terms of seed points s ∈ S on a Riemann surface.
Write either a Matlab or a Mathematica script that implements the construction of
Voronoï regions on a Riemann surface. “
In other words, the edges of the closed half planes used to construct a Voronoï
region are oriented relative to the distance between the interior point s and the nearby
neighbouring seed points r in S. Each edge of a Voronoï region is line segment of
one of the closed half planes midway between the seed points s and r . Each such
closed half plane spans the interior of the particular Voronoï region. The end result
of the construction of a Voronoï region is potent polytope that is a filled polygon.
Constructing a Voronoï region for each of the seed points in a finite, bounded region
X results in a cover of A. That is, a covering of X with Voronoï regions V (s) for
seed points s ∈ S is defined by
X⊆ V (s) (Region Covered with Voronoï Regions).
s∈S
24 1 Computational Geometry, Topology and Physics of Visual Scenes
Theorem 1.27 For a set of seed points S in a finite bounded region X of the Euclidean
plane, the construction of planar Voronoï regions with respect to each of the seed
points s ∈ X induces a cover of the region.
Example 1.28 A sample use of seed points in the construction of a Voronoï region
V ( p) is shown in Fig. 1.10. The construction starts with selection of a pair of seed
points such as p and q with a half plane π whose edge is midway between the
selected seed points as shown in Fig. 1.10a. The choice of upper versus lower half
plane depends on which half plane covers the interior of the particular Voronoï region.
The orientation of the edge of π depends on the locations of the seed points in the
immediate neighbourhood of p.
For example, the orientation of each of the halfplane edges in Fig. 1.10b is con-
trolled by the seed points q1 , q2 , q3 , q4 , q5 surrounding the point p in the region
V ( p). The resulting Voronoï region V ( p) is limited to the intersection of the five
half planes as shown in Fig. 1.10c. The dotted lines in Fig. 1.10c suggest those parts of
the infinitely long leading edges of the half planes on either side of the line segments
on V ( p). The end result is a polytope like the one shown in Fig. 1.10d. “
The subtle part of the pursuit of good tessellations of visual scenes is the appropriate
selection of seed points. A good tessellation of a visual scene is one which the
chosen seed points results in a covering of the foreground object shapes in a scene.
A planar shape A (denoted by shA) is a finite region of the Euclidean plane bounded
by a simple closed curve with a nonempty interior. A curve is simple, provided the
curve has no self intersections (loops). Let p be a point on a closed curve. A curve is
closed, provided there is a path on the curve that leads from any point p back to p.
Image object shapes correspond to the light-reflecting regions in a visual scene.
With that in mind, we have two basic choices in the selection of seed points that are
a source of a tessellation.
1.10 Centroids as Seed Points in the Interior of Shapes 25
Edge Pixels: A edge point is a point on a shape contour. Examples are corners and
Lowe keypoints. A corner is a point with an edge pixel with a gradient orientation
that differs sharply from its neighbours. Lowe keypoints were introduced by D.
Lowe in 1999 [37] and elaborated in [38]. A Lowe keypoint is an edge pixel
with high pixel edge strength. Let I mg be a visual scene image and let I mg(x, y)
be a pixel at location (x, y). The edge strength of a pixel (also called the pixel
gradient magnitude) is denoted by E(x, y) and defined by
∂ I mg(x, y) 2 ∂ I mg(x, y) 2
E(x, y) = + Pixel edge strength.
∂x ∂y
Centroids: The centroid of a finite bounded region of the plane is the center of
mass of the region. Let X be a set of points in a n × m rectangular 2D region
containing points with coordinates (xi , yi ) , i = 1, . . . , n in the Euclidean plane.
Then, for example, the coordinates xc , yc of a centroid in an n × n 2D rectangle
are
1
1
n n
xc = xi , yc = yi .
n i=1 n i=1
1
1
1
n n h
xc = xi , yc = yi , z c = zi .
n i=1 n i=1 h i=1
Example 1.29 (2D and 3D Region Centroids) In Fig. 1.17, the red dot • indicates
the location of a region centroid. Two examples are shown, namely, centroid • in
a 2D convex region in Fig. 1.17a and centroid • in a 3D region occupied with the
Wolfram Beethoven in Fig. 1.17b.
Centroids are always located in the interior of each image scene shape. For this
reason, centroids are ideal choices for seed points. Why? Tessellation polytopes will
be centered on the centroid in the interior of each shape. This becomes important in
the comparison of shapes either within the same visual scene or shapes in separate
video frames. By contrast, choosing edge pixels such as corners or keypoints as seed
points, leads to tessellation polytopes that are centered on shape contours, which is
not what we want. That is, edge-based tessellation polytopes tell us something about
shape contours but are not a good source of information about shape interiors. Notice
that shapes may have similar contours and yet have very different interiors.
26 1 Computational Geometry, Topology and Physics of Visual Scenes
(a) (b)
This section briefly illustrates the use of centroids as seed points in tessellating image
scene shapes. The basic steps in the construction of a centroid-based tessellation of
a visual scene are in given in Algorithm 2.
From the image object centroids found in Example 1.30, we derive the tessellation
show in Fig. 1.19. Following the steps in Algorithm 2, each time a centroid-based
polytope is constructed, it is superimposed on the image containing the centroids
found. The end result is shown in Fig. 1.20.
This section introduces the basics of cell complexes, which provide a basis for a
cellular form of computational geometry and lead to practical application of CW
topology in the study of shapes in visual scenes.
In the hands of Alexandroff [39], triangulation of plane regions led to the intro-
duction of a cellular topology in which a finite, bounded plane region is viewed as a
cell complex. In a topology of cellular complexes, a cell complex K is a Hausdorff
space and a sequence of path-connected 0-cells E (denoted by skE) are called skele-
tons in Cooke and Finney [40] (also called a CW complex or Closure-finite Weak
topology complex in Hatcher [41]). Briefly, a Hausdorff space is a nonempty set in
which every point resides in a neighbourhood that is disjoint from every other neigh-
bourhood of the points in the space. See Table 1.2 for an overview of the minimal
skeletons in a planar cell complex.
Recall that a space X is a nonempty set with particular properties such as cellular
(members of X are connected cells), topological (members of X are open sets in
which unions and intersections of subsets of X also belong to X ), contractible (X has
the homotopy type of a point), CW complex and so on. For a practical application
of CW complexes in terms of the persistence of cells over time, see Jaquette and
Kramár [42], who give an approach to constructing persistence diagrams, which are
very useful in tracking the conservation, deterioration and eventual disappearance of
physical cell complex shapes over time. Shape persistence occurs, provided a shape
retains its basic structure over time.
1.12 Cell Complexes in Computational Geometry and Topology 29
v1 e3
e5
e4
A finite, bounded planar shape A (denoted by shA) is a finite region of the Euclidean
plane bounded by a simple closed curve and with a nonempty interior [46]. The focus
here is on shape signatures that provide a fingerprint for members of a class of shapes.
A shape signature can either be a single feature value such as the radial distance
between the boundary points and the centroid of a shape as in El-ghazal, Basir and
Belkasim [47] or a shape feature vector as in Yang, Kpalma and Ronsin [48, Sect.
5.6, pp. 18–19]or a barcode as in Ghrist [49] and Peters [50].
Here, a vortex complex is a system of shapes within a shape4 that has interval-based
feature values that define a shape barcode (also called a Ghrist barcode) that persists
over time (see, e.g., [4, Sect. 2.6]). That is, a shape barcode is a collection of parallel
horizontal (or vertical) bars, each with a length representing the duration of a shape-
covering vortex complex feature that endures (persists). For a good introduction to
barcodes relative to shape persistence, see Ghrist [51, Sect. 5.13].
Example 1.33 (Barcode View of the Persistence of a Shape) A sample shape bar-
code is shown in Fig. 1.23. This barcode exhibits the persistence of combinations of
features of a shape over time. In this example, a sequence of 5 triangulated video
frames is represented. In each video frame, there is a collection of triangles with
a common vertex, which is the centroid of a rabbit. This collection of intersecting
triangles is an illustration of a cellular nerve structure. Each of the cell interiors in
this nerve has a dominant hue (red, green and blue) that varies over time. In addition,
the number of triangles and the maximum triangle area in this rabbit nerve also vary
over time. The interest here is in tracking the persistence of shape feature values
that appear and disappear in sequences of video frames over time. Notice that all
three triangle interior hues occur in only 2 out of the 5 frames. The total number of
triangles and the maximum triangle area in these two multi-hue nerves also persist
during the same temporal interval in which all 3 hues occur. In other words, there are
only two video frames in the sequence in which the rabbit nerve shape has matching
feature values. This rabbit nerve also exhibits a vortex cycle defined by the sequence
of edges opposite the centroid at the nerve center. “
Problem 1.34 K
The goal in this problem is to use a Ghrist barcode to track the persistence of the
feature values of a triangulated shape over a sequence of video frames in an offline
approach to the triangulation and tracking the persistence of shapes in video frames.
Do the following:
1o Capture a video of a changing visual scene.
2o Use Matlab to select a set of seed points that are centroids of the objects in each
video frame, offline.
3o Using the selected seed points, triangulate each video frame, offline.
4o Find a maximum nucleus cluster (MNC) on each video frame, offline.
5o Extract the dominant hue of each triangle in the MNC in each video frame,
offline.
6o Extract the total number triangles in the MNC in each video frame, offline.
7o Construct a Ghrist barcode that tracks the feature values of the MNC on the
sequence of video frames, offline. Comment on the resulting Ghrist barcode.
8o Repeat Step No. 1 for a second video of the same changing visual scene.
9o Comment on which MNC shape features tend to persist in the video frames.
Also, comment on which shape features repeatedly change over time.
“
The geometry of vortex cycles is related to the study shape signatures [50] and the
geometry of photon vortices by Litchinitser [52], overlapping vortices by Adelberger,
Dvali and Gruzinov [53], vortex properties of photons and electromagnetic vortices
formed by photons by Dzedolik [54] and vortex atoms introduced by Thomson (Lord
Kelvin) [55].
Overlapping 1-cycles in a vortex constitute an Edelsbrunner–Harer nerve within
the vortex. Let F be a finite collection of sets. An Edelsbrunner–Harer nerve [27,
Sect. III.2, p. 59] nerve consists of all nonempty subcollections of F (denoted by
NrvF) whose sets have nonempty intersection, i.e.,
NrvF = X ⊆ F : X
= ∅ (Edelsbrunner–Harer Nerve).
Example 1.35 (Two Forms of Vortex Complexes) Two different vortex complexes
vcyc A, vcycB are shown in Fig. 1.22. Vortex complex vcyc A contains a pair of non-
overlapping 1-cycles cyc A1 , cyc A2 . By contrast, vortex complex vcycB in Fig. 1.22
contains a pair of overlapping 1-cycles cycB1 , cycB2 with a common vertex, namely,
v13 . Let F be a collection of sets of edges in cycB1 , cycB2 . The pair of 1-cycles
in vortex complex vcycB constitute an Edelsbrunner–Harer nerve, since cycB1 ∩
cycB2 = v13 , i.e., the intersection of 1-cycles cycB1 , cycB2 is nonempty. The edges
of the cycles in both forms of vortex complexes define closed convex curves. “
Overlapping vortexes give rise to vortex nerves. That is, a vortex nerve A (denoted
by vNrvA) is a collection of intersecting vortexes (Fig. 1.23).
Fig. 1.23 Sample Ghrist barcode showing the persistence of shape features over time
shape interiors with polygons with known geometric features. Notice that every
1-cycle has a boundary that is a polygonal curve. With 1-cycle polygonal curve
boundaries, we can not only measure the boundary features (e.g., length, number of
edges, longest edge, shortest edge) but also the interior features (e.g., area, diameter,
maximum distance from the 1-cycle interior centroid, minimum distance from the
1-cycle interior centroid, minimum and maximum wavelengths of the interior pixels
in a colour image). For the details concerning the wavelength of light waves, see
Appendix A.22. For an approach to determining the wavelengths of colour image
pixels, see Appendix A.22.
Example 1.37 (Sample Image Vortex Nerves) An example of twin image vortex
nerves is shown in the painting5 in Fig. 1.25. The trio of vortex cycles in Fig. 1.25
intersect pairwise, forming a pair of image vortex nerves on the painting of the girl
with green scarf. “
5 Many thanksfor Alessandro Granata, Salerno, Italy for letting me use his painting in this study of
image geometry.
34 1 Computational Geometry, Topology and Physics of Visual Scenes
A number of simple results for vortex cycles come from the Jordan Curve Theorem.
Theorem 1.38 (Vortex Cycle Retract Theorem) A finite planar vortex cycle has a
retraction to a distinguished point.
Proof Let vcyc A be a planar vortex cycle. And let the family of maps rt : vcyc A −→
vcyc A (t ∈ [0, 1]) be defined by
1.14 Shape Barcodes Similar to Ghrist Barcodes 35
From the Jordan curved theorem (our Theorem 1.1), we know that the planar region
bounded by a filled vortex cycle is separated from what is outside the cycle and
contains contains at least one distinguished point in the interior of the cycle, namely,
the centroid of the vortex interior. Centroid p ∈ vcyc A is a distinguished point
in vcyc A, since every filled planar vortex cycle has a centroid, which is a unique
distinguished point within the vortex cycle. Hence, the centroid p is a retract of
vcyc A.
A space that has the homotopy type of a point is contractible.
Theorem 1.39 (Vortex Cycle Retract Theorem)
A finite planar vortex cycle defines a space that is contractible.
Problem 1.40 ®
Prove Theorem 1.39. Hint: Define a vortex nerve resulting from the collection of
1-cycles on a vortex complex. “
Problem 1.41 ®
Complete the sequence of Ziegler -to-Y transformations on Fig. 1.26a. “
Problem 1.42 ®
Prove (a) a planar shape with a single hole can be contracted to path-connected
barycenters between the shape boundary and hole boundary and (b) the result-
ing path-connected barycenters are nuclei of a sequence of intersecting nerve
complexes. “
36 1 Computational Geometry, Topology and Physics of Visual Scenes
Problem 1.43 ®
How many 2-holes are needed to destroy a 1-cycle, making it a shape boundary with
an empty interior? “
Problem 1.44 ®
The diameter of a 2-hole is the maximum distance between a pair of points on
the boundary of the 2-hole. What is the diameter of a 2-hole in a filled, planar n-
sided polytope that destroys a 1-cycle, making it a shape boundary with an empty
interior? “
This section points to the contrast between Delaunay triangles and Voronoï regions.
The story starts with the selection of seed points S that are the corners of cells in a
rectangular grid. Recall that a Voronoï region contains a single seed point, which is in
the interior of the region. Unlike a Voronoï region, each Delaunay triangle (denoted
by ) contains 3 seed points on its boundary, namely, the vertices of the . The steps
in the construction of a Delaunay triangle are given in Algorithm 3.
(a) (b)
Fig. 1.27 Delaunay triangles on a rectangular grid
Example 1.45 (Delaunay Triangles) A set of nine seed points S taken from the
intersection of evenly spaced grid lines are shown as + are shown in Fig. 1.27a. As
a result, the Delaunay Triangles derived from these seed points are evenly shaped
but not equilateral (see Fig. 1.27b). Each triangle with red edges in Fig. 1.27b is an
example of a Delaunay Triangle ( pqr ) for three seed points p, q, r in S formed
by the intersection of a collection of half planes in which each edge of ( pqr ) is
a line segment on the edge of one the half planes π ∈ . “
Problem 1.46 K
Design a Matlab or a Mathematica script that uses grid line intersections so that
none of the resulting triangles are irregular (each triangle is equilateral and each
triangle has sides with lengths that do not match the lengths of the sides of the other
triangles). “
Problem 1.47 K
Use Matlab to design a script to do the following:
1o Given a digital image, select a set S of seed points that are image corners.
2o Triangulate S.
3o Recall that the barycenter of a triangle is located at the intersection of the median
lines. Find the set B of barycenters of the triangles from Step No. 1.47.2.
4o False colour the barycenters found.
5o Triangulate B. “
38 1 Computational Geometry, Topology and Physics of Visual Scenes
Problem 1.48 K
Recall that the barycenter of a triangle is located at the intersection of the median lines.
Design a Matlab or Mathematica script to display the barycenter of each Delaunay
triangle on a rectangular grid. “
Let X be a finite, bounded, planar region with holes. Each non-hole subregion of X
has a centroid. Let S be set of centroids on X . In this section, we briefly contrast
Delaunay triangles and Voronoï regions on S. From Algorithm 3, each seed point p
in S is the vertex of a triangle ( pqr ). The vertices q, r in ( pqr ) are seed points
that are nearest vertex p.
Recall from Table 1.2 that a 2-cell (K 2 ) is a filled triangle and a K 1.5 triangle
is partially filled, containing a 2-hole. A 2-hole is a puncture in a finite, bounded
planar region. Triangulation of a planar region containing holes results in Delaunay
triangles with one or more holes in their interiors.
Example 1.49 (Centroidal Delaunay Triangle with a Hole its Interior) Let S be a set
of centroids on a finite, bounded planar region with holes. Each centroid is a center of
mass of a shape. Three shapes A, B, E (denoted by shA, shB, shE) are represented
by blobs in Fig. 1.28a. The centroid of each shape is represented by red cross hairs +
located at points p, q, r . The resulting centroidal Delaunay triangle ( pqr ) is shown
in Fig. 1.28b. This particular Delaunay triangle has a prominent hole H0 in its interior.
In addition, triangle ( pqr ) straddles the three shapes shA, shB, shE. “
(a) (b)
1.16 Barcodes Derived from Centroidal Delaunay Triangles 39
Example 1.50 (Shape Barcode Derived from a Centroidal Delaunay Triangle) Let
( pqr ) be a centroidal Delaunay triangle with centroidal vertices p, q, r and
barycenter b on finite, bounded, physical planar region with holes so that the triangle
straddles three shapes shA, shB, shE. In that case, each centroid corresponds to a
tiny, vertexlike physical mass with a wavelength, whenever the centroid is exposed
to light.
The barcode for ( pqr ) is denoted by bc( p, q, r, b, A). For example, let
Over time, the shape and hue wavelengths of a filled triangle with holes on a sequence
of snapshots of a visual scene will vary due to changing temperature and lighting
(e.g., sunlight) conditions. Evidence can be found in video frame sequences of typical
daytime visual scenes. In the study of shapes in visual scenes, it is important to have
some mechanism to track shape changes. This can be done with what is known as a
Ghrist barcode (see, e.g., the barcode showing changes in filled triangle over time in
Fig. 1.30). Two temporal intervals over which the features of a changing filled triangle
persist are represented by temporal intervals between the yellow · · · · · · · · ·
dotted lines in Fig. 1.30). “
40 1 Computational Geometry, Topology and Physics of Visual Scenes
Fig. 1.30 Ghrist barcode exhibiting filled triangle vertex wavelength persistence
Shape persistence. K Those features of a shape that persist over time are of
great interest in gaining an understanding of the resilience and character of
an object with a particular shape such as the recently discovered Martian south
polar cap lake (liquid water deep below the south polar ice cap on Mars) reported
by Clery [58]. This discovery was made recently by the Mars Advanced Radar
for Subsurface and Ionospheric Sounding (MARSIS), an instrument on the
European Space Agency’s Mars Express, which began orbiting Mars in 2003.
For more about Ghrist barcodes, see Sects. 1.13 and 1.21 and Ghrist [49] and,
especially, [51, Sect. 5.13, pp. 104–106 and pp. 202–205]. Shape persistence
is defined by a temporal interval in which a particular configuration of a shape
begins and ends its existence (Fig. 1.31). “
Example 1.52 (Delaunay Triangles Derived from Rectangular Grid Seed Points for
Voronoï Regions) A sample derivation of Delaunay triangles derived the intersection
of a 2×2 rectangular grid covered by Voronoï regions is shown in Fig. 1.33. “
42 1 Computational Geometry, Topology and Physics of Visual Scenes
Fig. 1.35 Sample triangulation of a set of seed points from a visual scene
Step 11. Repeat the construction for a new Delaunay triangle, starting with the
Seed Point step for each seed point p in S .
Step 12. Triangulation Result: From the Result step. Triangulation of the seed
points (see, e.g., Fig. 1.35).
Step 13. Triangulation Cover Result: Triangulation of a visual scene, partially
covered with Delaunay triangles (see, e.g., Fig. 1.36). “
The focus of this section is on the contrast between Voronoï tessellation and Delaunay
triangulation of a visual scene. A number of contrasting features of the these two
forms of computational geometry are given next.
Locations of Seed Points: In a Voronoï region V (s) of a seed point s, the seed
point is in the interior (inside the boundaries) of V (s). See, for example, the
single seed point in the interior of the Voronoï region in Fig. 1.38a versus the
seed points that are vertices of the Delaunay triangle in Fig. 1.38b. The Voronoï
region in Fig. 1.38a comes from the tessellation shown in Fig. 1.37. Can you see
which one? The Delaunay triangle in Fig. 1.38b comes from the triangulation in
Fig. 1.35.
Region Clusters: Two significantly different forms of regions clusters result from
tessellation versus triangulation of a set of seed points on a finite, bounded, planar
region. With a Voronoï form of region clustering, a Voronoí region is at the center
of the cluster (see, e.g., Fig. 1.39a). This means that the leaves (Voronoí regions
adjacent to a central region) intersect with the region at the cluster center but non-
adjacent pairs of leaves do not intersect. See, for example, the green heptagons
on opposite sides of the red heptagon in Fig. 1.40. Although each of the green
heptagons has an edge in common, the green heptagons do not intersect (they
have nothing in common). For this reason, a cluster of Voronoí regions is not a
nerve structure.
1.20 Spoke-Based Cell Complex Nerves 45
Fig. 1.37 Voronoi tessellation of seed points derived from the centroids on a visual scene
(a) (b)
(a) (b)
This section briefly introduces nerve spokes. To obtain a nerve from either a tes-
sellated or triangulated finite bounded region, we need to solve the problem of non-
adjacent polytopes having an edge or a vertex in common with a given polytope but yet
do not intersect. This problem has been solved by introducing polyform nerves [60,
Sect. 12.1, p. 320ff] and cluster spokes [61] (see, also, Ahmad and Peters [62, 63]).
A polyform nerve is a collection of sequences of connected polygons that have a
polygon in common.
46 1 Computational Geometry, Topology and Physics of Visual Scenes
Problem 1.56 K
Write either a Mathematica notebook or a Matlab script that highlights spokes and
spoke-based nerves on a visual scene. “
Problem 1.57 K
Write a Matlab script that highlights spokes and spoke-based nerves in a sequence
of frames in a video. To do this, use image centroids as seed points. Highlight in
yellow the spoke-based nerves on maximal nucleus clusters (MNCs) with spokes
that have radiating spokes with the same length in a sequence of video frames. That
is, for instance in a sample video, if an MNC on a video frame X has 2 spokes
skA, skB with lengths 5 and 8, respectively, and another video frame Y also has an
MNC with 2 spokes sk A , skB with lengths 5 and 8, then highlight in yellow the
MNCS in frames X and Y . Notice that an MNC with spokes is example of a spoke-
based nerve. Compare the centroid-based spokes with the results (finding video frame
MNCs with spokes) using corners as seed points in a sequence of triangulated video
frames. Again highlight in yellow those video frame MNC spoke-based nerves with
radiating spokes with the same length.
6 Many thanks to Braden Cross for providing this example of spokes on a triangulated image.
1.20 Spoke-Based Cell Complex Nerves 47
Observe that a centroid-based MNC covers the interior a shape containing a vertex
that is the nucleus of the MNC. By contrast, corner-based MNC covers part of the
interior of a shape shA in which the nucleus of the MNC is a corner on the edge
of the shape shA. Comment on the differences between the two forms of MNCs
(i.e., centroid-based MNC versus corner-based MNC) in covering a particular shape
in a video frame. Which form of MNC would be more useful in comparing and
classifying video frame shapes? “
In the case of clusters of Voronoí regions, each spoke contains the Voronoí region
at the center of a cluster plus one of the cluster leaves (a Voronoí region that has an
edge in common with the cluster polygon at the center). By contrast with Voronoí
region clusters, the center of a Delaunay triangle cluster is a vertex that is a seed
point. The leaves of a Delaunay triangle cluster are triangles (see, e.g., Fig. 1.39b). In
addition, the leaves in a Delaunay cluster have a common vertex, namely, the vertex
at the cluster center. Hence, a Delaunay cluster A is a nerve complex defined by
NrvA = ∈ A :
= ∅ (Delaunay Cluster Nerve).
Spokes on a nerve NrvA on a finite, bounded planar region resemble the spiraling
ripples on the surface of a liquid vortex. Recall that a nerve spoke is a sequence of
connected triangles. The triangle in a spoke sequence contains either an edge or a
vertex on the border of the nerve Nrv A. Each spoke extends outward from a bounding
edge of the nerve NrvA. For a Delaunay nerve NrvA(S) on a set of seed points S on
a triangulated rectangular-shaped region, the end of each spoke is either a vertex or
an edge not in NrvA(S) and closest to a region boundary.
48 1 Computational Geometry, Topology and Physics of Visual Scenes
The are many ways to construct such nerve spokes. One of the simplest construc-
tions is derived by alternatively choosing an edge or a vertex outside nerve Nrv A(S)
and closest to a region boundary. Each chosen cell (vertex or edge) is used to append
>∈
a triangle to a nerve spoke. For example, first choose an edge qr / NrvA on a triangle
( pqr ) with vertex p on the boundary of nerve NrvA(S). In that case, the nerve
spoke skA contains one triangle, namely, ( pqr ) extending outward from nerve
NrvA(S). Next, choose a vertex p not in spoke sk A, not in nerve NrvA(S) and clos-
est to one of the region boundaries. Then draw triangle (qr p ). The spoke skA now
contains a pair of triangles with a common edge, i.e., skA := ( pqr ) ∪ (qr p ),
which spirals outward from the nerve, going towards one of the region borders. The
steps in the construction of a Delaunay nerve spoke in a bounded rectangular region
are given in Algorithm 4.
Algorithm 4 confines the construction of a nerve spoke to vertices and edges close
to a particular rectangular region side. For a more extreme spoke in which a spoke
wraps around a bounded region containing a Delaunay nerve, remove the restriction
that the same side of the rectangular region is chosen each time so that either a new
spoke vertex or edge is chosen. See Problem 1.61 for the construction of a winding
spoke on a Delaunay nerve.
Example 1.58 (Sample Delaunay Nerve Spokes) ( p q r )Three sample Delaunay
nerve spokes are shown in Fig. 1.43. The construction of a nerve spoke in Algorithm 4
can be viewed in terms of the green highlight spoke in Fig. 1.43. For the orange and
blue nerve spokes in Fig. 1.43, a similar alternating vertex-edge selection technique
is used. The basic approach in all three spokes is to wind each spoke outward towards
a region border. “
Problem 1.59 K
Implement Algorithm 4 in either a Matlab script or Mathematica notebook in terms
of constructing spokes on a maximum nucleus cluster (MNC) on a triangulated visual
scene. Each MNC is an example of a Delaunay nerve NrvA. For the seed points, use
centroids. Highlight the nerve NrvA in yellow (use high opacity so the underlying
image can be seen underneath the highlighted nerve). Highlight with green each
spoke extending outward from the nerve NrvA. “
Problem 1.60 K
Implement Algorithm 4 in a Matlab script in terms of constructing spokes on a max-
imum nucleus cluster (MNC) on triangulated video frames in two different videos.
Choose videos that track changes in similar visual scenes. Each video frame MNC
is an example of a Delaunay nerve NrvA. For the seed points, use centroids. High-
light the nerve NrvA in yellow (use high opacity so the underlying image can be
seen underneath the highlighted nerve). Highlight with green each spoke extending
outward from the nerve NrvA. Then do the following:
1o From each video frame, extract the following spoke feature values:
(i) number of triangles in nerve NrvA.
(ii) maximum triangle area in nerve NrvA.
(iii) maximum number of triangles in the spokes on nerve Nrv A.
(iv) minimum number of triangles in the spokes on nerve Nrv A.
(v) maximum hue wavelength of the vertices in the spokes on nerve Nrv A.
(vi) minimum hue wavelength of the vertices in the spokes on nerve Nrv A.
2o Give a Ghrist barcode that shows the persistence of the feature values over a
sequence of video frames. See Fig. 1.23 for an example of a Ghrist barcode.
Hint: Select a sequence of 8 video frames in which the spoke feature values
50 1 Computational Geometry, Topology and Physics of Visual Scenes
vary over time. For the selected sequence, estimate the total time represented
by the Ghrist barcode. Also estimate the duration that each spoke feature value
persists, i.e., stays close to the same value.
3o Indicate which of the selected spoke feature values would be useful in classify-
ing the video frame region containing the shape covered by the Delaunay nerve
NrvA. “
Problem 1.61 K
Relax the fixed side restriction in Step 8 of Algorithm 4 for the construction of a
nerve spoke in Problem 1.59. Give a new version of Algorithm 4 (call it the winding
spoke algorithm). Implement the winding spoke algorithm in either a Matlab script
or Mathematica notebook in terms of constructing spokes on a maximum nucleus
cluster (MNC) on a triangulated visual scene. Each MNC is an example of a Delau-
nay nerve NrvA. For the seed points, use centroids. Highlight the nerve NrvA in
yellow (use high opacity so the underlying image can be seen underneath the high-
lighted nerve). Highlight with green each spoke extending outward from the nerve
NrvA. “
Problem 1.62 ®
Repeat the steps in Problem 1.60, giving a new implementation and repeat the steps
for the winding spoke algorithm from Problem 1.61. The result for this problem will
be both a new implementation as well as a new Ghrist barcode relative to the persis-
tence over time for the MNC winding spokes on a sequence of video frames. “
Delaunay triangles are quite different from ordinary triangles in classical Euclidean
geometry. This distinguished character of Delaunay triangles results from its method
of construction, which places it in the hierarchy of cellular complexes in homology,
namely, 0-cells (vertices), 1-cells (line segments), 1.5-cells (filled triangles with one
or more holes) and 2-cells (filled triangles) in the plane (see Table 1.2). Delaunay tri-
angles are also the faces of 3-cells (filled tetrahedrons having faces that are Delaunay
triangles) and 2.5-cells (partially filled tetrahedrons having faces that are Delaunay
triangles with holes) in 3-dimensional space. Placing Delaunay triangles in the hier-
archy of cellular complexes distinguishes the view of such triangles in this work from
the more conventional view Delaunay triangles that appear in Delaunay triangula-
tions of finite, bounded regions in either 2D or 3D space. In keeping with an interest
in the triangulation of visual scenes that contain shapes filled with holes, we include
Delaunay triangles that have punctures (holes) in their interiors.
1.22 Properties of Delaunay Triangulation 51
Example 1.63 (Delaunay Edge Property) A sample closed half plane π pq on a Delau-
nay triangle ( pqr ) is shown in Fig. 1.44. The half plane covers the vertex r that is
opposite triangle edge >pq. “
Cover property. Let S be a set of seed points and let ( pqr ) be a Delaunay
triangle ( pqr ) with vertices p, q, r that are seed points in S. Each Delaunay
triangle ( pqr ) is covered by three intersecting closed half planes.
Proof This follows from the edge property of a Delaunay triangle ( pqr ), since each
edge of the triangle is the boundary of a closed half plane that covers the triangle
vertex opposite edge, e.g., edge >
pq is the boundary of a closed half plane π pq that
covers vertex r .
52 1 Computational Geometry, Topology and Physics of Visual Scenes
λ p = (wavelength of p).
λq = (wavelength of q).
v P L = (number of vertices in path > p, q).
e P L = v P L − 1.
Problem 1.65 ®
K Give a sample shape edge path Ghrist barcode bcP pq for a triangulated visual
scene. Highlight the edge path that corresponds to barcode bcP pq in two pairs of
similar (not identical) visual scenes. “
Problem 1.66 K
For a selected shape in the triangulation of a visual scene, compute the minimum
values of vPL, ePL and spanPL between opposite shape contour points. Highlight
the edge path that corresponds to the Ghrist barcode bcP pq in two pairs of similar
(not identical) visual scenes. “
That is, an Alexandroff nerve NrvE(S) of a nonempty set of seed points S equals the
collection of all triangles ( pqr ) with p, q, r ∈ S whose intersection is a seed point
vertex (called the nucleus). The nucleus of an Alexandroff nerve is a vertex common
to the triangles in the nerve. In effect, an Alexandroff nerve is a cluster of triangles
attached to a nucleus. For this reason, such clusters are called nucleus clusters. An
Alexandroff nucleus cluster (ANC) is a collection of Delaunay triangles that have
a common vertex. In other words, a Delaunay cluster of intersecting triangles is a
characterization of an Alexandroff nerve that calls attention to the importance of
the nerve nucleus. This becomes interesting in the case where the ANC is maximal
(denoted simply by MNC).
Proof Let p be a seed point in set of seed points S on a finite, bounded surface
region. Assume that there is set A containing k seed points q1 , . . . , qk , k ≥ 3 nearest
p. Then, for each pair qi , qi+1 ∈ A, construct a Delaunay triangle ( pqi , qi+1 ). In
that case, each of the constructed triangles has p as a common vertex, which is the
nucleus of an Alexandroff nerve.
(a) (b)
Fig. 1.46 Alexandroff nerve NrvD ({s0 , s1 , . . . , s8 }) and one of its closed half planes
1.23 Alexandroff Nerves 55
In addition, the Delaunay triangles in NrvE(S) have a common vertex (see, e.g.,
nerve nucleus s0 in Fig. 1.46a).
Example 1.69 (Delaunay triangulation on seed points that are image centroids) A
sample morning glory vortex with raindrops on the petals is shown in Fig. 1.47a. The
vortex shape of the morning glory provides a catchment area that raindrops flow into.
Let S be a set of seed points on the flower image and let s0 be a seed point in S. A
sample Alexandroff nerve NrvD(S) on part of the flower catchment area is shown
in Fig. 1.47b. In this case, s0 is the nucleus of NrvD(S), i.e., the filled triangles in
NrvE(S) have s0 as a common vertex. “
Next, we turn out attention to maximal nucleus clusters (MNCs) in the triangu-
lation of visual scenes. An MNC with a nucleus s0 (denoted by M N C(s0 )) in a
triangulation is an Alexandroff nerve in which the number of triangles on a nucleus
vertex (the common vertex of the triangles in the nerve) is maximal. Alexandroff
MNCs are important, since such clusters of triangles cover that part of a triangulated
surface with the highest concentration of seed points nearest a particular seed point,
namely, the nucleus of an Alexandroff nerve. Depending on the choice of seed points,
an Alexandroff MNC will reveal quite a bit about the characteristics of either the
boundary region or the interior of a surface shape.
(a) (b)
Fig. 1.47 Delaunay nerve NrvE ({s0 , s1 , . . . , s8 }) and its closed half planes
56 1 Computational Geometry, Topology and Physics of Visual Scenes
The bc M N C barcode includes an entry for the MNC edge path length (the MNC
perimeter denoted by perim M N C ), which equals the sum of the lengths of the edges
opposite the nucleus in an MNC. A bc M N C barcode is constructed in the following
manner.
Problem 1.71 K
Using the Matlab script from Problem 1.47, do the following:
1o Select a pair of digital images containing similar (but not the same) visual scenes
(your choice).
2o Find the MNCs in the pair of selected images. Do this for 13, 21, 34, 55, 89
seed points.
3o Compare shapes covered by the MNCs found in the pair of images. The com-
parison should be made by checking which MNC covers most or all of the
underlying shapes. Hint: concentrate on that part of an underlying shape not
covered by each MNC you find.
7 Many thanks to M. Z. Amad for this rendering of an Alexandroff MNC on a drone video frame.
1.23 Alexandroff Nerves 57
4o Determine which choice of seed points from Step 1.71.2 gives the best results.
5o Speculate about what changes need to be made in the Matlab script so an MNC
covers most or all of the shapes under the MNC. “
Problem 1.72 ®
Construct an example of an Alexandroff nerve Ghrist barcode bc M N C in the case
where the nerve has the maximal number of triangles in the triangulation of a visual
scene. “
Problem 1.73 K
Select a pair of similar visual scenes that are triangulated. Highlight the MNCs so that
the underlying shapes covered by the MNCs in the selected pair of images are visible.
Give the MNC Ghrist barcode bc M N C for each the triangulated images. Find a pair
of triangulated visual scenes where the bc M N C barcodes are almost equal. “
In its original form, the split feasibility problem is defined in terms of finding a vector
p in a set P in space X and a matrix A in a space Y so that Ap is in a set Q in space
Y . For Censor, Elfving, Kopf and Bortfeld [65], sets P and Q were closed convex
sets.
In this section, split feasibility is formulated relative to pairs of triangulated video
frames f i , f j , j > i, with an MNC on each of the frames. The problem then is to
find a nerve B on frame f j with q ∈ sk cyclic B (cyclic skeleton boundary) on nerve B
and a nerve A on frame f i with p ∈ sk cyclic A on nerve A so that
dist (p, sk cyclic A) − dist (q, sk cyclic B) < th.
The focus here is on selecting an appropriate threshold th > 0 and on comparing the
distance between a vector p and a cell complex sk cyclic A on one triangulated video
frame with the distance between a vector q and a cell complex sk cyclic B on another
triangulated video frame.
For example, let X and Y be a pair of triangulated video frames. Let p be the
nucleus of an Alexandroff nerve NrvA with a boundary sk cyclic A on frame X . And let
th > 0 be a threshold. Using what is known as the Hausdorff distance between a
vector and set (denoted by, for example, dist (p, sk cyclic A) for the minimum distance
between p and sk cyclic A), the split feasibility problem reduces to finding a boundary
sk cyclic B on an Alexandroff nerve NrvB with a nucleus q on frame Y so that
58 1 Computational Geometry, Topology and Physics of Visual Scenes
th > 0.
p ∈ sk cyclic A on NrvA in video frameX.
q ∈ sk cyclic B on NrvB in video frameY.
dist (p, sk cyclic A) = min p − a : a ∈ sk cyclic A .
dist (q, sk cyclic B) = min q − b : b ∈ sk cyclic B .
dist (p, sk cyclic A) − dist (q, sk cyclic B) < th.
For the details concerning the Hausdorff distance, see Appendix A.8.
The solution to the split feasibility problem is illustrated for a specific pair of Alexan-
droff nerves in Example 1.74.
Example 1.74 (Split feasibility for a pair of Alexandroff nerves) A pair of Alexan-
droff nerves NrvA, NrvB with cyclic skeleton boundaries sk cyclic A, sk cyclic B on a pair
of video frames i and j are shown in Fig. 1.49. In this example, nerve Nrv A is a col-
lection of triangles with a common vertex vecp. Similarly, nerve NrvB is a collection
of triangles with a common vertex vecq. To see applicability of the split feasibility
problem here, consider the problem checking if the minimum distance between p and
the vertexes on the nerve boundary sk cyclic A is close to the minimum distance between
q and the vertexes on the nerve boundary sk cyclic B. To check if this every happens on
a sequence of triangulated video frames, choose a closeness threshold th > 0. Then
1.24 Split Feasibility Problem for Alexandroff Nerves on Video Frames 59
we need to compute the Hausdorff distances dist (p, sk cyclic A), dist (q, sk cyclic B) and
check if the following inequality holds true:
?
dist (p, sk cyclic A) − dist (q, sk cyclic B) < th.
For example, in Fig. 1.49, a red −−− line segment attached between nerve nucleus p
and vertex a on the cyclic skeleton boundary sk cyclic A has a length which is close to
the length of the red −−− line segment attached between nerve nucleus q and vertex
b on the cyclic skeleton boundary sk cyclic B. The closeness of this pair of line segments
is defined in terms of a chosen threshold. For an appropriate choice of threshold th,
the feasibility problem for nerves on a pair of video frames is solved. “
20 /* This completes the solution of the split feasibility problem for comparable Alexandroff
MNCshape s on a sequence of video frames.*/ ;
Problem 1.75 ®
Complete Algorithm 6 for the frames in a video by checking for a failed search rel-
ative to an initially selected MNCshape E, nucleus p, boundary vertex b ∈ sk cyclic E,
60 1 Computational Geometry, Topology and Physics of Visual Scenes
minimal distance dist ( p, sk cyclic E) in looking for and not finding a second maxi-
>
mal Alexandroff nerve MNCshape E with a minimal length segment p , b between
nucleus p and nerve boundary vertex b ∈ sk cyclic E on MNCshape E so that
dist (p, sk cyclic E) − dist (p , sk cyclic E < th. “
Problem 1.76 K
Recall from that an Alexandroff nerve shape NrvE on a triangulated digital image
is a collection of triangles with a common vertex. We obtain a maximal Alexandroff
nerve MNCshape E, provided the number of triangles in the nerve is maximal in
comparison with other nerves in the triangulation. Do the following:
1o Using your cell phone or digital camera, select a 1/2 min to 1 min video.
2o Implement Algorithm 6 for the frames in a video using Matlab. In your imple-
mentation of this Algorithm, be sure to highlight the following shapes in the
selected video frame.
(i) highlight centroids as red • dots on each triangulated video frame.
(ii) highlight triangles as green triangles on a maximal Alexandroff nerve
found on a triangulated video frame.
>
(iii) highlight triangles as red −−− minimal length segment p, b between on
maximal Alexandroff nerve found on an initial triangulated video frame.
>
(iv) highlight triangles as red −−− minimal length segment p , b between on
maximal Alexandroff nerve found on another triangulated video frame (if
such a video frame exists).
3o Display two video frames containing Alexandroff nerve shapes that solve the
split feasibility problem.
4o Display the minimal segments on the pair of nerve shapes MNCshape E,
MNCshape E found.
5o Display the difference dist (p, sk cyclic E) − dist (p , sk cyclic E ) and threshold th
in millimeters.
6o Give video used in solving this problem. “
8o Experiment Step: Give several examples of pairs of visual scene MNCs that
represent solutions to the split feasibility problem.
9o Illustrate the Experiment Step 1.77 for several different thresholds th. “
Problem 1.78 K
For the MNC on a triangulated visual scene, do the following:
1o Modify Algorithm 6 so that the wavelengths are computed for each nucleus
q and the boundary vertexes of a MNC found on a video frame. Recall from
Appendix A.22 how the wavelength λ photon of a photon [10, Sect. 8.2, p. 260]
is computed using
2π 2π
λ photon = | f − f |
= t − t .
|t−t |
|f − f |
To solve the split feasibility problem for pairs of wavelengths, find wavelengths
λ p , λq whose difference is less than the threshold th.
5o Give several examples of pairs of visual scene MNCs that represent solutions
to the split feasibility problem. “
This section briefly introduces a means of estimating the wavelength λ of colour pixel
in an RGB (red green blue) raster image or in an RGB video frame. Wavelength
is a nonlinear function of pixel hue. The first step in the calculation of λ is the
conversion of RGB image to HSV (Hue Saturation Value). Then we transform the
hue channel(h) according to the following equation which is an approximation of
the nonlinear mapping.
435 nm if h(i, j) > 0.7483,
λ(i, j) = −(h(i, j)−2.60836) (1.1)
0.004276
other wise.
Here we assume that the hue values are scaled between [0, 1]. The wavelengths
calculated by this equation (measured in nanometers nm, i.e., 10−9 m) are limited to
the range [435, 610 nm]. For more about this, see Ahmad and Peters [66, Sect. 4, p.
51] and the wavelength of a photon in Appendix A.22.
64 1 Computational Geometry, Topology and Physics of Visual Scenes
Problem 1.79 Split Feasibility of Pairs of Visual Scene of average MNC barycen-
ter Wavelengths. ®
Repeat the steps in Problem 1.77 for the average wavelengths of the MNC barycen-
ters. This is an important problem to consider, since the closeness of similar MNCs
may fail geometrically but succeed in terms of the wavelengths of the hues of the
barycenters on a pair of MNCs. The solution to this problem edges closer to a method
of comparing visual scene shapes partially covered by MNCs, where the nuclei of the
MCS are in the interior of a pair of object shapes. In addition, average wavelengths
of the MNC barycenters is a useful feature to consider in classifying MNC shapes
(for more about this, see Problem 7.18). “
Br ( p) = {q ∈ K : p − q < r } .
Of great interest in the study of the closeness of vortex cycles is the interior of a
shape, found by subtracting the boundary of a shape from its closure. In general,
the interior of a nonempty set A ⊂ X (denoted by intA) defined by
A nonempty set A is a closed set, provided A includes both its interior and its
boundary. In effect, a nonempty set A is closed, provided
Example 1.80 A collection of cells that are closed sets is shown in Fig. 1.51. In each
case, the boundary of a cell as well as the interior of a define the cell. For example,
the boundary of a vertex (0-cell) is the vertex itself. Similarly, the interior of a vertex
is also the vertex itself. The situation is more straightforward in the following cases:
1-cell (line segment): The boundary of a 1-cell are its end points and the interior
of a 1-cell is the set of points on the line drawn between the end points.
1-cycle on a Delaunay nerve: Connected closed sets of 0-cells (vertices) that
define sequences of 1-cells on a Delaunay nerve. Let cyc A be a 1-cycle with
vertices that are barycenters on an Alexandroff nerve Nrv A and let B = > pq be a
1-cell in cyc A. The interior of a 1-cycle equals surface region of the Alexandroff
nerve bounded by cycA and the boundary of cyc A is defined by
>
bdy(cyc A) = pq.
p,q∈cyc A
See, for example, the pair intersecting 1-cycles in Fig. 1.52. A 1-cycle on the
barycenters of an Alexandroff nerve is also called a barycentric 1-cycle.
1.5-cell: Filled triangle with holes. A 1.5-cell equals the union of its edges (bound-
ary) and its interior region bounded by its edges (i.e., a 1.5-cell interior is defined
by a finite, bounded punctured plane region bounded by its edges). A punctured
plane region is a plane region with points removed. In the case of the complex
plane, a punctured plane region is the complex plane with the origin removed.
2-cell: Filled triangle without holes. A 2-cell equals the union of its edges (bound-
ary) and its interior region bounded by its edges.
66 1 Computational Geometry, Topology and Physics of Visual Scenes
Example 1.81 Cycle A and cycle B in Fig. 1.52 are examples of skeletons that are
conn
connected. In that case, we write cyc A δ cycB. That is, cycle A and cycle are
connected, since this pair of cycles has a common edge. B Similarly, the boundaries of
nerves NrvA and NrvB in Fig. 1.52 are also examples of skeletons that are connected.
For this reason, we write
That is, nerves NrvA and NrvB are connected, since this pair of nerves has a common
vertex.
conn
K = δ ({0 − cell, 1 − cell, 1.5 − cell, 2 − cell} ∈ K ) ∪
conn
δ ({1 − cycle} ∈ K ) (cell complex∈ K )
union of connected cells in complex K .
cyc A ∈ NrvA,
cycB ∈ NrvB,
cl(cyc A) = cycA.
cyc A ∩ cycB = >pq
= p•——•q .
68 1 Computational Geometry, Topology and Physics of Visual Scenes
In this example, the 1-cycle with only one other 1-cycle in the triangulation. If we
expand the vertices on the extremities of the nerve Nrv A, the more 1-cycles in the
triangulation. “
Problem 1.84 ®
Expand nerve NrvA in Fig. 1.52 with barcentric 1-cycles on nerves adjacent to nerve
NrvB. Give the intersection of the barycentric filled 1-cycle cycB on nerve NrvB
with other barycentric 1-cycles in the expansion of nerve NrvA. “
Problem 1.85 K
In a Mathematica notebook, implement Example 1.83 for an MNC in the triangulation
of a selected visual scene. That is, do the following:
1o Select a visual scene img.
2o Find an MNC NrvA on a visual scene.
3o Find the barycenters of the triangles on Nrv A.
4o Draw 1-cycle cyc A on nerve NrvA.
5o Draw a nerve NrvA on one of the edges opposite the nucleus of nerve NrvA.
6o Find the barycenters of the triangles on nerve NrvB.
7o Draw 1-cycle cycB on nerve NrvB.
8o Highlight the interior of 1-cycle cyc A on nerve NrvA and the interior of 1-cycle
cycB on nerve NrvB.
9o Highlight the 1-cell (line segment) in the intersection of
1-cycle cyc A and 1-cycle cycB. “
Problem 1.86 K
In a Matlab script, implement the steps in Example 1.85 on the image frames of
a video. In other words, triangulate each video frame and then highlight a 1-cycle
cyc A on nerve NrvA on an MNC, highlight the interior of 1-cycle cyc A on nerve
NrvA and the interior of 1-cycle cycB on nerve NrvB. Also, for each triangulated
video frame, highlight the 1-cell (line segment) in the intersection of 1-cycle cyc A
and 1-cycle cycB. “
Problem 1.87 ®
Let A be a sub-complex of the cell complex defined by the triangulation of a visual
scene. Prove that A has the closure finiteness property. “
Problem 1.88 ®
Let A, B be sub-complexes of the cell complex defined by the triangulation of a
visual scene. Prove that that the nonempty intersection of any closed sub-complex
of K with the closure of another sub-complex B in K is closed. “
This section briefly pays tribute to P. Aleksandrov [Alexandroff], H. Hopf and J.H.C.
Whitehead who introduced the containment and intersection conditions for what are
known as CW complexes8 (see Fig. 1.53a). CW stands for Closure finite and Weak
[topology]. Basically, a cell complex in the plane is a collection of what are known
as skeletons (sequences of vertexes, edges and filled triangles). In the plane, a cell is
either a 0-cell (vertex) or 1-cell (edge) or 2-cell (filled triangle).
Containment Condition for a CW complex. In a CW complex, it always the
case that a cell on a cell in complex K is a member of the complex K (this is
the Alexandroff–Hopf containment condition). “
A cyclic skeleton is an example of a closed cell. It is always the case that the intersec-
tion of two closed cell complexes sk cyclic A, sk cyclic B on a cell complex K is a closed cell
on sk cyclic A and on sk cyclic B. This intersection can either be a sequence of one or more
edges common to sk cyclic A, sk cyclic B or a single vertex common to sk cyclic A, sk cyclic B.
Notice that a vertex (0-cell) or an edge attached between vertexes (1-cell) is a closed
cell.
1.27 CW Complexes and Their Origin 71
The 0-cell q is closed, since it contains both its boundary (namely, itself) and its
interior (also, itself). And q is a closed cell on both > >r .
pq, q, “
Problem 1.90 ® Given a cell complex K , prove that following cases satisfy the
Alexandroff–Hopf Intersection Condition:
1o Let skA, skB on K be skeletons (sequences of connected edges) with a common
edge, i.e.,
skA ∩ skB = common 1-cell.
Problem 1.92 K Given a cell complex K resulting from the triangulation of a set
of seed points on planar region. Let sk cyclic E be a cyclic skeleton on the boundary of
a hole on K . Assume that int(sk cyclic E) (interior of the cyclic skeleton) is a hole. Is
sk cyclic E a closed cell complex? “
72 1 Computational Geometry, Topology and Physics of Visual Scenes
Origins of the two conditions for CW complexes. The conditions for a closure
finite, weak topology of cell complexes were introduced by Alexandroff and
Hopf [67, Sect. III, starting on page 124]. Pavel Sergeevich Aleksandrov [known
as P. Alexandroff outside Russia], 1896–1982, wrote about 300 scientific works.
For example, in 1924, Alexandroff introduced locally finite covering, used as a
basis for his metrisability of topological spaces. His collaboration with H. Hopf
began in 1926, culminating in a seminal work on Topology in 1935. “
Heinz Hopf (see Fig. 1.53b),9 1894–1971, worked not only on the topology of
cell complexes with P. Alexandroff, he also is well-known for his work on algebraic
topology, especially on homology classes and vector fields.
John Henry Constantine Whitehead (see Fig. 1.53c),10 FRS, 1904–1960, intro-
duced the representation of projective spaces, 1930, supervised by O. Veblen, formal-
ized in 1939 the two Alexandroff–Hopf conditions for a CW complex [pp. 315–317]
[68].
J.H.C. Whitehead introduced an elaboration of the Alexandroff–Hopf CW
topology more than 80 years ago, culminated in his paper published in 1939 [pp.
315–317] [68] and elaborated in 1949 [15, Sect. 5, p. 223] and applied in the study
of shape signatures by Peters [50, Sect. 2.4, p. 81]. A cell complex K has a topology
τ that is a CW topology, provided τ has the closure finiteness and weak topology
properties. A good introduction to CW topology is given by Hatcher [41, pp. 519–
521] and by Jänich [69, Sect. VII.3, pp. 95–96].
Problem 1.93 ®
Let K be a cell complex defined by the triangulation of a visual scene. Prove that
that K has a weak topology. “
Image segmentation is the partition of an image into a set of meaningful regions based
on some criteria. This is an observation by Fehri, Velasco-Forero and Meyer [70]. In
this section, we consider the segmentation of an image based on the Alexandroff–
Hopf Topology of Complexes. The basic approach is to partition each image into
two types of closed sets. Here are several examples.
1-cycle on NrvA and let cycB be a 1-cycle whose vertices are nuclei of the nerves
along the boundary of NrvA. Let vcycE denote a collection of nesting, non-
concentric 1-cycles. The result is a vortex complex vcyc(cycA) defined by
barycentriccyc A ∈ NrvA.
bdy(NrvB) ∩ bdy(NrvA)
= ∅.
p = nucleus of NrvB.
vcycE(cyc A) = cyc A ∪ nuclei p -cycB ⊃ barycentriccycA.
The story starts with a mapping of either form of Delaunay triangle (i.e., either
filled 1.5-cells or filled 2-cells) to a distinguished point. Recall that a finite, bounded
region A of the plane contracts to a unique point x ∈ A, provided there is a family
of continuous maps f t : A −→ A, t ∈ I (I is an index set) so that each f i (x) sends
11 Many thanks to Alessandro Granata, Salerno, Italy for contributing a copy of his painting for this
(a) (b)
(a) (b)
Sources 1 This section briefly indicates introductory sources of the ideas and
approaches to as well as further reading in computational geometry and topology.
Computational Geometry Mesh generation : Elelsbrunner [59].
Polytopes : Ziegler [20].
Visual scene geometry, tessellation, triangulation : Peters [64].
Cell complexes :
A gentle introduction to cell complexes: Jänich [29, Chap. VII].
Best introduction to the topology of complexes and tiling with cells: Alexan-
droff [39, Sect. 1, pp. 1–11].
Definition of a complex: Cooke and Finney [40, Sect. 1.1].
Topological properties of cell complexes: Hatcher [41, Chap. 0, pp. 5–10].
Digital image geometry :
Application in the study of shapes in digital images: Ahmad and Peters [62].
Delaunay triangulation and Voronoï tessellation: Peters [60, Sect. 4.1, pp.
122–125].
Tiling :
Voronoï tessellations: A good introduction to tiling and other good things is
1.30 Sources and Further Reading 79
Grünbaum and Shephard [18] (see, also, [72]). See Sect. 5.4, starting on p.
250, on Dirichlet tilings on a set of seed points (what we are calling Voronoï
tessellations on a set of seed points). For the construction of a Dirichlet tessel-
lation on seed points on a finite, bounded plane regions, see Green and Sibson
[73].
Triangulation :
An excellent introduction to triangulation (including Delauny triangulation)
and its application, see Hjelle and Dæhlen, [74]. A principal part of this book
is Delauny triangulation. It includes a chapter on algorithms for Delauny
triangulation (Chap. 4) and a chapter on constrained Delauny triangulation
(Chap. 6). The usual view of Delaunay triangulation as the dual of Voronoï
tessellation is given in the book.
Observation 1 Delaunay triangulation.
Unfortunately, an important difference these two forms of tiling a surface hide an
important difference between them. From a topology of complexes perspective,
notice that the vertices of a Delaunay triangle are seed points (our 0-cells). By
contrast, the edges of a Voronoï region on a set of seed points do not include any
seed points. Instead, a Voronoï region has a single seed point in its interior. For
this reason, a seed point in a Voronoï region is not path-connected to any other
seed point in a Voronoï tessellation. By contrast, every seed point in a Delaunay
triangulation is path-connected to any other seed point in the triangulation. That
is, given a Delaunay triangle vertex p and other vertex q in the triangulation,
there is a sequence of edges between p and q. In other words, from a seed point
perspective, a Voronoï tessellation is not path connected. In other words, a
Delaunay triangulation is a complex with a CW topology defined on it. Further,
every sub-complex of a Delaunay triangulation has a CW topology defined on
it.
In addition, every Delaunay triangle has a distinguished point in its interior,
namely, the barycenter of the triangle. An Alexandroff nerve complex results
from the intersection of the triangles on any given vertex in the triangulation.
A barycentric 1-cycle complex results from the edges connected between the
barycenters of the triangles in every Delauny nerve. “
Delaunay triangulations: See Grünbaum and Shephard [18, p. 266] and Green
and Sibson [73, p. 173]. To construct a Delaunay triangle, start with three nearest
Voronoï regions on 3 seed points (i.e., adjacent polygons with seed points in their
interiors) and join the seed points with edges to form the triangle.
Other Forms of Tiling : Many other forms of tiling a finite, bounded planar region
are possible with shapes with known features such as contour, interior holes and
hue.
Example 1.101 A sample tiling with known shapes and features is shown in Fig. 1.57.
The kangaroo head B in region X is descriptively similar to the kangaroo head A in
region Y . This similarity stems from the fact that both heads have the same contour
and the same number and types of holes, but these shapes have differing interior
hues. For about tiling finite bounded, plane regions with different known shapes, see
Naimpally and Peters [17, Sect. 14.5, starting on p. 227]. “
Problem 1.102 K
Use Matlab to tile a visual scene with a known shape such as a polygon so each tile is
the same polygon. That is, each of the tiles will be a polygon the same number sides
and the same interior area. Define a proximity function that can be used to compare
two different visual scene regions X and Y using the distance between distinguished
points on the polygons in the two regions. Do the following. (1) Give an example of
a pair of planar regions X , Y in a visual scene where the proximity function is not
zero and greater than some fixed threshold. This indicates that visual scene regions
X , Y are different. (2) Give an example of a pair of planar regions X , Y in a visual
scene where the proximity function is zero or less than some fixed threshold. This
1.30 Sources and Further Reading 81
indicates that visual scene regions X , Y are alike. (3) Also solve (1) and (2) in terms
of a proximity function for a region in two different visual scenes. (4) Repeat steps
(1), (2), (3) on pairs of video frames. That is, solve (1) and (2) for pairs of video
frames that have similar tiled region and dissimilar tiled regions. “
Problem 1.103 ®
Repeat the steps in Problem 1.102 in tiling a visual scene with a known shapes such
as a polygon with different numbers of sides or different interior areas. “
For examples of other forms of tiling that does not depend on either Voronoï
tessellation or Delaunay triangulation, see [75]. See, for instance, carta marmorizzata
(marbled paper), pp. 164–165, and disegno cachemire (cashmere design), 166–167
for unusual decompositions of plane regions into artistically pleasing, space-filling
shapes.12
Applications of Computational Geometry:
Topology shape metrics are introduced by Barth, Niedermann, Rutter and
Wolf [76]. In this paper, drawings, maps, and images are rectangulated with grid
lines or orthogonal radial grids (concentric circles with known radii and between-
circle distances) to obtain a shape metric. The approach is comparable to the
derivation of proximity functions in the solution of split-feasibility problems in
case where rectangulated shapes are compared.
Robot motion planning is accomplished by decomposing the crawl space for a
robot into convex polygons with pathways between the polygons. See, for exam-
ple, de Berg, Cheong, van Kreveld and Overmars [77, Chap. 13, starting on p.
284].
Network localization is the focus of the paper by Dai, Shen and Win [78]. This
paper introduces an application of path connectedness relative to vertices either
in the interior of a planar polygon on network nodes or in triangulating network
nodes.
Nerve clusters on fMRI tessellations introduced by Peters, İnan, Tozzi and
Ramanna [79]. This paper uses computational proximity in nucleus clustering in
Voronoï tessellations. See, also, colour image segmentation using Voronoï tessel-
lations by Hettiarachchi and Peters [80] and brain tissue tessellation by Peters,
Tozzi and Ramanna [81].
Video frame tessellation and triangulation, leading to the detection of nerve
complexes useful in comparing video frames, is introduced by Peters [64].
Shape nerve complexes are derived from planar shape triangulations by
Peters [23]. Digital image object shape approximation is introduced by Ahmad
and Peters [62].
Computational Topology
Basics: Good, solid introduction of the basic topics: Edelsbrunner and Harer [27].
Persistence of time-varying systems:
Good overview of applications: Munch, [82].
Introduction to persistence barcodes: Ghrist [51, Sect. 5.13, pp. 104–106].
Path connectedness: Kaczynski and Mischaikov and Mrozek [24, Sect. 1.1,
Chap. 8 and Sect. 12.4]
Shape:
Introduction to shapes with holes: Peters [60, Sect. 5.3, pp. 148–150].
Introduction to retracts: Hatcher [41, Chap. 0, pp. 1–4].
References
1. Carrière, M., Oudot, S.: Joint contour nets. IEEE Trans. Vis. Comput. Graph. 20(8), 1100–1113
(2014)
2. Weisstein, E.: Closed curve. Wolfram MathWorld (2018). http://mathworld.wolfram.com/
ClosedCurve.html
3. Jordan, C.: Cours d’analyse de l’École polytechnique, Tome I-III.? Editions Jacques Gabay,
Sceaux (1991). Reprint of 1915 edition, Tome I: MR1188186, Tome II: MR1188187, Tome
III: MR1188188
4. Peters, J.: Proximal vortex cycles and vortex nerve structures. Non-concentric, nesting, possibly
overlapping homology cell complexes. J. Math. Sci. Model. 1(2), 56–72 (2018). ISSN 2636-
8692. www.dergipark.gov.tr/jmsm. See, also, https://arxiv.org/abs/1805.03998
5. Baldomir, D., Hammond, P.: Geometry of Electromagnetic Systems. Clarendon Press, Oxford
(1996). xi+239 pp. Zbl 0919.76001
6. Zangwill, A.: Modern Electrodynamics. Cambridge University Press, Cambridge (2013).
xxii+977 pp. ISBN: 978-0-521-89697-9, MR3012344
7. Nye, J.: Natural Focusing and Fine Structure of Light. Caustics and Dislocations. Institute of
Physics Publishing, Bristol (1999). xii+328 pp. MR1684422
8. Nakahara, M.: Geometry, topology and physics. Institute of Physics, Bristol, UK (2003).
xxii+573 pp. ISBN: 0-7503-0606-8, MR2001829
9. Worsley, A., Peters, J.: Enhanced derivation of the electron magnetic moment anomaly from
the electron charge from geometric principles. Appl. Phys. Res. 10(6), 24–28 (2018). https://
doi.org/10.5539/apr.v10n6p24
10. Susskind, L., Friedman, A.: Quantum Mechanics. The Theoretical Minimum. Penguin Books,
UK (2014). xx+364 pp. ISBN: 978-0-141-977812
11. Gruber, F., Wallner, G.: Polygonization of line skeletons by using equipotential surfaces–a
practical descriptions. J. Geom. Graph. 18(1), 105–114 (2014). MR3244043
12. Descartes, R.: Le monde de Mr. Descartes ou Traité de la lumiere, et des autres principaux
objets des autres sens avec un discours de l’action des corps, et un autre des fièvres, composez
selon let principes du mème auteur [The world of Mr. Descartes or the treatise of the light and
other main objects of the senses: with a speech of local movement, and another of the fevers,
compose according to the principles of the same author]. T. Girad, Paris, France (1664). http://
doi.org/10.3931/e-rara-18973
13. Boyer, C.: History of Analytic Geometry. Scripta Mathematica, NY (1656). Ix+291 pp.,
MR0081235
References 83
14. contributors, W.: Gravitational Wave. Wikipedia, The Free Encyclopedia (2018). https://en.
wikipedia.org/w/index.php?title=Gravitational_wave&oldid=845420183
15. Whitehead, J.: Combinatorial homotopy. I. Bull. Am. Math. Soc. 55(3), 213–245 (1949). Part
1
16. Willard, S.: General Topology. Dover Pub., Inc., Mineola (1970). Xii + 369pp., ISBN: 0-486-
43479-6 54-02, MR0264581
17. Naimpally, S., Peters, J.: Topology with Applications. Topological Spaces via Near and Far.
World Scientific, Singapore (2013). Xv + 277 pp., Amer. Math. Soc. MR3075111
18. Grünbaum, B., Shephard, G.: Tilings and Patterns. W.H. Freeman and Co., New York (1987).
Xii+700 pp., MR0857454
19. Renze, J., Uznanski, D., Weisstein, E.: Half plane. Wolfram MathWorld (2018). http://
mathworld.wolfram.com/Half-Plane.html
20. Ziegler, G.: Lectures on polytopes, Graduate Texts in Mathematics, 152. Springer, New York
(1995). x+370 pp. ISBN: 0-387-94365-X, MR1311028
21. Tozzi, A., Peters, J.: Topological assessment of unidentified moving objects. MDPI Preprints
2019(020160), 1–7 (2019). https://doi.org/10.13140/RG.2.2.10252.56960
22. Peters, J.: Two forms of proximal physical geometry. Axioms, sewing regions together, classes
of regions, duality, and parallel fibre bundles. arXiv 1608(06208), 1–26 (2016). To appear in
Adv. Math.: Sci. J. 5(2) (2016), Zbl 1384.54015
23. Peters, J.: Proximal planar shapes. Correspondence between triangulated shapes and nerve com-
plexes. Bull. Allahabad Math. Soc. 33 113–137 (2018). MR3793556, Zbl 06937935. Review
by D, Leseberg (Berlin)
24. Kaczynski, T., Mischaikov, K., Mrozek, M.: Computational Homology, Applied Mathematical
Sciences, vol. 157. Springer, New York (2004). xvii+480 pp. ISBN 0-387-40853-3/hbk, Zbl
1039.55001
25. Larman, D.: Paths on polytopes. Proc. Lond. Math. Soc. 20, 161 (1970)
26. Morris, Jr W.D.: Lemke paths on simple polytopes. Math. Oper. Res. 19, 780–789 (1994).
MR1304624
27. Edelsbrunner, H., Harer, J.: Computational Topology. An Introduction. American Mathematical
Society, Providence (2010). xii+241 pp. ISBN: 978-0-8218-4925-5, MR2572029
28. Hatcher, A.: Algebraic Topology. Cambridge University Press, Cambridge, UK (2002).
Xii+544 pp. ISBN: 0-521-79160-X; 0-521-79540-0, MR1867354
29. Jänich, K.: Topology. With a chapter by T. Bröcker. Translated from the German by Silvio
Levy. Springer, New York (1984). ix+192 pp. ISBN: 0-387-90892-7 54-01, MR0734483
30. Munkres, J.: Elements of Algebraic Topology, 2nd edn. Perseus Publishing, Cambridge (1984).
ix + 484 pp. ISBN: 0-201-04586-9, MR0755006
31. Shinbrot, M.: Fixed–point theorems. Sci. Am. 214(1), 105–111 (1966). https://www.jstor.org/
stable/24931240
32. Weyl, H.: The Concept of a Riemann Surface. [Die Idee der Riemannschen Fläche (German)].
Dover, Mineola, New York (2009). Translated by G.R. MacLane, xi+191 pp. ISBN-13: 978-
0-486-47004-7, MR0166351
33. Voronoï, G.: Sur un problème du calcul des fonctions asymptotiques. J. für die reine und
angewandte Math. 126, 241–282 (1903)
34. Voronoï, G.: Nouvelles applications des paramètres continus à la théorie des formes quadra-
tiques. premier mémoire. sur quelques propriétés des formes quadratiques positives parfaite.
J. für die reine und angewandte Math. 133, 97–102 (1908)
35. Voronoï, G.: Nouvelles applications des paramètres continus à la théorie des formes quadra-
tiques. deuxièm mémoire. researches sur les parallélloèdres primitifs. J. für die reine und
angewandte Math. 134, 198–287 (1908). JFM 39.0274.01
36. Mughal, A., Libertiny, T., Schröder, G.: How bees and foams respond to curved confinement:
level set boundary representations in the surface evolver. arXiv 1611(10055v1), 1–28 (2016)
37. Lowe, D.: Object recognition from local scale-invariant features. In: Proceedings of the 7th
IEEE International Conference on Computer Vision, vol. 2, pp. 1150–1157 (1999). https://doi.
org/10.1109/ICCV.1999.790410
84 1 Computational Geometry, Topology and Physics of Visual Scenes
38. Lowe, D.: Distinctive image features from scale-invariant keypoints. Int. J. Comput. Vis. 60(2),
91–110 (2004). https://doi.org/10.1023/B:VISI.0000029664.99615.94
39. Alexandroff, P.: Elementary Concepts of Topology. Dover Publications, Inc., New York (1965).
63 pp., translation of Einfachste Grundbegriffe der Topologie [Springer, Berlin, 1932], trans-
lated by Alan E. Farley, Preface by D. Hilbert, MR0149463
40. Cooke, G., Finney, R.: Homology of Cell Complexes. Based on Lectures by Norman E. Steen-
rod. Princeton University Press, Princeton; University of Tokyo Press, Tokyo (1967). xv+256
pp. MR0219059
41. Hatcher, A.: Algebraic Topology. Cambridge University Press, Cambridge (2002). xii+544 pp.
ISBN: 0-521-79160-X, MR1867354
42. Jaquette, J., Kramár, M.: On ε approximations of persistence diagrams. Math. Comp. 86(306),
1887–1912 (2016). MR3626542
43. Krantz, S.: A Guide to Topology. The Mathematical Association of America, Washington,
D.C. (2009). ix + 107 pp. The Dolciani Mathematical Expositions, 40. MAA Guides, 4, ISBN:
978-0-88385-346-7, MR2526439
44. Peters, J.: Two forms of proximal, physical geometry. Axioms, sewing regions together, classes
of regions, duality and parallel fibre bundles. Advan. Math. Sci. J. 5(2), 241–268 (2016). Zbl
1384.54015, reviewed by D. Leseberg, Berlin
45. Pudykeiwicz, J.: Examples of vortical structures. Researchgate (2018). https://www.
researchgate.net/post/What_are_examples_of_vortexes_in_the_physical_world
46. Peters, J.: Proximal planar shapes. Correspondence between shape and nerve complexes. arXiv
1708(04147v1), 1–12 (2017)
47. El-ghazal, A., Basir, O., Belkasim, S.: Farthest point distance: a new shape signature for fourier
descriptors. Signal Process.: Image Commun. 24, 572–586 (2009)
48. Yang, M., Kpalma, K., Ronsin, J.: A survey of shape feature extraction techniques. HAL
archives-ouvertes.fr (2008). http://hal.archives-ouvertes.fr/hal-00446037
49. Ghrist, R.: Barcodes: the persistent topology of data. Bull. Amer. Math. Soc. (N.S.) 45(1),
61–75 (2008). MR2358377
50. Peters, J.: Proximal planar shape signatures. Homology nerves and descriptive proximity. Adv.
Math: Sci. J. 6(2), 71–85 (2017). Zbl 06855051
51. Ghrist, R.: Elementary Applied Topology. University of Pennsylvania (2014). Vi+269 pp.
ISBN: 978-1-5028-8085-7
52. Litchinitser, N.: Structured light meets structured matter. Sci., New Ser. 337(6098), 1054–1055
(2012)
53. Adelberger, E., Gruzinov, A.: Structured light meets structured matter. Phys. Rev. Lett. 98,
010,402–1–010,402–4 (2007)
54. Dzedolik, I.: Vortex properties of a photon flux in a dielectric waveguide. Tech. Phys. [trans.
from Zhurnal Tekhnicheskol Fiziki] 50(1), 135–138 (2005)
55. Kelvin, W.T.L.: On vortex atoms. Proc. R. Soc. Edin. 6, 94–105 (1867)
56. H. Boomari, M.O., Zarei, A.: Recognizing visibility graphs of polygons with holes and internal-
external visibility graphs of polygons. arXiv 1804(05105v1), 1–16 (2018)
57. Singh, S., Gairola, U.: Coordinatewise commuting and weakly commuting maps, and extension
of jungck and matkowski contraction principles. J. Math. Phys. Sci. 25(4), 305–318 (1991).
MR1168798
58. Clery, D.: Liquid water spied deep below polar ice cap on mars. Sci. Mag. 1–2 (2018). http://
www.sciencemag.org/news/2018/07/liquid-water-spied-deep-below-polar-ice-cap-mars
59. Edelsbrunner, H.: Geometry and Topology for Mesh Generation. Cambridge Monographs on
Applied and Computational Mathematics, vol. 7. Cambridge University Press, UK (2001).
xii+177 pp. ISBN: 978-0-521-68207-7, MR2223897
60. Peters, J.: Computational Proximity. Excursions in the Topology of Digital Images, Intelligent
Systems Reference Library, vol. 102 (2016). Xxviii + 433 pp. https://doi.org/10.1007/978-3-
319-30262-1, MR3727129 and Zbl 1382.68008
61. Peters, J.: Proximal nerve complexes. A computational topology approach. Set-Value Math.
Appl. 1(1), 1–16 (2017). ISSN: 0973-7375, arXiv preprint arXiv:1704.05909
References 85
62. Ahmad, M., Peters, J.: Proximal cech complexes in approximating digital image object shapes.
Theory and application. Theory Appl. Math. Comput. Sci. 7(2), 81–123 (2017). MR3769444
63. Ahmad, M., Peters, J.: Delta complexes in digital images. Approximating image object shapes.
arXiv 1706(04549v1), 1–20 (2017)
64. Peters, J.: Foundations of Computer Vision. Computational Geometry, Visual Image Structures
and Object Shape Detection, Intelligent Systems Reference Library, vol. 124. Springer Inter-
national Publishing, Switzerland (2017). i-xvii, 432 pp. https://doi.org/10.1007/978-3-319-
52483-2, Zbl 06882588 and MR3768717
65. Censor, Y., Elfving, T., Kopf, N., Bortfeld, T.: The multiple-sets split feasibility problem and
its applications for inverse problems. Inverse Probl. 21(6), 2071–2084 (2005). MR2183668
66. Ahmad, M., Peters, J.: Maximal centroidal vortices in triangulations. A descriptive proximity
framework in analyzing object shapes. Theory Appl. Math. Comput. Sci. 8(1), 38–59 (2018).
ISSN 2067-6202
67. Alexandroff, P., Hopf, H.: Topologie. Springer, Berlin (1935). Xiii+636pp
68. Whitehead, J.: Simplicial spaces, nuclei and m-groups. Proc. Lond. Math. Soc. 45, 243–327
(1939)
69. Jänich, K.: Topologie. (German) [Topology], 8th edn. Springer, Berlin (2005). x+239 pp. ISBN:
978-3-540-21393-2, MR2262391
70. Fehri, A., Velasco-Forero, S., Meyer, F.: Segmentation hiérarchique faiblement supervisée.
arXiv 1802(07008v1), 1–4 (2018)
71. Gersho, A., Gray, R.: Vector Quantization and Signal Compression. Kluwer Academic Pub-
lishers, Boston (1992). xii + 732 pp
72. Grünbaum, B., Shepherd, G.: Tilings with congruent tiles. Bull. (New Ser.) Am. Math. Soc.
3(3), 951–973 (1980)
73. Green, P., Sibson, R.: Computing Dirichlet tessellations in the plane. Comput. J. 21(2), 168–173
(1978). MR0485467, includes both Dirichlet tessellations and Delaunay triangulations
74. Hjelle, Ø., Dæ hlen, M.: Triangulations and Applications. Mathematics and Visualization.
Springer, Berlin (2006). ISBN 978-3-540-33260-2, MR2262170
75. Ltd., A.: Disegnum. Prospettiva, simmetria, curve, arte celtica e islamica, sezione aurea. Alexian
Limited, Milano, Italy (2014). ISBN 978-88-518-0249-3, https://www.illibraio.it/libri/
76. Barth, L., Niedermann, B., Rutter, I., Wolf, M.: Towards a topology-shape-metrics framework
for ortho-radial drawings. arXiv 1703(06040v1), 1–35 (2017)
77. de Berg, M., Cheong, O., van Kreveld, M., Overmars, M.: Computational Geometry. Algorithms
and Applicaitons, 3rd edn. Springer, Berlin (2008). ISBN 978-3-540-77973-5, https://doi.org/
10.1007/978-3-540-77974-2
78. W. Dai, Y.S., Win, M.: A computational geometry framework for efficient network localiza-
tion. IEEE Trans. Inform. Theory 64(2), 1317–1339 (2018). MR3762623, includes detailed
algorithms
79. Peters, J., İnan, E., Tozzi, A., Ramanna, S.: Bold-independent computational entropy assesses
functional donut-like structures in brain FMRI images. Front. Hum. Neurosci. 11, 1–38 (2017).
https://doi.org/10.3389/fnhum.2017.00038, https://doi.org/10.3389/fnhum.2017.0003
80. Hettiarachchi, R., Peters, J.: Voronoï region-based adaptive unsupervised color image segmen-
tation. arXiv cs.CV 1604(00533v1), 1–2 (2016)
81. Peters, J., Tozzi, A., Ramanna, S.: Brain tissue tessellation shows absence of canonical micro-
circuits. Neurosci. Lett. 626, 99–105 (2016). https://doi.org/10.1016/j.neulet.2016.03.052
82. Munch, E.: Applications of persistent homology to time varying systems. Ph.D. thesis, Duke
University, Department of Mathematics (2013). Supervisor: J. Harer, MR3153181
Chapter 2
Cell Complexes, Filaments, Vortexes
and Shapes Within a Shape
Abstract This chapter revisits cell complexes, nerve structures and shapes in digital
images. To make sense of the anti-symmetry of holes and non-holes whose boundaries
define image object shapes, we represent shapes with completely understandable
connected structures called cell complexes. The beneficial outcome of this approach
to understanding the enigmas inherent in massive video frames point-clouds is our
introduction of shape fingerprints that are very simple collections of things called
filament skeletons with lucid measurable and comparable properties. The gist of this
approach is a plain-speaking way to compare shapes lurking in the point-clouds in
digital images.
In bounded planar regions, the building blocks of a cell complex are vertices (0-
cells), edges (1-cells) and filled triangles (1.5 cells or 2-cells). By selecting a set of
distinguished surface 0-cells, we can triangulate a bounded surface region containing
the selected vertices using Delaunay’s approach. Recall that such 0-cells are called
seed points. That is, a seed point is a surface point used as a vertex in a surface
triangulation or as a pivot in the construction of filled polygons in a Voronoï surface
tessellation. In the case of a triangulation, a cell complex is the result. A cell complex
is a collection of path-connected vertices that lie on what is known as a Hausdorff
space. Briefly, a space is Hausdorff, provided separate points live in separated (non-
intersecting) balls with nonzero radii. A penultimate example of a cell complex is a
Delaunay triangulation.
Every cell in a cell complex is a closed set, i.e., each cell in a complex has a
boundary and a nonempty interior. In effect, every cell is an example of a shape. In
the case of a 1-cell, the vertices are its boundary and its interior is defined by the
edge between the vertices. For either 1.5 cell (filled triangle with holes in its interior)
or 2-cell (conventional filled triangle), the edges of the triangle are its boundary and
whatever lies between the edges define its nonempty interior.
The focus here is on the decomposition of finite, bounded surface regions into col-
lections of connected cells that cover surface shapes and holes. Such collections
of connected cells (vertices) are called cell complexes. Every cell complex entirely
fills a bounded planar surface region with vertices, edges and filled triangles. A sur-
face shape is any bounded surface region with nonempty interior. Typically, surface
shapes have holes in their interiors. This is especially true of physical surface shapes.
By contrast with surface shapes, a surface hole is a bounded region with an empty
interior (Table 2.1).
What is commonly known as a shape (namely, a contour) is, by definition, a
hole. In other words, a surface hole is identified with its contour (the edges on its
boundary). Holes are very important in distinguishing between shapes, depending on
the sizes and number of holes in the shape interiors. Any dark region in a visual scene
is considered a hole, since each dark region absorbs light (i.e., a flow of photons1
colliding with such a hole, falling into a dark region). The dark regions in the Martian
surface in Fig. 2.1 are examples of holes.
A distinguished sequence of circular path-connected vertices that lie on a pla-
nar cell complex is called a vortex. Every vortex has a continuous boundary and a
nonempty interior. Hence, a vortex is an example of a shape. The simplest vortex is
a single sequence of circular path connected vertices with no vortices in its interior.
The yellow path connected vertices in Fig. 2.1 is an example of a simple barycen-
tric vortex. A barycentric vortex is defined by a sequence of connected triangle
barycenters that form a circular path. Vortexes are of great interest because of their
common occurrence in physical surfaces. The boundary of every physical surface
hole is an example of a virtual vortex. A physical surface hole boundary is a virtual
vortex, provided we draw edges between the centroids of the particles that lie on
the boundary of a physical hole. A vortex that is a collection of nesting, usually
non-concentric vortexes is an example of a system of shapes in a shape.
1 Xavier Oudet observed that light is just a flow of photons [1]. A photon is a quantum of electromag-
netic energy whose size is determined by its wavelength. For more about this, see Appendix A.22.
2.3 Video Frames, Hausdorff Spaces and CW Complexes 89
Fig. 2.1 Barycentric vortex on a triangulated Mars Express radar image showing Martian sub-
surface lake
Recall that every space is a nonempty set with particular properties such as being
finite and closed (a space with a boundary).
In this work, each space is Hausdorff. This means that distinct points reside in
balls (also called neighbourhoods) that are disjoint from each other, i.e., one can find
a ball containing each point so that none of the balls intersect. Let R2 denote the
Euclidean plane. In the plane, a ball with radius r and containing a point p (denoted
by Br ( p)) is defined by
Br ( p) = q ∈ R2 : p − q < r (Ball containing a point).
The ball Br ( p) of point p is open, since it does not include its boundary. Otherwise,
a planar ball of a point that includes its boundary is closed. Obviously, in the case
where every vertex in a planar cell complex lives in an open ball, then that complex
is a Hausdorff space.
Every physical surface is a collection of particles (something with nonzero mass)
separated by holes in spacetime. A hole in a physical surface is that part of the surface
containing no particles. Holes have zero mass. It is assumed that every particle on a
physical surface lives in a surface region resembling a donut [doughnut].
Lemma 2.1 A finite, bounded physical surface is a Hausdorff space.
Proof Let X be a finite, bounded physical surface containing particles p, q that are
nearest neighbours. Every particle on X is separated from particles that are its nearest
90 2 Cell Complexes, Filaments, Vortexes and Shapes …
..
.
In this section, we take a brief look at the Closure finite and Weak topology prop-
erties that characterize what is known as a CW complex. The term weak topology
results from the fact that such complexes require only what is known as the inter-
section property (every intersection of closed sub-complexes of a cell complex K is
nonempty as well as closed).
K−1 ⊂ K0 ⊂ K1 ⊂ K2 ⊂ · · · ⊂ Kn .
Kn ⊂ K is closed.
K = Kk .
0≤k≤n
Recall that the closure of a set equals both the boundary of a set and the interior of
the set. For example, Let >pq be a 1-cycle (represented by p•——•q). The boundary
of >
pq is the pair of vertices p, q. And the interior of >pq is the edge —— attached
between p and q.
Closure Finiteness Property of CW Complexes For a CW complex, we need
to have the closure of the collection of skeletons Kn intersect only a finite
number of skeletons in the space. This requirement is easy to achieve in a cell
complex with a finite number of path-connected vertices. In a CW complex,
every skeleton is attached to one or more other skeletons. For example, a 0-
skeleton K 0 (set of vertices) is a subspace of 1-skeletons K 1 (set of edges), since
each vertex in a CW complex is attached to an edge. Similarly, a 1-skeleton K 1
is a subspace of 2-skeletons K 2 (set of filled triangles), since each edge in a CW
92 2 Cell Complexes, Filaments, Vortexes and Shapes …
Example 2.4 Sample triangle clusters are shown in Fig. 2.3. The vertices in triangles
A and B are path-connected and have edge > pq in common. The combination of
the two triangles attached to each other via an edge form a skeleton. Every pair of
vertices in this skeleton is path-connected. In the case of a vertex in A and a vertex
in B, there is a path between the pair of vertices via edge > pq.
The three triangles in Fig. 2.3b belong to a triangle cluster, thanks to connections
between the triangles provided by edges > pq and > pr. “
94 2 Cell Complexes, Filaments, Vortexes and Shapes …
This section carries forward our first look at CW complex skeletons by considering
triangle clusters that are nerve complexes. Let sk( Ai ), 1 < i ≤ n be a skeleton. A
collection of skeletons in a CW complex defines a skeletal nerve A (denoted by
skNrvA), provided
Skeletal nerve
Example 2.5 A collection of different forms of skeletal nerves, each having a com-
mon vertex, is shown in Fig. 2.4.
For example, collections of triangles attached to a vertex are shown in Fig. 2.4a
and b. Many examples of this type skeletal nerve appear in Sect. 1.23. A filament
skeletal nerve is shown in Fig. 2.4c. In this case, a collection of six filaments have
nonempty intersection, since the filaments have a common vertex. For example, in
Fig. 2.4c, filaments with vertices that are seed points are shown. These filaments
have nonempty intersection defining a skeletal nerve, since these filaments have a
common seed point vertex p. The skeletal nerve in Fig. 2.4d combines filament
skeletons and a triangle skeleton (three with filled triangles attached to a common
vertex). The nucleus of this hybrid form of skeletal nerve is the vertex p. “
The basic steps in the construction of a filament skeletal nerve are given in Algo-
rithm 7. In this algorithm, various forms of triangulation are possible. For example
and for simplicity, filament skeletal nerves can be constructed on the Delaunay tri-
angulation of a finite, bounded planar region. That is what is done in Example 2.6.
Problem 2.7 Choose an image displaying a visual scene such as the Martial surface
in Fig. 2.6. Implement the steps in Algorithm 7, using Delaunay triangulation of a set
of image centroids. For the Vertex selection step in Algorithm 7, use the nucleus p
of a maximal nucleus cluster NrvA in the triangulation. For the Filament skeleton
selection step in Algorithm 7, define a filament skeleton skE by drawing a sequence
of connected line segments starting at nucleus p and ending at the nucleus q of a
nerve cluster nearest nucleus p in Nrv A. “
Problem 2.8 K
Repeat the steps in Problem 2.7 for each of the frames in a video. Highlight filaments
in the maximal filament skeleton in each of the frames. “
Problem 2.9 Choose an image displaying a visual scene such as the Martial surface
in Fig. 2.6. Implement the steps in Algorithm 7, using Delaunay triangulation of a
set of image centroids. For the Vertex selection step in Algorithm 7, use the nucleus
p of a maximal nucleus cluster in the triangulation. For the Filament skeleton
selection step in Algorithm 7, define a filament skeleton skE by drawing a sequence
of connected line segments starting at nucleus p and ending at the nucleus q of a
nerve cluster with center q furthest from nucleus p. “
Problem 2.10 K
Repeat the steps in Problem 2.9 for each of the frames in a video. Highlight filament
skeletons attached to nerve nuclei in each of the frames. “
drawing a sequence of line segments starting at nucleus p and ending at the nucleus
q of a nerve cluster with center q with a wavelength λq satisfying the following
condition:
Filament Terminating Condition
λ p − λq
> th.
For each pair of intermediate vertices in the filament skeleton skE, choose an edge
containing vertices that are closest to each other. This will sometimes result in over-
lapping filament skeletons. Give examples of filament skeletons in two different
images. “
Problem 2.12 K
Repeat the steps in Problem 2.11 for each of the frames in a video. Highlight filament
skeletons attached to nerve nuclei in each of the frames. “
Problem 2.13 (Centrality of nerve nuclei in a social network) Choose a set of seed
points that are nodes of interest in a social network. Implement the steps in Algo-
rithm 7, using Delaunay triangulation of a set of nodes of interest. For the Vertex
selection step in Algorithm 7, use the nucleus p of a maximal nucleus cluster in
the triangulation. Let λ p be the degree of centrality from nucleus p. The degree of
centrality of a social network node is an importance score based on the number of
98 2 Cell Complexes, Filaments, Vortexes and Shapes …
links held by the node. Choose a threshold th > 0. For the Filament skeleton selec-
tion step, define a filament skeleton skE by drawing a sequence of line segments
starting at nucleus p and ending at the nucleus q of a nerve cluster with center q with
degree of centrality λq satisfying the following condition:
For each pair of intermediate vertices in the filament skeleton skE, choose an edge
containing vertices that are closest to each other. This will sometimes result in over-
lapping filament skeletons. Give examples of filament skeletons in two different
social networks. “
Problem 2.14 Choose a set of seed points that are nodes representing the centroids of
stars in a galaxy. Implement the steps in Algorithm 7, using Delaunay triangulation of
a set of image centroids. For the Vertex selection step in Algorithm 7, use the nucleus
p of a maximal nucleus cluster in the triangulation. Let λ p be the wavelength of the
light emanating from of nucleus p. Choose a threshold th > 0. For the Filament
skeleton selection step, define a filament skeleton skE by drawing a sequence of
line segments starting at nucleus p and ending at the nucleus q of a nerve cluster
with center q with a wavelength λq of the light emanating from nucleus q satisfying
the following condition:
For each pair of intermediate vertices in the filament skeleton skE, choose an edge
containing vertices that are closest to each other. This will sometimes result in over-
lapping filament skeletons. Give examples of filament skeletons in two different
galaxies. “
A versatile form skeletal nerve results from attaching a combination of filament
skeletons and filled triangles to single vertex. The model for this form of skeletal
nerve is shown in Fig. 2.4d. An obvious place to look for this form of skeletal nerve in
triangulated visual scenes is a combination of those triangles and filament skeletons
that radiate out from the nucleus of a Delaunay nerve.
Example 2.15 (Hybrid skeletal nerve on a Delaunay triangulation of a Visual Scene)
An example of a hybrid form of skeletal nerve on a Delaunay triangulation is shown
in Fig. 2.7. Let S be set of seed points that are centroids on the Martian surface shown.
Each centroid is represented by a +. In this Delaunay skeletal nerve, one first selects
a seed point p of interest. The selected seed point will be the nucleus of a Delaunay
nerve, since every seed point in a triangulation is a vertex of a collection of filled
triangles. Attach to the selected seed point, one or more filled triangles and one or
more filament skeletons. Those selected triangles attached to p ∈ S form a triangle
2.6 Skeletal Nerves 99
skeleton skA. Similarly, those filament skeleton skeletons attached to p form another
filament skeleton skB. Each filament skeleton in skB begins with an edge > pq with
vertex p and a vertex that is the midpoint q of the edge opposite p. In each of the
filament skeletons in Fig. 2.7, one or more additional edges of interest are attached
to >pq. The interpretation of the phrase of interest depends on the application. As a
result,
skA ∩ skB = p,
the total mass m t of the photons reflected from visual scene surfaces and time t. That
is, there is a correspondence between each skeleton and the combined mass of the
photons giving rise to a vertices in a skeleton at a particular time. First, consider the
structure and length of a photon.
m ph ≈ 1.5 × 10−41 g. “
Photon Mass Compared with Electron Mass. Ryutov [7, B429] observes that
the upper bound of photon mass m ph is 22 orders of magnitude less than the
electron mass. In this case, 22 orders of magnitude equals 1022 . Let m be the
mass of an electron. Then m ph < 10122 m. “
Agarwal [8, p. 628] observes that the deviation of the photons of polarized laser
light on reflection is due to the force created by the mass of the photon at the contact
point of reflection. Force can only be created if a photon has mass. A photon of zero
mass cannot create any force at the contact point of reflection and will not deviate.
The mass of [a] spinning photon creates the force to turn at the contact point of
reflection resulting in the deviation of photon and change the direction of photon.
2
The energy of a system is measured in units of joules or kg · ms2 . Max Planck
introduced a constant h to simplify the representation of system energy. For Planck,
Planck’s constant
−34 m2
h = 6.6 × 10 kg · 2 .
s
Later generations of physicist’s refined Planck’s constant, using
2.7 Photon Energy and Skeletal Nerve Energy 101
Fig. 2.7 can span the vertices on a triangulated surface K so that there is a choice of
either seed points and non-seed points as end-vertices of the filament skeleton sk A
on skNrvA. By contrast, a triangle with the same vertex p in a skeletal nerve skNrvA
is limited in terms of that part of a triangulated surface covered by the triangle.
With a hybrid skeletal nerve such as the one in Fig. 2.7, we introduce a proximity
function on the filament skeleton that provides a means of comparing skeletal nerves
on different triangulated images. Let V be a set of vertices on a triangulated image
and let th > 0 be a threshold. For example, let λ p , λq be the hue wavelengths
of beginning vertex p and ending vertex q, respectively, on a filament skeleton
on the skeletal nerve skNrvA. Then consider the wavelength proximity function
f : V × V −→ R defined by
Also, let skNrvB be a hybrid skeletal nerve on a triangulated surface B and let
skNrvB contain a filament skeleton with end vertices p , q . Then skeletal nerve
skNrvA is considered similar to skeletal nerve skNrvB with respect to its filament
skeletons, provided
Filament Similarity Condition
f ( p, q) − f ( p , q )
< th.
And let K be the set of triangles in a skeletal nerves skNrvB. Then skeletal nerve
skNrvA is considered similar to skeletal nerve skNrvB with respect to its triangle
skeletons, provided
Putting these two similarity measures together, we arrive at a measure of the proximity
of a pair of skeletal nerves,2 defined by
2 Many thanks to M. Z. Ahmad for suggesting this formulation of the skeletal nerve similarity
measure.
2.9 Proximity of Skeletal Nerves 103
Problem 2.16 K
Using Matlab or a comparable tool, triangulate the frames X in a video displaying a
changing visual scene such as the one in Fig. 2.8. Implement the proximity functions
f ( p, q), g(K ) for the wavelength and triangle area proximity functions on X .
steps in Algorithm 7, using Delaunay triangulation of a set of image centroids. For
the Vertex selection step in Algorithm 7, use the nucleus p of a maximal nucleus
cluster in the triangulation. For the Filament skeleton selection step in Algorithm 7,
define a filament skeleton skE by drawing a sequence of connected line segments
starting at nucleus p and ending at the nucleus q of a nerve cluster with center q
furthest from nucleus p. “
104 2 Cell Complexes, Filaments, Vortexes and Shapes …
This section carries forward the notion of a vortex cycle that appears in Sect. 1.13,
introducing skeletal vortexes in the context of skeletal nerves in CW complexes.
In general, a vortex is a collection of nesting, non-concentric 1-cycles (also called
filament skeletons). In this work, a vortex filament is an oriented edge, i.e., an edge
that is either oriented in a single direction or is bi-directional. A filament oriented in
a particular direction in a skeletal vortex is the analog of a physical vortex filament
along which a particle moves in a particular direction. Such structures are of interest
in dynamical systems, particle physics, quantum vortex filaments (vortex filament
motion with drag in Abhinava and Guhaby [10, Sect. 3.2, p. 12ff]) and in fluid
mechanics (velocity along an edge in a spinning fluid such as water plunging down
a drain with the Coriolis effect, i.e., swirling in a direction conforming to the Earth’s
rotation, described by G.-G. Coriolis in 1835). For examples of vortex filaments, see
Cottet and Koumoutsakos [6, Sect. 3.2.3, p. 68ff]. Unlike common forms of non-
intersecting vortexes, filament skeletons in a skeletal vortex have a vertex in common.
In sum, a skeletal vortex on a triangulated surface is a collection of oriented filament
skeletons with at least one common vertex or at least one common edge.
K
skVB
skVA
skA1 p p
skB1
skA2 skB2
skB3
Each of the skeletal nerves in Fig. 2.4 represent the birth a skeletal vortex. Mindful
of the possibility of changes in the skeletons in a physical vortex, each first instance
of a skeletal vortex marks the birth of a vortex in spacetime. The evolution of a
skeletal vortex can be witnessed by tracking each vortex in a sequence of snapshots
in triangulated video frames.
A skeletal vortex A is a collection of filament skeletons with a common vertex
p (denoted by skVA), defined by
Proof Let skVA be a skeletal vortex. By definition, the filament vortices in skVA
have nonempty intersection. This gives the desired result.
Recall that a Delaunay nerve is a collection of filled triangles that have a vertex
in common. A skeletal vortex skVA differs from an ordinary Delaunay nerve NrvA,
inasmuch as skVA is restricted to filament skeletons that have a vertex p in common.
Skeltal Vortexes are constructed by repeated application of Algorithm 8 to obtain
a collection of filament skeletons that have a vertex in common for a set of seed
points on a finite, bounded surface region.
With skeletal vortexes modeling the behavior of light waves reflected off surfaces,
expansions and contractions of the vortices are common. This becomes apparent
in comparing what happens to light waves in a sequence of video frames. The light
reflected from visual scene surfaces is changing. There are many factors that influence
reflected light. Movements of objects such as tree leaves, animals and birds in a visual
scene, atmospheric disturbances, water vapor in the air, and surface temperatures are
common factors. The expansion of skeletal vortexes leads to collisions. Colliding
skeletal vortexes lead to the growth of skeletal nerves.
Example 2.19 (New Nerves from Colliding Skeletal Vortexes) The expansion of the
pair of skeletal vortexes skVA, skVB in Fig. 2.9 leads to collisions. For example, in
Fig. 2.10, the filament skeletons in this pair vortexes bump into each other at vertex
p. From Lemma 2.18, skVA, skVB are skeletal nerves by themselves. We have
The result of the collision the skeletal vortexes skVA, skVB is the advent of a new
nerve, namely, skeletal nerve skNrvE with nucleus p. “
We can expect the lifespan of a skeletal nerve due to colliding skeletal vortexes,
to be short. A new skeletal nerve such as the one in Example 2.11 will have a
short lifespan in the case where the expansion of the underlying filament skeletons
continues. In other words, a skeletal nerve resulting from colliding filament skeletons
persists only for a short time. This phenomenon can readily be observed in a sequence
of video frames reflecting changes in a visual scene.
K
skVB
skVA p
skA1 p
skB1
skA2 skB2
skB3
K
skVB
skVA p
skA1 p
skB1
skA2 skB2
skB3
In keeping with an interest in skeletal vortexes that model changing filament skele-
tons, it is often the case that filaments in a pair of colliding vortexes do not intersect.
This leaves the door open for colliding filaments in pairs of sub-vortexes to intersect,
sometimes.
Example 2.20 Taken as whole, the pair of vortex skeletons skVA, skVB in Fig. 2.11
do not form a single compound skeletal nerve, since the filament skeleton skB3
intersects skA1 but does not intersect skA2 . On the other hand, we obtain a new
skeletal nerve, provided we consider the intersection of filament skeletons skB1 , skB2
with skVA, i.e.
Example 2.21 A sample frame from a video of a cat playing the pieces in the Chi-
nese board game Mahjong ( )3 is shown in Fig. 2.12. The pair of vortex skele-
tons skVA, skVB in Fig. 2.12 do not form a single compound skeletal nerve, since
the filament skeleton skB3 does not intersect sk A3 . On the other hand, we obtain
a new skeletal nerve, provided we consider the intersection of filament skeletons
{skB1 , skB2 } with {skA1 , skA2 }, i.e.
3 Many thanks to Enze Cui for this tessellation of the video frames for this example.
108 2 Cell Complexes, Filaments, Vortexes and Shapes …
Collections of filament skeletons that intersect the same cell complex are called
Gemini complexes. That is, a Gemini complex A (denoted by skGA) is a collec-
tion of filament skeletons looping over a finite bounded planar surface and which
are attached to same cell complex. For example, skGH = {skA, skB} is Gemini
complex, provided filament skeleton skA intersects a cell complex E, which also
intersects filament skeleton skB. In the case where a pair of filament skeletons have
a cell complex in common, we have strong Gemini skeletons. That is, a strong
Gemini complex is a collection of filament skeletons that have a cell complex in
common. For example, the Gemini complex skGH = skA , skB is a strong
Gemini complex, provided skA ∩ skB = E for some cell complex E .
Example 2.22 (Gemini skeletons) There are several instances of Gemini skeletons
in the crowd scene4 in Fig. 2.13. Let skA, skB, skE, skG, skH be filament skeletons
as shown in Fig. 2.13. We have
skG ∩ skH
= ∅ skA, skE have an edge in common.
skA ∩ skB = ∅ non-intersecting Gemini filament skeletons.
skA ∩ skE
= ∅ skA, skE have an edge in common.
skB ∩ skE
= ∅ skB, skE have an edge in common.
That is, skG, skH are strong Gemini skeletons, since this pair of filament skeletons
have an edge in common. Similarly, skA, skE and skE, skB are strong filament
skeletons. The filament skeletons skA, skB are Gemini skeletons, since both skele-
tons intersect skE. However, sk A, skB have nonempty intersection. Hence, sk A, skB
are not strong Gemini skeletons. “
From a proximity and shape classification perspective, the detection of both Gem-
ini and strong Gemini filament skeletons on either triangulated or tessellated surfaces
is important. From the features of both types of Gemini complexes, we can begin to
characterize surface shapes covered by the skeletons.
We can use the fact that a filament complex is itself a cell complex to obtain the
following result.
Proof Filament skeletons in a strong Gemini complex intersect the same filament
complex. By definition, a filament skeleton is a cell complex. Hence, the desired
result follows.
From Example 2.22, we find that although skA, skB in Fig. 2.13 are Gemini
skeletons, they do not have common parts, i.e., skA, skB are not filament skeletons
in a strong Gemini complex. Hence, the converse of Theorem 2.23 does not hold.
Example 2.24 (Strong Gemini skeleton) There are many instances of strong Gemini
complexes in Fig. 2.12. Here are two examples
Problem 2.25 K
In an implementation, select a set S of seed points in a colour image X . Tessel-
late X and identify the maximal nucleus clusters (MNCs) on the tessellated image.
Experiment with S until more than one MNC is found in the same image. The crowd
scene in Fig. 2.13 is an example. Draw the filament skeletons on the MNCs found.
Highlight
1o Pair of skeletons in a strong Gemini complex.
2o Pair of Gemini skeletons that are not strong Gemini filament skeletons. “
Problem 2.26 ®
Repeat the steps in Problem 2.25 for the triangulation of a colour image X based on
a selected set of seed points in the colour image. “
Problem 2.27 ®
Repeat the steps in Problem 2.25 for the tessellation of the frames in a video based
on a selected set of seed points in the video frames. “
Problem 2.28 ®
Repeat the steps in Problem 2.25 for the triangulation of the frames in a video based
on a selected set of seed points in the video frames. “
Notice that strong Gemini complexes take us in the direction of nerve complexes,
again.
2.13 Gemini Complexes and Gemini Nerve Structures 111
An oriented filament skeleton is a filament skeleton with vertices that have a par-
ticular ordering. This ordering can be either one-directional or bi-directional. For
example, in a planar oriented filament skeleton, there is what Alexandroff terms a
particular sense of rotation [11, Sect. II.13, p. 13].
The observer sees itself moving towards (mapped to) e4 (denoted by p −→ e4 ),
starting the following motions:
p −→ e4 ,
e4 −→ e3 ,
e3 −→ e2 ,
e2 −→ e1 ,
e1 −→ p.
K
skVB
a1 a2
skVA p b3
b4
a0 p
skA1 b1 skB1 b2
a5 skA2 a4 skB2
skB3
0 p = 1 · e4 := 1
= 1 + 1 · e3 := 1 + 1,
= 1 + 1 + 1 · e2 := 1 + 1 + 1,
= 1 + 1 + 1 + 1 · e1 := 1 + 1 + 1 + 1,
= (1 + 1 + 1 + 1 + 1) p := (5)mod 5 + 0 p := 0 + 0 p,
= 0p
In other words, the observer at a nucleus p sees itself as part of a cyclic behaviour.
And all members (vertices) of the cycle are path-connected to itself. In fact, the
rotating motion of a filament skeleton in a skeletal nerve such as skVA defines the
nucleus of the nerve. In the sequel, we will find that p is the generator of a cyclic
group. For more about this, see Sects. 3.6 and 3.17. “
Example 2.30 A pair of oriented skeletal vortex filament skeletons skA2 , skB1 with
common vertex p, is represented in Fig. 2.14. We have
skA2 := {a0 , a1 , a2 , p, a4 , a5 , p} .
skB1 := p, b1 , b2 , b3 , p , b4 . “
A principal interest in this study of the cyclic behavior of skeletal nerves is the light
wave propagation evident in the sequences of vertices in oriented filament skeletons
that define the nerve structures of shapes in triangulated video frames. The reflection
of light from the surface of an object in a changing visual scene is concentrated in
the interiors of shapes recorded in each video frame. That part of a visual scene that
2.14 Oriented Filament Skeletons 113
has the highest density of surface punctures (holes) is found by the concentration
of the centroids of the holes, one of which will be the nucleus in maximal nucleus
clusters (MNCs) easily detected in the triangulated surfaces.
Dennis [19] on light as a vector disturbance, Planck spectrum, and the photon
interpretation of polarization of light fields.
Shape: The study of shape is a huge topic with many surprising hills (ranging from
art to retracts theory) and valleys (the tricky interpretation of the role of photons
in the creation of video frames). See, for example:
Borsuk 1970 Lectures on the theory of shape [20], on retracts, on the close con-
nection between topology and geometric intuition.
Borsuk and Dydak 1980 paper on the shape theory [21], a survey of the most
important notions, results and problems in the theory of shape.
Ghrist on a not-for-the-faint-hearted applied topology [22], which includes topo-
logical data analysis, leading to the introduction of homology barcodes (pic-
tographs), a single descriptor for topological evolution and what is known as
persistent homology (read shape persistence over time).
Ghrist [23] on visualizing the shape of point-cloud data with pictographs called
homology barcodes.
Peters and Ramanna [24], on shape detection, image shape geometry, spatial and
descriptive proximities of shapes.
Peters [25] on proximal planar shapes and the correspondence between triangu-
lated shapes and nerve complexes.
Peters [26], on defining shape barcodes, short history of topological nerves and
homology nerve complexes.
Ahmad and Peters [27], on spoke complexes, maximal nucleus clusters (MNCs),
shape signatures and approximating image object shapes.
Surface Holes and Non-Holes that Define Shapes: There is an apparent asym-
metry between holes and non-holes in physical shapes, which we witness in most
visual scenes. This asymmetry is analogous to the contrast between baryonic mat-
ter (i.e., matter that includes protons, neutrons and all the objects composed of
atomic nuclei) characterized by positive electric charge and baryonic anti-matter
characterized by negative electric charge. For a good introduction to the matter-
antimatter asymmetry in elementary physical systems, see Sasso [28]. Recall that
a shape is a finite bounded surface region with a nonempty interior. The paradox
here is that a shape acquires it character and its definition from the holes in its
interior. A physical surface without a hole reflects light that varies, depending
on the surface material and curvature. By contrast, a physical surface with a hole
absorbs light. A physical surface with a boundary and with an interior that is partly
non-punctured (non-hole) and partly punctured (sprinkled with holes), defines a
shape.
2.15 Sources, References and Additional Reading 115
References
1. Oudet, X.: Light as flow of photons. Technical Report, Université Paris-Sud (2018). https://
www.researchgate.net/profile/Xavier_Oudet
2. Cooke, G., Finney, R.: Homology of cell complexes. Based on lectures by Norman E. Steenrod.
Princeton University Press and University of Tokyo Press, Princeton, NJ, USA and Tokyo, Japan
(1967). xv+256pp., MR0219059
3. Krantz, S.: A Guide to Topology. The Mathematical Association of America, Washington,
DC (2009). ix+107pp., The Dolciani Mathematical Expositions, 40. MAA Guides, 4. ISBN:
978-0-88385-346-7, MR2526439
4. Willard, S.: General Topology. Dover Pub., Inc., Mineola, NY (1970). Xii+369pp. ISBN: 0-
486-43479-6 54-02, MR0264581
5. Naimpally, S., Peters, J.: Topology with Applications. Topological Spaces via Near and Far.
World Scientific, Singapore (2013). Xv+277pp., Amer. Math. Soc., MR3075111
6. Cottet, G.H., Koumoutsakos, P.: Vortex Methods. Theory and Practice. Cambridge University
Press, Cambridge, UK (2000). xiv+313pp. ISBN: 0-521-62186-0, MR1755095
7. Ryutov, D.: Using plasma physics to weigh the photon. Plasma Phys. Control. Fusion 49,
B429–B438 (2007). https://doi.org/10.1088/0741-3335/49/12B/S40
8. Agarwal, N.: Experimental proof of mass in photon. J. Mod. Phys. 6(5), 627–633 (2015).
https://doi.org/10.4236/jmp.2015.65068(Opensource)
9. Susskind, L., Friedman, A.: Quantum Mechanics. The Theoretical Minimum. Penguin Books,
UK (2014). xx+364pp. ISBN: 978-0-141-977812
10. Abhinava, K., Guhaby, P.: Inhomogeneous Heisenberg spin chain and quantum vortex filament
as non-holonomically deformed NLS systems. arXiv 1703(02353v3), 1–15 (2017)
11. Alexandroff, P.: Elementary Concepts of Topology. Dover Publications, Inc., New York (1965).
63pp., translation of Einfachste Grundbegriffe der Topologie (Springer, Berlin, 1932), trans-
lated by Alan E. Farley , Preface by D. Hilbert, MR0149463
12. Dager, R., Zuazua, E.: Wave propagation, observation and control in 1-D flexible multi-
structures. Math’ematiques & Applications (Berlin) (Mathematics & Applications), vol. 50.
Springer-Verlag, Berlin (2006). x+221pp. ISBN: 978-3-540-27239-9, MR2169126
13. Gupta, S., Srivastav, S.: Matlab program for energy of some graphs. Int. J. Appl. Eng. Res.
12(20), 10145–10147 (2017). ISSN 0973-4562
14. Jänich, K.: Topologie (Topology), 8th edn. Springer, Berlin (2005). x+239pp. ISBN: 978-3-
540-21393-2, MR2262391 (in German)
15. Hatcher, A.: Algebraic Topology. Cambridge University Press, Cambridge, UK (2002).
xii+544pp. ISBN: 0-521-79160-X, MR1867354
16. Rowland, T.: Manifold. Wolfram MathWorld (2018). http://mathworld.wolfram.com/
Manifold.html
17. Peters, J.: Computational proximity. Excursions in the topology of digital images. Intell.
Syst. Ref. Libr. 102 (2016). Xxviii+433pp. https://doi.org/10.1007/978-3-319-30262-1,
MR3727129 and Zbl 1382.68008
18. Taylor, A.: Analysis of quantised vortex tangle. Ph.D. Thesis, University of Bristol, Bristol,
England (2017). Supervisor: M. Dennis
19. Dennis, M.: Topological singularities in wave fields. Ph.D. Thesis, University of Bristol, H.H.
Wills Physics Laboratory, Bristol, England (2001). Supervisor: M. Berry, http://www.bris.ac.
uk/physics/media/theory-theses/dennis-mr-thesis.pdf
20. Borsuk, K.: Theory of Shape. Monografie Matematyczne (Mathematical Monographs), vol.
59. PWN—Polish Scientific Publishers (1975). MR0418088, Based on K. Borsuk, Theory of
shape, Lecture Notes Series, No. 28. Matematisk Institut, Aarhus Universitet, Aarhus (1971),
MR0293602
21. Borsuk, K., Dydak, J.: What is the theory of shape? Bull. Austral. Math. Soc. 22(2), 161–198
(1980). MR0598690
22. Ghrist, R.: Elementary Applied Topology. University of Pennsylvania (2014). Vi+269pp. ISBN:
978-1-5028-8085-7
116 2 Cell Complexes, Filaments, Vortexes and Shapes …
23. Ghrist, R.: Barcodes: the persistent topology of data. Bull. Amer. Math. Soc. (N.S.) 45(1),
61–75 (2008). MR2358377
24. Peters, J., Ramanna, S.: Shape descriptions and classes of shapes. A proximal physical geometry
approach. In: Stańczyk, B., Zielosko, U., Jain, L. (eds.) Advances in Feature Selection for Data
and Pattern Recognition, pp. 203–225. Springer (2018). MR3895981
25. Peters, J.: Proximal planar shapes. Correspondence between triangulated shapes and nerve
complexes. Bull. Allahabad Math. Soc. 33, 113–137, : MR3793556, Zbl 06937935. Review
by D, Leseberg (Berlin) (2018)
26. Peters, J.: Proximal planar shape signatures. Homology nerves and descriptive proximity. Adv.
Math. Sci. J. 6(2), 71–85 (2017). Zbl 06855051
27. Ahmad, M., Peters, J.: Proximal Cech complexes in approximating digital image object shapes.
Theory and application. Theory Appl. Math. Comput. Sci. 7(2), 81–123 (2017). MR3769444
28. Sasso, D.: Inhomogeneous Heisenberg spin chain and quantum vortex filament as non-
holonomically deformed NLS systems. viXra 1411(0413v2), 1–18 (2017)
Chapter 3
Shape Fingerprints, Geodesic Trails and
Free Abelian Groups on Skeletal Vortexes
Abstract This chapter takes another look at filament skeletons, skeletal vortexes and
skeletal nerves in cell complexes. The focus here is on the group theory underlying a
Computational Topology of digital images (CTdi). A digital image is an example of
what is known as a shape space. A space is any nonempty set of points. A shape space
is a collection of sets of points X and each particular configuration (arrangement of
the points) in a subset of X defines a shape. A digital image shape space is a
collection of digitized optical sensor values that provide a record of the hue angles
of pixels in the Hue Saturation Value colour space. There is a 1-to-1 correspondence
between the pixel hue angles and the wavelengths of light reflected from the surfaces
in a visual scene at a given instant in spacetime. It is this 1-to-1 correspondence that
leads to a deeper view of skeletal complexes on triangulated video frames.
Shapes of spaces were introduced by Zomorodian [1, Sect. 1.2, p. 3]. Many shapes
of digital images that CTdi reveals are largely hidden from casual examination of
the images. This hidden character of digital images motivates the decomposition
of the images into recognizable geometric shapes that make it possible to measure
and compare shapes in image subregions. This decomposition can be accomplished,
for example, by triangulating on selected points on an image shape space. CTdi
includes a provision of algorithms (methods) that give the steps to get specific results
concerning image object shapes. This chapter also introduces Betti-Nye vortex cycles
on visual scenes and a whimsical view of persistent Betti numbers. A selection of
shape complex, skeleton and other symbols used in this chapter are given in Table 3.1.
A skeleton would start with a starting vertex v0 connected to a vertex v1 and the
length of the segment >v0 , v1 so that every other segment in skeleton sk A is a multiple
of this initial segment, which is a generator a (denoted by a) defined on an ordered
set of k path-connected vertices in skA, namely,
k ordered vertices
V = {v0 , v1 , . . . , vi , . . . , vk−1 } .
generator a = minimum filament-length
a = v1 − v0 .
3.2 Discovering Generators of Oriented Filament Skeletons … 119
and moving in the reverse direction along the same filament leads to
m i copies of −a
(−a) + (−a) + · · · + (−a) = −m i a.
maps to
>
v0 , v1 = −→ 1a.
maps to
>
v0 , v1 + >
v1 , v2 = −→ 1a + m 1 a.
....
..
maps to
> >
v0 , v1 + · · · + vk−2 , vk−1 −→ 1a + · · · + m k−1 a.
The + between path edges reads attach to. A path is formed by starting with an
edge >v0 v1 , where vertex v0 is the beginning of the leading edge in the path and >v1 , v2
is the next edge in the set of ordered vertices in a filament skeleton. Recall that the
edge >v0 , v1 is called a filament. The term filament is used to describe an edge in a
vortex cycle and to call attention to the physical side of things in triangulated pictures
(visual scene snapshots). By writing a multiple m i a of the generator a, we ignore
the varying fractions of the generator distance in any filament length and focus on
summing the multiples of the generator length.
For every filament v> >
i−1 vi in skA, there is a corresponding vi vi−1 in the reverse
direction (we allow traveling in either the forward or reverse direction on the skele-
ton). Hence, for every m i a, there is an additive inverse −m i a. That is, we can always
write m i a − m i a = 0. Notice also that there is a zero motion associated with every
set of oriented, path-connected vertices. When we write
zero motion
>
vi−1 vi + 0 −→ m i a + 0 = m i a (0 = additive identity).
Example 3.1 The exterior and interior of a pair of vortexes on the surface of a seashell
are represented in Fig. 3.1a, b, respectively.1 From a computer vision perspective,
the dark regions in these images are holes (surface regions that absorb light (flow
of photons) colliding with the surface). The white regions in these images are those
parts of the surface the reflect light (the so-called object regions).
To learn more about surface shapes such as the seashell vortex in Fig. 3.1a, we
do the following.
1o Select a set of seed points S containing the centroids of some of the surface
object regions.
2o Perform Delaunay triangulation on the points in S. The end result is a cell
complex K that is a collection of filled triangles s (2-cells). The vertices of
each triangle are seed points in S that are nearest each other on the seashell
surface.
3o Find one or more maximal nucleus clusters (MNCs) on K . Recall that a trian-
gulation MNC is a maximal collection of triangles with a common vertex.
4o Find the barycenters of the MNC triangles and the triangles on the triangles
bordering the MNC.
5o Draw a filament skeleton skA that represents a set of ordered barycenters that
are path-connected on skA.
2 Many thanks to M. Z. Ahmad for supplying the Matlab script used to draw a barycentric filament
skeleton.
122 3 Shape Fingerprints, Geodesic Trails and Free Abelian Groups …
a 0 = e,
a n = a 0 ◦ a 1 ◦ a 2 ◦ · · · ◦ a n−1 = e, and
a i ◦ a j = a i+ j , if i + j ≤ n, and
a i ◦ a j = a i+ j−n , if i + j > n.
Herstein [13, Sect. 2, p. 29] observes that a geometric realization of a cyclic group
with n elements is to let a group generator be a rotation through an angle of 2 πn on the
circumference of a unit circle. The binary operation ◦ is abelian, since a ◦ b = b ◦ a
for all members a, b in G. In CTdi shape theory, matters are simplified by letting
the operation ◦ be addition modulo n, i.e., ◦ = +modn. For (a + b)mod n, recall
that addition modulo n equals the remainder after division of the sum by n. So, for
example, let (G, +mod 2) be a set of integers that includes 5, 8. Then
The section gives a brief introduction to CTdi from a picture shape analysis per-
spective. The terms picture and digital image are used interchangeably. By digital
image, we mean a raster image. For many applications such as gaming, the focus on
images shifts from raster images to vector images. Commonly used symbols in this
chapter and beyond are given in Table 3.1. A complete list of symbols used in this
book is given at the beginning of the Subject Index at the end of this volume.
The story starts with the decomposition of a picture (aka digital image) into a
collection of triangles connected between selected keypoints on a picture. From
such a decomposition of a picture into a collection of connected vertexes, edges and
triangles, we gain access to some hidden stuff in a picture. A direct result of the
triangulation of a picture is simplicial complex, which is a covering of a picture
with a collection of non-overlapping triangles uniquely determined by their vertices.
A picture X is covered by a cell complex A, provided X ⊆ A, provided the set
of picture points is a subset of the complex A. A cell complex that results from
a triangulation of a finite bounded region is called a complex (aka, filled trian-
gle complex). Hiding in most pictures with modest complexity is a collection of
tiny as well as big shapes caught in the mesh of triangles that we paint onto each
picture.
This section introduces some of the basic building blocks in a computational topology
of digital images, starting with cells and cell complexes in a -complex (sequences
of connected 1-cells painted on a picture shape). From basic structures such as ver-
tices and edges on picture shapes, we then gain access to richer structures such as
sequences of connected closed arcs (called closed 1-cells) along a shape boundary
or sequences of connected arcs between open disks (called 2-cells) or the tiniest of
picture shape structures called 0-cells (vertexes, aka vertices). It is an easy next step
to start considering chains of arcs (1-complexes) with each chain linking one vertex
(0-cell) to another vertex via a sequence of arcs between a pair of vertices.
Briefly, a chain is a finite collection of cell complexes expressed as a formal
sum. From this, we obtain a chain complex containing a connected path between
vertices. In the case where a chain complex contains a simple, closed connected path,
picture regions are then surrounded by measurable paths. For more about chains, see
Appendix A.3. In unraveling the hidden structures in picture shapes, we are mainly
interested in what are known as 1-chains. A 1-chain is a formal sum of connected
arcs (1-cells) in a -complex.
For planar shapes, the geometry of a shape is found by covering a shape with cell
complexes. A cell complex (denoted by -complex) is a collection of connected
cells, which are constructed from splicing together vertices and edges along paths
called cycles in an image. For each given image shape, the basic approach is to start
with a set of 0-cells (vertices) and construct a cell complex by attaching edges (1-
126 3 Shape Fingerprints, Geodesic Trails and Free Abelian Groups …
cells) to each of the vertices. After we have obtained a -complex, we can either
determine the geometry of a image object shape in terms of the cellular covering of
a shape or we can begin tracing different types of paths determined by sequences
of edges from one vertex to another vertex in the -complex. Every -complex
contains different types of paths useful in classifying shapes.
We are interested in the construction of cycles (simple, closed, connected paths
that lie on a planar shape). The motivation for doing this is that we arrive at a tractable
approximation of some part of a shape. Shapes themselves can be very complex. This
is true of even the simplest shapes.
Shape fingerprint.
K By constructing cellular cycles on shapes, we obtain a fingerprint of a
shape, something characteristic of a shape that results from the restrictions of
the boundaries of each shape. A shape fingerprint is a distinctive connected
path from one part of a shape to another part of a shape. Shape fingerprints tell
us about a shape without our having to look at every part of the shape. Here is
a conjecture to try to prove or disprove.
Conjecture 3.3 Every physical object has at least one unique shape finger-
print. “
Conjecture 3.4 Every digital image has at least one unique shape fingerprint. “
defined by
(a) (b)
Fig. 3.4 Constructing shape cycles from sequences of connected closed 1-cells
The result of a Face-2-face map is a closed arc, an edge in a shape cycle. Grafting
one closed arc onto another one is accomplished by a sequence of Face-2-face maps.
Let v be a 0-cell as shown in Fig. 3.4b. Then the intersection of cell maps produces
> >
a simple closed cycle (path) fragment vv ∪ v v shown in Fig. 3.4b, namely,
Recall that the intersection of two sets A, B is the set of all points
common
to the
two sets (represented by A ∩ B). In our case, each Face-2- f ace( v, v ) is a set of
points in a closed arc. Each pair of connected arcs has a 0-cell (vertex) in common.
In this case, the vertex v is common to the pair of Face-2-face maps.
Continuing this sequence of Face-2-face maps, we obtain the cycle labelled a,
which is a connected path with a traversal that begins and ends in vertex v. The
end result of a complete sequence of Face-2-face maps is a shape cycle with edges
labelled 1a, 2a, 3a and 0a (end of cycle). In effect, a traversal of this cycle has the
appearance of a spinning clock hand on a 4-h clock (the count starts over when we
complete the traversal of the segment labelled 0a). “
Navigating back and forth along a closed connected path (what we are calling a shape
cycle) requires bi-directional 1-cells (aka oriented arcs). Recall that a bi-directional
>
oriented arc (also, oriented edge) vv is an arc that can be traversed from end point
v to the other end point v and can also be traversed from v to v. The spin behaviour
of an oriented arc entails a traversal of the arc in the forward direction, followed
by a traversal of the edge in the reverse direction. Although the terms arc and edge
are interchangeable, arcs are preferred over edges, since picture edges (and all other
edges in the physical world) tend to be curved rather than straight. The notion of a
straight edge is a carry-over from Euclidean geometry, which is a simplified view of
what a surveyor sees through a transit-level.
Example 3.6 (Spinning on Oriented Closed Arcs) Two forms of oriented arcs are
shown in Fig. 3.5, namely,
>
1o A single oriented closed arc vv (labelled 1a in the positive direction and −1a in
the reverse (negative) direction) is shown in Fig. 3.5a. A rest state is part of every
spinning behaviour. In the spinning behavior of an oriented arc, a rest state is a
state in which there is no spinning. A rest state (no spinning) is represented by
0a (not shown in Fig. 3.5a). A spin on this arc is carried out by traversing the
arc from vertex v to v by following the arrow −→ and then traversing this arc
from vertex v back to v by following the arrow −→. The complete spinning
behaviour of an oriented arc is represented by the set
(a) (b)
0a (rest state) :
traverse arc 1a traverse arc −1a
−→ −→ :
0a (rest state).
> >
2o A pair of connected oriented arcs vv (labelled 1a) and v v (labelled 2a) are
shown in Fig. 3.5b. Traversal in the reverse direction from v to v (labelled −2a)
and then from v to v (labelled −1a) is also shown in Fig. 3.5b. A spin on this
pair of connected arcs is carried out by traversing the arc from vertex v to v by
following the arrow −→ and then traversing this arc from vertex v back to v
by following the arrow −→. The complete spinning behaviour of an oriented
arc is represented by the set
0a (rest state) :
traverse arc 1a traverse arc 2a
−→ −→
traverse arc −2a traverse arc −1a
−→ −→ :
0a (rest state).
The spinning behavior on oriented arcs can be compressed into table form. This is
done by viewing each traversal along an edge as a form of addition modulo 2 on the
coefficients of the arc labels. Let a and b be integers. Recall that addition (a + b)
modulo 2 equals the remainder after division of a + b by 2. Then, for example, we
express the complete spinning behavior on a single oriented arc in the following
tables.
+mod2 0a 1a −1a +mod2 0 1 −1
0a 0a 1a −1a 0 0 1 −1
1a 1a 0a −0a maps
to 1 1 0 −0
−1a −1a 0a −0a −→ −1 −1 0 −0
of these tables. It is also the case that the +mod 2 addition operation on a sequence
of connected oriented arcs gives us two simple examples of cyclic groups. “
Problem 3.7 ®
Express the spinning behavior of the pair of connected arcs in Fig. 3.5b in table form
using addition modulo 2. “
In effect, the mapping ‡ glues together closed arcs to form connected closed arcs. A
cell complex is a collection of connected closed arcs that define connected paths on
a shape surface. Cell complexes are roadways that we can travel over from one part
of a shape to another part of a shape. If a connected closed arc permits a traversal that
begins and ends at the same vertex, then that connected closed arc defines a simple
closed path called a cycle. A cell complex that is is a single cycle is constructed as
shown in Algorithm 9.
3.7 Construction of Shape Cycles in Cell Complexes 131
Recall that the C in CW topology reads closure finite, since we are dealing with
closed arcs on cycles. The W reads Weak topology, since we are usually only con-
cerned with intersecting arcs on picture shape cycles. This is considered a weak
topology, since the union of open sets in an ordinary topology is not part of what
we consider. The main reason for this is that intersecting cycles yield new forms of
cyclic groups, whereas the union of shape cycles generally does not yield a cyclic
group. Each cycle gets the name cell complex, because we work with cells that are
open arcs in a simplified form of picture topology in constructing shape cycles.
This is an important shift from the inspection of vertices in a triangulation of a
picture to 0-cells that are the basic building blocks in simple closed connected paths
(cycles) on picture shapes. Similarly, a 1-cell is an open arc (or open edge, i.e., an
edge without its endpoints) in a chain takes on a new character as the glue in forming
cell complexes that are simple closed paths on a picture shape. The main motivation
in shifting our attention from vertices and edges on a triangulated picture shape to
0-cells as the endpoints of 1-cells in simple connected paths is that the characteriza-
tion, comparison and classification of shapes in pictures is simplified. That is, with
triangulation, the focus is on covering a shape with triangles and detecting collec-
tions of intersecting triangles with a common vertex (called Alexandroff nerves).
By contrast, the covering condition goes away with cell complexes that form cycles
(simple, closed, bounded, connected paths) swimming around on picture shapes. In
effect, we provide a simple means of detecting shape fingerprints by considering
cycles on shape surfaces.
132 3 Shape Fingerprints, Geodesic Trails and Free Abelian Groups …
With the construction of closed arcs that are cycles, we gain a means of isolating
interior regions of shapes caught within the boundary of a cell complex (aka chain
complex). Two types of cycles on shapes are commonly considered in the study of
image object shapes, namely,
Cycles that are Boundaries of Holes: A shape hole is a bounded image region
containing pixels with uniform intensity. This means that shape holes have many
appearances. Holes are uniformly dark or uniformly light image regions. The
edges of the triangles in a triangulated shape are a convenient source of approxi-
mate shape boundaries.
Example 3.8 (Shape Hole Boundary) A sample kangaroo shape hole Ho boundary
(denoted by bdyHo ) is shown in Fig. 3.6a. The boundary bdyHo is defined by the
edges of a triangle. “
Cycles that are not Boundaries of Holes: A cycle on a shape A (denoted by
cyc A) is a simple, closed, connected path that either surrounds a hole with a
boundary or does not surround a shape hole.
Example 3.9 (Shape Cycle) A sample shape cycle cyc A is shown in Fig. 3.6b. In
this example, cyc A contains no visible holes in its interior. “
It is important part of CT to consider only cycles surrounding holes as boundaries.
This makes sense, since a hole is basically an empty space with a edge. In declaring
that a cycle surrounding a hole to be a boundary, we are usually dealing with an
approximation of the edge of a hole. Other cycles that sweep through the shape
space of an image are not boundaries of holes.
bdyHo
cycA
cycB
Example 3.10 (Shape Cycle Surrounding Multiple Holes) A shape cycle cycB sur-
rounding several picture holes is shown in Fig. 3.7. In this example, cycB contains
visible holes in its interior, namely, the ear, eye and mouth holes in the kangaroo
shape. “
The end result of triangulating a picture is the division of a picture into separate
regions containing image shapes. The basic idea is to make a transition from merely
limiting our view of a triangulated picture to collections of triangular picture regions
to a close look at the interior of picture triangles, which are open sets, i.e., trian-
gle interiors without the bounding edges, which is where the topology of digital
images starts. For more about chains of arcs, see Flegg [15, p. 43]. The approach
can be viewed as a succession of maps (projections of picture points to vertices,
arcs and intersection triangles): -complex is a covering of a shape shA (denoted by
cov(shA)), provided (Fig. 3.8)
i.e., shape shA is entirely in the interior of the cover (is a proper subset of) cov(shA)
or shA occupies the same region of the plane as the cover cov(shA).
A map is a correspondence between objects such as a picture and its triangulation.
Every -complex covering an image object shape shA contains structures called
nerve complexes (denoted by NrvA).
Definition 3.11 (Alexandroff Nerve Complex) Let shA be a planar shape covered by
-complex cov(shA) and let v be a vertex in the cover cov(shA). Then an Alexandroff
134 3 Shape Fingerprints, Geodesic Trails and Free Abelian Groups …
NrvA
N rvB
N rvG
Definition 3.15 (Open Ball) Let shA be planar shape covering a cell complex K
with a vertex c ∈ K and let radius r > 0. A simplicial ball Br (c) is defined by
open ball
Br (c) = {y ∈ shA : c − y < r } . (Open Ball).
Br (c) is an open ball, since Br (c) does not have a bounding edge. “
Example 3.16 (Sample Open Balls) A pair of open balls partially covering of a kan-
garoo shape shA is shown in Fig. 3.11. Ball Br (c) in Fig. 3.11a partially covers the
head of the kangaroo, which offers a simplified way to compare one such head shape
with another one. This can be done by using the area of a ball a crude approximation
of a head shape. By contrast, Br (c) in Fig. 3.11b with radius r > r and vertex
center c covers a larger portion of shape shA. “
Depending on the application, balls that include their boundaries (i.e., closed
balls) are useful in analyzing cell complexes.
(a) (b)
(a) (b)
Definition 3.18 (Closed Ball) Let shA be planar shape covered a cell complex K
with a vertex c ∈ K and let radius r > 0. A closed ball denoted by cl(Br (c)) is
defined by
Closed Ball
cl(Br (c)) = {y ∈ shA : c − y ≤ r } .
cl(Br (c)) is a closed ball, since cl(Br (c)) includes its bounding edge. “
Example 3.19 (Sample Closed Balls) A pair of closed balls partially covering a
kangaroo shape shA is shown in Fig. 3.12. Ball cl(Br (c)) in Fig. 3.12a partially
covers the head of the kangaroo, which offers a simplified way to compare one such
head shape with another one in terms of the distances of the parts of a head to the
bounding edge or to the interior of the closed ball. Again, this can be done by using
the area as well as the bounding edge of the closed ball as a crude approximation of
a head shape. By contrast, cl(Br (c)) in Fig. 3.12b with radius r > r and vertex
center c covers a larger portion of shape shA. “
(a) (b)
From the perspective of a triangulated planar shape with round edges, a C̆ech nerve
is superior to an Alexandroff nerve, since the outer edges of the balls in a C̆ech nerve
tend to conform more readily to the shape edges.
Example 3.20 (Sample C̆ech nerves) Let K be a cell complex covering the kan-
garoo shape in Fig. 3.13a. A sample C̆ech nerve C ech r (K) partially covering a
triangulated kangaroo shape shA is shown in Fig. 3.13a. Selected vertices in K are
a source of centers for the balls in the C̆ech nerve. In Fig. 3.13a, each ball center is
indicated with a •. In constructing a C̆ech nerve, it is necessary to identify all balls
that intersect (have one or more points in common). A second sample C̆ech nerve
C ech r (K) with a radius r > r is shown in Fig. 3.13b. This C̆ech nerve completely
covers the kangaroo head. For this reason, C ech r (K) is more interesting than the
smaller C̆ech nerve in Fig. 3.13a. “
3.11 Multiple Balls in a Cech Nerve 139
C̆ech Complex
cxC̆echr (S) = C̆echr (S) : C̆echr (A) = ∅ .
A∈2 S
Recall that each C̆ech nerve C̆echr (S) is a collection of intersecting balls, each its
own center and with radius r . Here, the focus shifts from the triangulation of vertices
in constructing a cell complex, to the triangulation of the centers of the balls in the
collection of intersecting C̆ech nerves.
Example 3.21 (Sample C̆ech Complex) Let S be a set of vertices (each vertex in
S is represented by a red •. A sample C̆ech complex on S partially covering the
kangaroo shape is shown in Fig. 3.15. This C̆ech complex cxCech r (S) contains
three overlapping C̆ech nerves, i.e.,
cxCech r (S) = {Cech r (A1 ), Cech r (A2 ), Cech r (A3 )} = ∅,
3
Fig. 3.15 C̆ech complex cxC ech r (S) = C ech r ( Ai ∈ 2 S )
i=1
This section introduces trails of edges (paths) between shapes. A trail is a sequence
of connected edges whose union is homeomorphic to a single line segment spanning
between the initial vertex and ending vertex in a trail, introduced by Boltyanskiĭ and
Efremovich [17, Sect. 1.4, p. 11].
Let X, Y be nonempty sets. Recall that a mapping h on X into Y (denoted by
h : X −→ Y ) has the property that for each x ∈ X there corresponds exactly one
y ∈ Y . Let a δ b read a is close to b, for elements a, b ∈ X . The set Y is called
the range (also image) of the mapping h. The set X is called the domain (also
pre-image) of the mapping h.
A mapping h : X −→ Y is continuous, provided, for elements a, b ∈ X ,
whenever a δ b (closeness of a and b), then h(a) δ h(b) (closeness of h(a) and h(b)),
i.e. h is continuous, h maps close points in X to close points in Y . The notation
h(X ) reads h(X ) equals a subset of Y , provided h maps X into Y . In the into case,
h(X ) ⊂ Y . If h maps X onto Y , then h(X ) = Y . Onto mappings are called surjective
mappings.
A mapping h : X −→ Y is one-to-one written 1-1 (also injective or invertible),
provided every element of the range of h corresponds to a unique member of the
domain X of h. In other words, if the mapping h is 1-1, then every image in the range
of h has a unique pre-image in the domain of h. The inverse of the mapping h is
denoted by h −1 . Whenever the mapping is invertible, then h −1 (x) = x. A mapping
that is both 1-1 and onto is called a bijection. For a good introduction to mappings,
see the Gellert Encyclopedia [18, Sect. 14.5, starting on p. 325].
Example 3.22 (Types of 1-1, onto mappings) A picture of an invertible h : X −→ Y
mapping that resembles a hair comb with a handle on each side of the comb where
3.13 Homeomorphic Mappings and Trails Between Nerves 141
(a) 1-1, onto (b) 1-1, not (c) not (d) not
onto 1-1, onto 1-1, not
onto
The mapping h is 1-1, since every point in X is mapped to a single point in > p1 p4 .
The map h is a mapping on X onto > p1 p4 , since all points in the range > p1 p4 are
included in the map. Let A δ B for A, B ∈ X read A is near B. Observe that h is
continuous, since x δ x in X maps to h(x) δ h(x ) for h(x), h(x ) in the range > p1 p4 .
Finally, h −1 is also continuous, since h −1 (x) δ h −1 (x ) maps to x δ x . Hence, trail
X is homeomorphic to the wiggly segment > p1 p4 . “
Lemma 3.25 Every pair of vertices in a triangulated planar region has trail between
them.
Lemma 3.26 There is a trail between every pair of nerves in a triangulated, finite,
bounded, planar region.
Proof Let NrvA( p), NrvB(q) be a pair of nerves with nuclei p, q, respectively,
in a triangulated planar region. From Lemma 3.25, there is a trail between the
nuclei p, q.
Example 3.27 (Trail between nerves) A sample trail > pq between nuclei p, q in a
pair of nerves NrvA( p), NrvB(q) is shown in Fig. 3.18. “
For example, observe the many possible geodesic trails between shapes in a Napoli
breakfast3 in Fig. 3.19.
Example 3.28 (Geodesic trails between shape vertices) A triangulation of the Napoli
breakfast in Fig. 3.19 is shown in Fig. 3.20a. A sample geodesic trail > pq between
vertices p, q in a pair of shapes shA( p), shB(q) is the sequence of red — segments
shown in Fig. 3.19. A particular vertex p on, for example, shape shA in Fig. 3.20b
is denoted by shA( p). Similarly, a vertex q on shape shB in Fig. 3.20c is denoted
by shB(q). The highlighted yellow region labelled shB in Fig. 3.20c indicates a
nerve with the maximum number of triangles that have the nucleus q in common.
Depending on the selection of the vertices on a pair of shapes, the geodesic trails
will vary in length. Here are sample geodesic trails between shA( p) and shB(q).
>
pq = geodesic trail between nuclei p and q,
>
pp = geodesic trail between shB nucleus p and shB vertex p ,
>
p q = geodesic trail between shB vertex p and shB nucleus q,
>
p p = geodesic trail between shB vertex p and shB vertex p .
3 Many thanks to R. Tozzi and A. Tozzi for the Napoli breakfast picture.
3.14 Geodesic Trails Between Shapes 145
The choice of vertices on the ends of a geodesic line will depend on the part of each
shape of greatest interest. “
Theorem 3.29 There is a geodesic trail between every pair of shapes on a triangu-
lated, finite, bounded, planar region.
Proof Let cxK be a triangulation on a set vertices K of a bounded planar region. By
definition, every planar shape shA has a boundary bdy(shA) that is a simple closed
curve. Select shapes shA, shB ∈ cxK . Select a vertex p on the boundary bdy(shA)
146 3 Shape Fingerprints, Geodesic Trails and Free Abelian Groups …
and a vertex q on the boundary of bdy(shB). From Lemma 3.26, there is a trail >
pq
between p and q. By choosing the shortest segment between each pair of possible
vertices on the trail in >
pq, we obtain a geodesic trail between p and q.
Problem 3.30 ®
Prove that there is a shortest segment for each of the segments in a trail between
triangulated planar shapes. “
Problem 3.31 K
Give an algorithm to find and highlight a geodesic trail between a pair of shapes in
a triangulated bounded region in each frame of a video. Use Matlab® to implement
your algorithm on the frames of a video. “
Video frame object shapes change ever so slightly, provided video camera move-
ment is minimal and the components of a visual scene change very little over short
time intervals. Of considerable interest is a trajectory that maps continuous shape
change in the form of a geodesic trail between selected vertices in a triangulated
frame. The collection of recorded changes in a shape during a movie defines a
shape space. A shape space is a record of shape change vertices in a sequence
of video frames. A shape change geodesic trail is a sequence of segments between
changing vertices recorded in shape space. For more about this, see Faraway and
Trotman [23].
Problem 3.32 K
Give an algorithm to compute the length of a shape change geodesic trail between
vertices on a triangulated bounded video frame region over a sequence of video
frames. Shape change is represented by a record of changes in shape boundary
vertices with a corresponding change in the lengths and gradients of the segments
between selected pairs of shape boundary vertexes. Use Matlab® to implement your
algorithm on a selection of 10 different video frames. “
Example 3.33 (Sample Simple Closed Curve) A sample simple closed curve A is
shown in Fig. 3.21. An interior point int p is represented by the bulls-eye and an
exterior point extq is represented by the bulls-eye in Fig. 3.21. “
3.15 Elementary Shapes 147
In effect, a simple closed curve provides the contour of an elementary planar shape
that encloses a distinct interior region as well as a region of the plane external to an
elementary shape. From the Jordan Curve Theorem 1.1, we get the following result
for all planar elementary shapes.
Theorem 3.34 (Di Concilio–Guadagni–Peters Theorem) Every elementary planar
shape divides the plane into two parts.
Proof By definition, the boundary of an elementary planar shape is a simple closed
curve. The shape boundary determines a separation between the interior planar region
and exterior planar region of a shape. Hence, from the Jordan Curve Theorem 1.1,
an elementary shape divides the plane into two parts.
The Jordan Curve Theorem extends to surfaces in 3-dimensional space in which
every simple closed surface (a surface that does not fold into itself) divides space
into two regions the same as a planar simple closed curve does [6, p. 3].
4 Many thanks to Alexander Yurkin for supplying this self-portrait of his father.
3.15 Elementary Shapes 149
Example 3.35 (Sample Digital Image Holes) An Italian Emme Poste vehicle parked
outside the Salerno train station is shown in Fig. 3.23. Sample holes are shown in
Fig. 3.23b. The locations of the centroids of the holes are shown in Fig. 3.23c. In
each case, the centroid of a hole is represented by a red dot •. “
Example 3.36 (Sample Low and High Intensity Image Holes) The cabin showing
the driver of an Italian Emme Poste vehicle contains a number holes that contain
uniformly low intensity pixels as shown in Fig. 3.24a. Sample bounded regions con-
taining uniformly high intensity pixels are shown in Fig. 3.24b. “
Theorem 3.37 Every boundary of a planar physical shape can be decomposed into
a cyclic chain-wise set of pixels.
Proof The proof is by construction. Select any set M containing pixels and an arc
A connected between p, q ∈ M. Assume M contains at least three pixels. Then
connect p to an arc segment on M, connected between p and r ∈ M ∖ q or q ∈ M.
Repeat this, until the last arc selected has q as an endpoint.
150 3 Shape Fingerprints, Geodesic Trails and Free Abelian Groups …
(a) Emme Poste dark holes (b) Emme Poste illuminated Holes
Fig. 3.24 Sample dark and illuminated (white light) image holes
In this section, shapes found in triangulated planar regions can be gathered up into
collections those shapes that have affinities with each other. This is done by imposing
a proximity relation on a triangulated region.
Let X be a nonempty set and let 2 X denote the collection of all subsets in X with
A, B ∈ 2 X .
Recall that a proximity relation δ on a collection of sets 2 X is a set of ordered pairs
of subsets in the collection 2 X . If (A, B) ∈ δ, one also observes that δ holds for the
ordered pair (A, B) and one writes this in the form A δ B. We can construct a Leader
uniform topology on X (see Leader [29]). To this, select all sets B ∈ 2 X that are
near each given set A ∈ 2 X . The end result is a collection of sub-collections of 2 X
so that each subcollection contains members with the property that A δ B. If A, B
have matching descriptions, then we write A δΦ B to indicate that A is descriptively
near B. Let X be a planar region and let cxK be a simplicial complex derived from
the triangulation of X on a set of vertices K in X . The subscript Φ comes from a
description mapping Φ : 2 X −→ Rn defined by
topology, select filled triangle, e.g., A1 in Fig. 3.24a. Notice that Sample bounded
regions containing uniformly A1 has a very dark region. Then compare the descrip-
tion of A1 with the other curved triangles in Fig. 3.24a and look for matching
descriptions. For example,
A1 δΦ A2 (A1 is descriptively near A2 ),
A1 δΦ A3 (A1 is descriptively near A3 ),
A1 δΦ A4 (A1 is descriptively near A4 ).
This section briefly considers algebraic structures called cyclic groups that are deriv-
able from the contours and skeletons of physical shapes. This is done by decomposing
shape contours and skeletons into collections of unit length arcs.
An arc on a planar image shape boundary is analogous to what B. L. McAllister
called a lump in a Peano continuum (any measurable closed interval on a continuous
curve) [28, p. 337]. In this section, we consider decomposing simple closed curves
into arcs of equal length. This paves the way for the introduction of algebraic struc-
tures called cyclic groups derived from shape boundaries. To do this, we first tackle
group structures such as ordinary, garden variety groups with no special properties
as well as Abelian groups. First, we consider ordinary groups, which are prevalent
structures associated with various shapes, including physical shapes.
Definition 3.39 (Group) A group is a pair (G, ◦), where G is a nonempty set with
a binary operation ◦ defined on G. Let a, b, c ∈ G. G is a group, provided
Closure : a ◦ b ∈ G for all a, b ∈ G.
Identity Element : There is an identity element e ∈ G so that a ◦ e = a for all
a ∈ G.
Associativity : (a ◦ b) ◦ c = a ◦ (b ◦ c) for all a, b ∈ G.
Inverse Element : There is an inverse element −a ∈ G so that a ◦ −ae for all
a ∈ G. Notice that an inverse −a is the negative of element a ∈ G and can have
many different forms, depending on the choice of the binary operation ◦ and the
nature of the elements in G. “
A group binary operation is also called the product and denoted by either ◦
or by ·. Recall that a mapping from a set X to a set Y is a subset of M of X × Y
such that, for every x ∈ X , there is a unique element y ∈ Y such that the ordered
pair (x, y) is in M [13, Sect. 2, p. 10]. We write π : X −→ Y to indicate that π is a
mapping on X into Y , i.e., π maps X into Y .
A binary operation on G is a mapping ◦ : G × G −→ G that maps the product
G × G into G (i.e., ◦ is closed). In general, let X be a nonempty set. A mapping
◦ of X × X into X is a binary operation on X . The identify element is a special
element (typically denoted by e) in group (G, ◦), which leaves any other element in
G unchanged after the a · e for all a ∈ G. A good introduction to group theory is
given by Herstein [13].
Example 3.40 Let G be the set of integers 0, ±1, ±2, ±3, . . . and let ◦ = + (addition
operation). “
Problem 3.41 ®
Verify that the group properties are satisfied for the additive (G, +) in Example 3.40.
154 3 Shape Fingerprints, Geodesic Trails and Free Abelian Groups …
A ∩ B is the set of all elements in A that are also in B. For a good introduction to
set theory, see Moschovakis [30]. Notice that the empty set ∅ in included in 2 A . Let
Ac (complement of A ⊆ G) denote the set of elements not in A and G c is the set of
all elements not in G. In terms of the intersection ∩, Ac is the inverse of A and ∅ is
the identity element. That is, A ∩ Ac = ∅. “
Problem 3.43 ®
Verify that the group properties are satisfied for (G, ∩) in Example 3.42.
Example 3.44 Let X be a set of shapes in the rabbit sketch in Fig. 3.26, i.e.,
8
X = X1 · · · ∪ Xi ∪ · · · ∪ X8 = X i (Union of subsets in rabbit shape), where,
i=1
X 1 = rabbit shape contour (boundary),
X 2 = eye,
X 3 = inner ears,
X 4 = inner tail,
X 5 = inner rear paw,
3.17 Cyclic Groups Derived from Shape Contours and Skeletons 155
Problem 3.45 ®
Verify that the group properties are satisfied for (X, ∩) in Example 3.44.
Example 3.46 Let G be the set of all pixel intensities in a digital image I mg. Assume
that all pixel intensities in I mg range from 0 to 255. The residue or remainder of
x mod m means the remainder after the division of x by m. The number m is called
the modulus. This is sometimes called clock arithmetic. Let x mod 255 = mod [x,
255] denote the remainder after division of x by 255. For example, 55 mod 255 =
mod [55, 255] = 55 (remainder after dividing 55 by 255). “
Problem 3.47 ®
Verify that the group properties are satisfied for (G, mod255) in Example 3.46.
Problem 3.49 ®
Verify that the group properties are satisfied for (G, ∩) in Example 3.48. Hint: The
intersection of any pair of subimages in an image I mg is one of the subsets in 2 I mg .
Definition 3.50 (Abelian Group) Let G be a group with a binary operation ◦ defined
on G. Let a, b ∈ G. G is an Abelian group, provided
Commutativity : a ◦ b = b ◦ a for all a, b ∈ G. “
Problem 3.51 ®
Prove that the group (G, ∩) in Example 3.44 for the subsets of the rabbit shape
Problem 3.52 ®
Prove that the group (G, mod255) in Example 3.46 (clock arithmetic on pixel inten-
sities) is Abelian. Hint: Check the properties of union mod255.
Problem 3.53 ®
Prove that the group (G, ∩) in Example 3.48 (∩ on 2 I mg ) is Abelian. Hint: Check
the properties of intersection ∩ on G.
Consider decomposing a circle contour (boundary) into arcs so that each arc is a
>
multiple of the smallest arc. For example, the length of arc AC in Fig. 3.27 is double
>
the length of arc AB. Similarly,
> >
AD = 3 AB.
>
The arc AB is an example of what Giblin [10, Sect. A.6, p. 216] terms a generator
in a cyclic group. Let (G, +) be a group and let q ∈ G. The element q is a generator
of G, provided every x ∈ G can be written as a multiple of q, i.e., x = nq for some
integer n.
Example 3.54 Let G equal the set of integers under +. For x ∈ G a positive integer,
x = n · 1. If x ∈ is a negative integer, then x = −n · 1 and if x = 0, then x = 0 · 1.
Hence, 1 is generator for the integers. “
Definition 3.55 (Cyclic Group [10]) A group (G, +) is cyclic, provided there exists
an element a ∈ G such that any b ∈ G is of the form na for some integer n ∈ Z+ ∪ 0
(n is either a positive integer or zero). The element a is called a generator of G and
−a is the inverse of the generator a. “
Example 3.56 Let G equal the set of integers 0, ±1, ±2, ±3, . . . and let ◦ = +
(addition operation) with generator +1. “
3.17 Cyclic Groups Derived from Shape Contours and Skeletons 157
Problem 3.57 ®
Verify that (G, +) in Example 3.56 is a cyclic group. Hint: Use the fact that (G, +)
from Problem 3.41 is a group.
Example 3.58 Let G ar c equal the set of arcs on the circumference of the circle
>
in Fig. 3.27 so that each arc in G is a multiple of arc AB (generator) and let + be
defined on the set of arcs in G. “
Problem 3.59 ®
Verify that (G ar c , +) in Example 3.58 is a cyclic group. Hint: First prove that
(G ar c , +) is a group. Then observe that each arc b ∈ G ar c is a multiple of AB.
The set G ar c is very restricted. Notice that there are many other arcs on the circle
boundary in Fig. 3.27 that are not included in the set of arcs in G ar c .
Example 3.60 A sketch of a rabbit physical shape is given in Fig. 3.28a (call it shape
shA). The contour of shA is given in Fig. 3.28b (call it shape bdy(shA)). The skeleton
of shA is shown in the interior of the barebones rabbit shape in Fig. 3.28c. The rabbit
physical skeleton (by itself) is given in Fig. 3.28d. “
In decomposing a physical contour into arcs, the goal is to decompose the contour
bdy(shA) of a shape A into a collection G of connected arcs with the same length
so that G covers bdy(shA). To do this, it is necessary to select an arc generator with
some unit length such as 1 mm. Then every arc in G will be n×1 mm. This is possible,
provided the unit arc length is small enough. Since we only require bdy(shA) ⊆ G,
we allow for a small overlap in cases where a contour cannot be evenly divided into
arcs with same length.
(a) physical shape (b) shape contour (c) skeletonized (d) barebones
shape skeleton
Theorem 3.62 (Physical Contour Cyclic Group) Every planar physical contour has
an arc generator for a cyclic group.
Example 3.64 The skeleton of a physical shape shA is shown in the interior of the
barebones rabbit shape in Fig. 3.28d (call it shA). The rabbit physical skeleton
shA (by itself) is shown partially decomposed in arcs of unit length in Fig. 3.28d.
A partial decomposition of the physical contour is shown in Fig. 3.30. The assumption
> >
made here is that arc AE = 5 AB. “
3.17 Cyclic Groups Derived from Shape Contours and Skeletons 159
Theorem 3.65 (Physical Skeleton Cyclic Group) Every planar physical skeleton
has an arc generator for a cyclic group.
Proof Let shA be the skeleton of a physical shape A. Decompose shA into a
collection G of connected arcs a of the same length so that G covers shA and
every length of every arc G is a multiple of a. That is, if b is an arc in skeleton G,
then b = na, n ∈ Z+ ∪ {0}. Hence, by definition, (G, +) is a cyclic group.
A finitely generated (f.g.) cyclic group (G, +) is a cyclic group with generators
q1 , . . . , qn , n ≥ 1, provided, for x ∈ G,
x = λ1 q1 + · · · + λi qi + · · · + λk qk , k ≤ n, q1 , . . . , qk ∈ G, λi ∈ Z(integers).
x = λ1 q1 + · · · + λk qk = 0, q1 , . . . , qk ∈ G implies λ1 = · · · = λk = 0.
The operation + is Abelian. That is, for any pair elements x, y ∈ G, we have
means that every member g in G with k > 0 generators x1 , x2 , , . . . , xi . . . xk
can be written as a linear combination of the generators in the following way.
A free Abelian group (G, +) = ({x} , +) with one generator x is just a cyclic
group.
Since every member of G is a linear combination of the generators q1 , . . . , qn
with multipliers λ1 + · · · + λk , the set of generators is called a basis for G. That is,
a linear combination is a sum of the basis elements where each term of the sum is
a multiple of a basis element. Let G 2 be a free group with 2 generators q1 , q2 .
For example, let q1 be a collection of 89 connected line segments and let q1 be a
collection of 233 connected line segments. So, for example, a group member g ∈ G 2
is defined by
g = λ1 q1 + λ2 q2 = 89 · q1 + 21 · q2
= connected segments from q1 , q2 .
For more about free groups, see Giblin [10, Sect. A.10, p. 218].
Example 3.66 A pair of generators for a free cyclic group G 2 is represented by the
connected vortexes shown in Fig. 3.31. We write G 2 , since the rank of this free group
is 2, i.e., G has 2 generators represented by q1 , q2 in Fig. 3.31. “
q t
p
r
162 3 Shape Fingerprints, Geodesic Trails and Free Abelian Groups …
pq + >
∂ = > > (Boundary of a 2-cell).
pr + qr
After triangulation of a finite, bounded planar region containing holes, the bound-
ary of each hole is the sum of its oriented bounding edges. By adding λ summands
that are oriented edges of a hole (denoted by λ∂), we obtain a representation of the
boundary of a hole, i.e.,
λ terms
λ∂ = ∂ + · · · + ∂ (Oriented Boundary of a Hole).
Most images contain more than one hole. The boundary of each hole contains paths
defined by a sequence of connected 0-cells such as a sequence of edges on pairs of
neighbouring barycenters on triangles along the border of a each hole. Each boundary
containing connected, oriented barycenter edges defines a chain. Let par tiali be
the ith path (connected edges) along a border. A boundary-chain c is a sum of paths
is defined by
c= ∂i .
Each coefficient λi is an integer that indicates the number of oriented cells (e.g.,
directed edges) on the boundary i of a hole. In defining an additive group based on
n-chains, addition is performed on the λi coefficients.
To provide a basis for an additive group, all coefficients λi are mapped to λi modk,
where k is the number of terms in an n-chain Cn . “
Addition mod k is represented by + [k] (shorthand for +mod k). For example,
Problem 3.69 ®
Let C5 be a 5-boundary chain with
Give a table representing the (C5 , +[5]) chain group for addition modulo 5.
“
Problem 3.70 K
Write a Mathematica notebook to do the following:
1o Select an image from a collection 3 digital images.
2o Using the centroids of image holes as sites, triangulate the selected image. Dis-
play the triangulated image.
3o Using edges as cells bordering each hole, highlight a sequence of connected cells
on each hole boundary. Display the highlighted cells.
4o Compute the coefficients for an n-chain:
n hole boundaries
Cn = λ1 ∂1 + · · · + λi ∂i + λi+1 ∂i+1 + · · · + λn ∂n (n-chain).
That is, Compute λi mod n for each λi coefficient in Cn . Display Cn with the
computed coefficients.
5o Compute the sums
i, j = bdy sizes
λi + λ j mod 2
Since directed edges form the hole boundaries in a chain, then, by reversing the
direction, we obtain the inverse of a chain Ci , i.e.,
Example 3.71 The IR image shown in Fig. 3.33a has two holes indicated by • discs
covering the holes. “
Two types of chains are introduced in this section, namely, a n-chain called a cycle
(i.e., a chain with n cycles, denoted by Z n ) and a n-boundary that is the boundary of
a n + 1-chain, leading to a group of n-boundaries denoted by Bn = Bn (K ).
Recall that an n-connected cyclic chain complex Zn is a sum of connected cyclic
n oriented cells in a cell complex K . A chain Z n is called an n-cycle chain, provided
its boundary ∂Cn equals zero, i.e., ∂Cn = 0. In other words, an n-cycle is a n-chain
with an empty boundary [2, Sect. IV.1, p. 80].
An n-boundary is an n-chain Bn that is the boundary of an n + 1-chain so that
λi ∂i ∈ Ck+1 ,
n-boundary = n-chain
Bn = λ1 ∂1 + · · · + λi ∂i + λi+1 ∂i+1 + · · · + λn ∂n .
The two types of chains are used to define homology groups. Each boundary
λi i of a hole is a cycle, which is a generator of a cyclic group. A collection of
connected boundaries in a cell complex K defines a free cyclic group called the
boundary group (denoted by (Bn , +) or simply by Bn (K ) or Bn ) for a collection
of n boundaries on a simplicial complex K . The boundaries λi ∂i are generators of
the boundary group Bn (K ). Notice that λi ∂i + λi+1 ∂i+1 commutes with addition,
i.e.,
commutes
λi ∂i + λi+1 ∂i+1 = λi+1 ∂i+1 + λi ∂i (Abelian property).
Example 3.72 Three 2-dimensional chains on an IR image are shown in Fig. 3.33
containing a pair of holes. Each hole is bounded by circular green region (see, e.g.,
Fig. 3.35) with the centroid of the hole indicated by a red •. In Fig. 3.33, ∂C2 = 0,
i.e., C2 has an empty boundary. The 2-chain Z 2 lies between B2 and C2 . B2 is a
2-chain boundary, since B2 contains only 2 hole boundaries (Fig. 3.34). “
Every oriented filament skeleton is the boundary of a shape. Recall that an oriented
filament skeleton skA is a filament skeleton with vertices that have a particular
ordering. The ordering of the vertices in a filament skeleton represents motion along
the filaments, from some starting point to a vertex a in skA. And reverse motion
from vertex a back to where we started the motion is also possible.
Example 3.73 Recall from Sect. 2.10 that a skeletal vortex on a triangulated surface
is a collection of filament skeletons with either a common vertex or a common
edge. Each of the oriented filament skeletons skA1 , skB1 in the skeletal vortexes
skVA, skVB in Fig. 3.36 is a boundary of a shape with nonempty interior. For
these two filament skeletons, the unlabeled vertices are represented by • dots and the
unspecified nonempty interior is represented by grey .
And each filament skeleton has a single generator. In the case of a skeletal vortex,
the filament skeletons in the vortex have a common vertex. The filament skeletons
skA1 , skB1 , each with its own single generator, are represented by cyclic groups
G sk A1 (◦), G skB1 (◦ ), where ◦, ◦ are binary operations on the groups members. In
each case, the group members are represented by the unlabeled • dots in Fig. 3.36. A
3.21 Filament Skeleton Cyclic Group 167
cyclic group results from the forward rotation (in a clockwise direction) starting, for
example, with vertex a in G sk A1 (◦), moving round the skeleton and arriving at a. “
a ◦ −a = 0 clockwise rotation to a
◦ (added to)
counterclockwise rotation from a to obtain −a
sums to 0 no motion.
In other words, the group representing each of the filament skeletons has an
identity element 0 and, for each rotation to a vertex x, there is reverse rotation
from x back to where we started the rotation, giving us the inverse element −x. To
complete the picture of a filament skeleton cyclic group, we need to verify that the
group operation ◦ is associative.
Proposition 3.74 The filament skeleton cyclic group operation ◦ is associative.
(x ◦ y) ◦ z = y ◦ z
=z
= x ◦ (y ◦ z)
=x◦z
=z
Recall from Sect. 2.6 that a skeletal nerve skNrvA is a collection of skeletal vortexes
skVA with nonempty intersection in a CW complex. That is,
skNrvA = skV A : skV A = ∅ (skeletal nerve).
We are interested in skeletal nerves that are collections of oriented filament skeletons
in a skeletal vortex. Recall from Sect. 2.10, a skeletal vortex is a collection of oriented
filament skeletons with a common vertex or edge. By definition, a skeletal vortex
is a skeletal nerve. From Sect. 3.21, we know that a bi-directional oriented filament
skeleton makes it possible to derive a cyclic group. Hence, a skeletal vortex is a nerve
represented by a collection of cyclic groups. Having arrived at this point, we can look
forward to deriving a free Abelian group representation of each skeletal vortex and
each skeletal nerve.
Recall from Sect. 3.18 that a free Abelian group is the direct sum of a collection
of cyclic groups. Since every skeletal nerve is a collection of skeletal vortexes with
a common vertex, we can also derive a free Abelian group representation on every
skeletal nerve.
Lemma 3.75 Each skeletal vortex has its own Betti number.
Theorem 3.76 The Betti number of a skeletal nerve equals the sum of the Betti
numbers of its skeletal vortexes.
Bsk Nr v = B1 + · · · + Bi + · · · + Bn ,
Example 3.77 (Skeletal Vortex Nerve with Betti Number 3) A skeletal vortex nerve
skNrvE is represented by a pair of nesting, non-concentric vortexes skVA, skVB
with a cusp filament filamentE attached between 3a = 0e on skVA and 4b = 2e on
skVB in Fig. 3.37. skNrvE is represented by the group
with three generators, namely, a , e . a. In this vortex nerve group, the Betti
number for skVA = 1, since skVA has one generator, namely, a. Similarly,
skVE = skVB = 1, since each of these vortexes has a single generator, namely,
e for filamentE and b for skVB. Hence, from Theorem 3.76, skNrvE has a Betti
number equal to 3. “
maps to
G sk A1 (◦) −→ G sk A1 (+2 ).
maps to
G skB1 (◦) −→ G skB1 (+2 ).
Recall that a free Abelian group G(+2 ) is the direct sum of its cyclic groups. In effect,
skVA is a free Abelian group G(+2 ) that is the direct sum of its cyclic groups. For
simplicity, we define the binary operation in G as addition modulo 2 (denoted by
+2 ) on the coefficients of members of G. For example, let G be on a pair skeletal
skVA1 , skVB1 in Fig. 3.36. with generators {a , b} This means that for an element
x ∈ G, x can be written as
maps to
x = na + mb −→ n +2 m, a sum, which is an integer 0 or 1. “
The next thing to consider is a practical application of what we have learned about
free Abelian groups on either skeletal vortexes or on skeletal nerves.
Recall from Sect. 2.10 that every skeletal vortex nerve (collection of intersecting
oriented filament skeletons) embodies one or more cycles. Every oriented filament
skeleton can be represented by an additive cycle group. In its simplest form, addition
modulo 2 is on the coefficients of the sum of a pair of group elements. From the
combination of cycles derived from oriented shape boundaries and holes in shape
interiors, free abelian groups can be derived. Such groups provide a measure of
surface shape changes over time. Typically, surface shapes are represented by vortex
complexes. The Betti number of each free abelian group tells us the number of group
generators associated with vortex complexes that may or many not persist over time,
depending on the evolution of the corresponding surface shapes.
This is an intuitive view of Betti numbers that is based on the analogy between
vortex nerves and reflected light from curved surfaces, which is analogous to what
is known as the coffee cup light caustic introduced by Nye [33]. Briefly, an optical
caustic is an envelope of light rays either reflected or refracted by a curved surface
3.23 Betti-Nye Optical Vortex Nerves and Persistent Betti Numbers 171
and the projection of that envelope of rays on another surface (observed by Lynch and
Livingston [34]). A coffee cup caustic results from an envelope of rays of sun light
reflected from the inner curved surface of a filled cup of coffee and the projection of
the envelope onto the coffee surface [33, Sect. 2.1, pp. 9–12] (see, e.g., Fig. 3.39). In
our case, the cusp of the projected envelope from the light caustic is represented by
a cusp filament connected between vortexes. For this reason, a pair of nesting, non-
concentric vortexes with cusp filaments attached between them is called a Betti-Nye
optical vortex nerve (Fig. 3.38).
Taken together, the path-connectedness of the vertexes attached to each other
by edges have the appearance of the cusps in a coffee cup caustic. These path-
172 3 Shape Fingerprints, Geodesic Trails and Free Abelian Groups …
Example 3.78 A sample optical vortex nerve5 on the triangles of a maximal nerve
complex (MNC) of a triangulated video frame, is shown in Fig. 3.40. Briefly, notice
that there is a pair of nesting, non-concentric vortexes with 9 cusp filaments connected
between inner and outer vortex vertexes, each with its own generator. Hence, from
Theorem 3.76, the Betti number equals 2 + 9 = 11 for this sample nerve. “
5 Many thanks to Arjuna P. H. Don for this sample triangulated video frame.
3.24 Optical Vortex Nerve Viewed as Intersecting Equipotential Lines 173
Fig. 3.40 Optical vortex nerve on a MNC barycenters of a triangulated video frame
Problem 3.79 ®
Give a graphical representation of the actions performed by a vfVOT. “
Problem 3.80 ®
Give a graphical representation of the vortex nerve constructed by a vfVOT. “
Problem 3.81 ®
Give a cyclic group G(< a >, +6 ) representation (in table form) of an oriented
skeletal filament skA surrounding the MNC in a triangulated video frame. Assume
that the generator of G is a and that skA has 6 vertices 0a, 1a, 2a, 3a, 4a, 5a
and the +6 represents modulo 6 (addition on a 6-hour clock). See, for example, a
representation of G(< a >, +6 ) in Fig. 3.41. “
3a
3.24 Optical Vortex Nerve Viewed as Intersecting Equipotential Lines 175
For more about optical tweezers in electrodynamics, see Zangwill [36]. A vfOT is
modelled as an optical vortex nerve skNrvK defined by the intersection of captured
points of light in filament skeletons surrounding a maximal nucleus cluster (MNC)
on one triangulated video frame (read skeletal vortex skVB) and in filament skeletons
surrounding a maximal nucleus cluster (MNC) gather the captured points of light in
a second video frame (read skeletal vortex skVB), i.e.,
See, for example, a representation of an optical skeletal nerve, see Fig. 3.42. In
practice, the operation of an optical vortex nerve shape-capture derived from pairs
of video frames would resemble a Curtis-Greer optical vortex [37].
To see what I mean, consider the equipotential lines of an electric field in a
typical environment filled with light-reflecting surfaces. These lines are represented
as non-concentric, nesting circles (read cycles that define cyclic groups, combined to
form a transient free Abelian group) in Baldomir and Hammond [38, Sect. 5.1, pp. 96–
97]. The Betti number of this electric field free Abelian group would be continuously
changing in spacetime. Typically, equipotential lines are viewed as smooth curves in
an electric field, which are continuously changing. The smooth form of equipotential
lines can be projected into the plane with a function f : Z −→ R2 defined by
z−1
f (z) = e z − .
z+1
Problem 3.83 ®
What is the Betti number for the free Abelian group that represents skNrvK in
Fig. 3.42. Justify your answer. “
Problem 3.84 K
Design video frame Optical Tweezers (vfOT) which constructs a vortex nerve
skNrvK from a pair of video frames. Let frame A, frameB be a pair of video frames.
This vortex nerve would be constructed from the intersection of captured points of
light (filament skeletal vertexes) in the filament skeletons in a largescale skeletal
vortex skVA surrounding a maximal nucleus cluster (MNC) in video frame frame A
and the captured points of light in the filament skeletons in a skeletal vortex skVB
surrounding a maximal nucleus cluster (MNC) in video frame frameB. The term
largescale skeletal vortex means that there are large number of vertices in each of
the filament skeletons in such a skeletal vortex. The basic idea is to derive a new shape
from the shapes covered by skVA in frame A and skVB in frameB. Implement your
vfOT in Matlab. Your vfOT should do the following:
1o Construct a pair of largescale skeletal vortexes skVA, skVB on a selected MNC
on a pair of adjacent video frames frame A, frameB.
2o vfOT: Construct a skeletal nerve skNrvK , which equals the collection
{skVA, skVB}, which have nonempty intersection.
3o Give the Betti number of the skeletal nerve skNrvK .
4o Display the skNrvK shape by highlighting those points of light in skVA, skVB.
Hint: skNrvK should resemble the optical vortex nerve in Fig. 3.42. “
Each equipotential line can be view discretely as a filament skeleton with many
path-connected vertices. From a filament skeleton perspective, a collection of non-
concentric nesting filament skeletons (discrete form of equipotential lines) is a vortex
nerve. See also, for example, Meijer [39] for the toroidal geometry of what happens
in the larger physical world.
The intuitive view of Betti-Nye vortexes carries over to skeletal nerves that can
be found in spacetime structures in most visual scenes.
Problem 3.86 ®
Let skNrvX be defined by intersecting skeletal vortexes skVA, skVB in Fig. 3.36,
where
What is the Betti number for the free Abelian group that represents skNrvX . Justify
your answer. “
For simplicity, we consider only vortex nerves with filament skeletons that are fine
and coarse contours of maximal nucleus clusters (MNCs) on either a tessellated or
triangulated video frame. Fine and coarse contours of a MNC on a tessellated surface
were introduced in Peters [12, Sect. 8.9, p. 262]. We use the result from Lemma 2.18
in the study of vortex nerves, i.e., a skeletal vortex is a skeletal nerve.
178 3 Shape Fingerprints, Geodesic Trails and Free Abelian Groups …
Example 3.87 (Betti number of a free Abelian group for a skeletal vortex nerve) A
sample skeletal vortex skVA on a maximal nucleus cluster (MNC) of a tessellated
video frame7 is shown Fig. 3.43. Notice that skVA is a collection of non-concentric,
nesting filament skeletons with a common vertex in the upper right portion of the
vortex. From Lemma 2.18, skVA is a vortex nerve. Each filament skeleton is repre-
sentable as cyclic group with a single generator. For this reason, skVA is representable
as a free Abelian group with Betti number equal to 2. “
7 Many thanks to Enze Cui for the video frame in this example.
3.24 Optical Vortex Nerve Viewed as Intersecting Equipotential Lines 179
Example 3.89 (Betti number of a free Abelian group for colliding skeletal nerves)
The tessellated video frame8 in Fig. 3.44 shows colliding filament skeletons sk A, skB
on a pair of close MNCs. From Lemma 3.88, skA, skB are examples of skeletal
vortexes with a common vertex. As in Fig. 3.36, these colliding filaments define a
skeletal nerve skNrvE.
From what we observed earlier in Sect. 3.22, the skeletal vortex nerve skNrvE
can be represented as a free Abelian group G skNrvE (+2 ) defined in terms of addition
modulo 2 on the coefficients on multiples of the cyclic group generators for sk A, skB.
The Betti number for skNrvE equals 2, since G skNrvE (+2 ) consists of two filament
skeletons, each with a single generator. “
Betti numbers:
Tucker and Bailey [40, p. 20], on the history of Betti numbers.
Edelsbrunner and Harer [2, Sect. IV.1, p. 81], Betti number as the rank of the pth
homology group.
Cooke and Finney [41]
Cell complexes:
Cooke and Finney [41, Chap. I, starting on p. 1], a brief but very good introduction
to cell complexes. See, especially, [41, Sect. 1.2, p. 1ff and Sect. 1.2, p. 2ff] on
Classes of Shapes.
K Each known triangulated shape belongs to a class of shapes that is used
to classify unknown triangulated shapes. Unlike the conventional Delaunay
triangulation of spatial regions, the proposed triangulation results in cells that
are filled triangles, derived from the intersection of half spaces, where the edge
of each half space contains a line segment connected between vertices called
sites (generating points). A straightforward result of this approach to image
geometry is a rich source of simple descriptions of plane shapes of image objects
based on the detection of nerve complexes that are maximal or MNCs. Shape
classes spring from collections of triangulated shapes that have affinities
with each other as a result of their similar descriptions. A shape class is a
lot like a collection of paintings by the same artist. The end result of this work
is a proximal physical geometric approach to detecting members of classes
shapes. “
References
1. Zomorodian, A.: Topology for Computing. Cambridge University Press, Cambridge (2005),
xiii+243 pp. ISBN:9780-0-521-13609-9
2. Edelsbrunner, H., Harer, J.: Computational Topology. An Introduction. American Mathematical
Society, Providence (2010), xii+241 pp. ISBN:978-0-8218-4925-5, MR2572029
3. Zomorodian, A.: Topological data analysis. In: Advances in Applied and Computational Topol-
ogy. Proceedings of Symposia in Applied Mathematics, vol. 70, pp. 1–39. American Mathe-
matical Society, Providence (2012). MR2963600
4. Rote, G., Vegter, G.: Computational topology: an introduction. In: Boissonnat, J.D., Teillaud,
M. (eds.) Effective Computational Geometry for Curves and Surfaces, pp. 277–312. Springer,
Berlin (2007), Xii+343 pp. ISBN: 978-3-540-33258-9, MR2387755
5. Zomorodian, A.: Introduction to Computational Topology. Stanford University (2017). http://
graphics.stanford.edu/courses/cs468-04-winter/
6. Alexandroff, P.: Elementary Concepts of Topology. Dover Publications, Inc., New York (1965),
63 pp., translation of Einfachste Grundbegriffe der Topologie [Springer, Berlin, 1932], trans-
lated by Alan E. Farley, Preface by D. Hilbert, MR0149463
7. Jänich, K.: Topology. With a chapter by T. Bröcker. Translated from the German by Silvio
Levy. Springer, New York (1984), ix+192 pp. ISBN: 0-387-90892-7 54-01, MR0734483
8. Hatcher, A.: Algebraic Topology. Cambridge University Press, Cambridge (2002), xii+544 pp.
ISBN: 0-521-79160-X, MR1867354
182 3 Shape Fingerprints, Geodesic Trails and Free Abelian Groups …
9. Ghrist, R.: Elementary Applied Topology. University of Pennsylvania (2014), Vi+269 pp.
ISBN: 978-1-5028-8085-7
10. Giblin, P.: Graphs, Surfaces and Homology, 3rd edn. Cambridge University Press, Cambridge
(2016), Xx+251 pp. ISBN: 978-0-521-15405-5, MR2722281, first edition in 1981, MR0643363
11. Peters, J.: Computational proximity. Excursions in the topology of digital images. Intell.
Syst. Ref. Libr. 102 (2016), Xxviii+433 pp. https://doi.org/10.1007/978-3-319-30262-1,
MR3727129 and Zbl 1382.68008
12. Peters, J.: Foundations of Computer Vision. Computational Geometry, Visual Image Structures
and Object Shape Detection. Intelligent Systems Reference Library, vol. 124. Springer Inter-
national Publishing, Switzerland (2017), i-xvii, 432 pp. https://doi.org/10.1007/978-3-319-
52483-2, Zbl 06882588 and MR3768717
13. Herstein, I.: Topics in Algebra, 2nd edn. Xerox College Publishing, Lexington (1975), Xi+388
pp. MR0356988; first edition in 1964, MR0171801 (detailed review)
14. Peters, J.: Computational Proximity. Excursions in the Topology of Digital Images. Intelligent
Systems Reference Library, vol. 102. Springer, Berlin (2016), viii+445 pp. https://doi.org/10.
1007/978-3-319-30262-1
15. Flegg, H.: From Geometry to Topology. Crane, Russak & Co., Inc., New York (1974), xii+186
pp.; published by Dover Publications, Inc., Mineola (2001), xiv+186 pp. ISBN: 0-486-41961-4,
MR1854661
16. Alexandroff, P.: Simpliziale approximationen in der allgemeinen topologie. Math. Ann. 101(1),
452–456 (1926). MR1512546
17. Boltyanskiı̆, V., Efremovich, V.: Intuitive Combinatorial Topology. Springer, New York (2001),
Xii+141 pp. ISBN: 0-387-95114-8, trans. from the 1982 Russian original by A. Shenitzer,
MR1822150
18. Gellert, W., Küstner, H., Hellwich, M., Kästner, H.: The VNR Concise Encyclopedia of Math-
ematics. Van Nostrand Reinhold Co., New York (1977), 760 pp. (56 plates). ISBN: 0-442-
22646-2, MR0644488; see Mathematics at a glance, A compendium. Translated from the
German under the editorship of K. A. Hirsch and with the collaboration of O. Pretzel, E. J. F.
Primrose, G. E. H. Reuter, A. Stefan, A. M. Tropper and A. Walker, MR0371551
19. Hilbert, D., Cohn-Vossen, S.: Geometry and the Imagination. AMS Chelsea Publishing, New
York (1952), Ix+357 pp. ISBN: 978-0-8218-1998, trans. by P. Nemény, MR0046650
20. Rowland, T., Weisstein, E.: Geodesic. Wolfram Mathworld (2017). http://mathworld.wolfram.
com/Geodesic.html
21. Weyl, H.: Raum. Zeit. Materie. (German) [Space. Time. Matter], 7th edn. Springer, Berlin
(1988), Xvi+349 pp. ISBN: 3-540-18290-X, MR0988402
22. Ahmad, M., Peters, J.: Geodesics of triangulated image object shapes. Approximating image
shapes via rectilinear and curvilinear triangulations, pp. 1–28 (2017). arXiv:1708.07413v1
23. Faraway, J., Trotman, C.A.: Shape change along geodesies with application to cleft lip surgery.
J. R. Stat. Soc. Ser. C 60(5), 743–755 (2011)
24. Yurkin, V.: Natural Dialectics. Local. Moscow, Russia (1940–1994). Unpublished monograph,
translated by Alexander Yurkin
25. Adhikari, M.: Basic Algebraic Topology and Its Applications. Springer, Berlin (2016),
Xxix+615 pp. ISBN: 978-81-322-2841-7, MR3561159
26. Pranav, P., Edelsbrunner, H., van de Weygaert, R., Vegter, G.: The topology of the cosmic web
in terms of persistent Betti numbers. Mon. Not. R. Astron. Soc. 1–31 (2016). https://www.
researchgate.net
27. Peltier, S., Ion, A., Haxhimusa, Y., Kropatsch, W., Damiandn, G.: Computing homology group
generators of images using irregular graph pyramids. In: Escolano, F., Vento, M. (eds.) Graph-
Based Representations in Pattern Recognition, pp. 283–294. Springer, Berlin (2007). Zbl
1182.68334
28. McAllister, B.: Cyclic elements in topology, a history. Am. Math. Mon. 73, 337–350 (1966).
MR0200894
29. Leader, S.: On clusters in proximity spaces. Fundam. Math. 47, 205–213 (1959)
References 183
30. Moschovakis, Y.: Notes on Set Theory. Undergraduate Texts in Mathematics, 2nd edn. Springer,
New York (2006), Xii+276 pp. ISBN: 978-0387-28722-5, MR2192215; first edition in 1994,
MR1260432 (detailed review)
31. Alexandrov, P.: Combinatorial Topology. Graylock Press, Baltimore (1956), xvi+244 pp. ISBN:
0-486-40179-0
32. Rotman, J.: The Theory of Groups. An introduction, 4th edn. Springer, New York (1965, 1995),
xvi+513 pp. ISBN: 0-387-94285-8, MR1307623
33. Nye, J.: Natural Focusing and Fine Structure of Light. Caustics and Dislocations. Institute of
Physics Publishing, Bristol (1999), xii+328 pp. MR1684422
34. Lynch, D., Livingston, W.: Color and Light in Nature. Cambridge University Press, Cambridge
(2001). ISBN: 978-0-521-77504-5
35. Worsley, A.: The formulation of harmonic quintessence and a fundamental energy equivalence
equation. Phys. Essays 23(2), 311–319 (2010). https://doi.org/10.4006/1.3392799. ISSN 0836-
1398
36. Zangwill, A.: Modern Electrodynamics. Cambridge University Press, Cambridge (2013),
xxi+977 pp. ISBN:978-0-521-89697-9/hbk, Zbl 1351.78001
37. Curtis, J., Grier, D.: Structure of optical vortices. Phys. Rev. Lett. 90, 133,901 (2003). http://
physics.nyu.edu/grierlab/vortex5c/vortex5c.pdf
38. Baldomir, D., Hammond, P.: Geometry of Electromagnetic Systems. Clarendon Press, Oxford
(1996). xi+239 pp. Zbl 0919.76001
39. Meijer, D.: Processes of science and art modeled as a horoflux of information using toroidal
geometry. Open J. Philos. 8, 365–400 (2018). https://doi.org/40.44236/ojpp.2018.84026
40. Tucker, W., Bailey, H.: Topology. Sci. Am. 182(1), 18–25 (1950). http://www.jstor.org/stable/
24967355
41. Cooke, G., Finney, R.: Homology of Cell Complexes. Based on Lectures by Norman E. Steen-
rod. Princeton University Press and University of Tokyo Press, Princeton (1967), xv+256 pp.
MR0219059
42. Whitehead, J.: Combinatorial homotopy. I. Bull. Am. Math. Soc. 55(3), 213–245 (1949).
Part 1
43. Switzer, R.: Algebraic Topology – Homology and Homotopy. Springer, Berlin (2002), xii+526
pp. Zbl 1003.55002
44. Alexandroff, P.: An Introduction to the Theory of Groups. Translated from the German by Hazel
Perfect. Blackie & Son Ltd, London-Glasow (1954, 1959, 1968), ix+112 pp. MR0277594
45. Zaka, O.: Image understanding and applications of symmetry groups. J. Algebra Comput. Appl.
1(1), 20–30 (2011). MR2862509
46. Peters, J.: Proximal planar shapes. Correspondence between triangulated shapes and nerve com-
plexes. Bull. Allahabad Math. Soc. 33 113–137 (2018). MR3793556, Zbl 06937935. Review
by D, Leseberg (Berlin)
47. Peters, J., Ramanna, S.: Shape descriptions and classes of shapes. A proximal physical geometry
approach. In: Stańczyk, U., Zielosko, B., Jain, L. (eds.) Advances in Feature Selection for Data
and Pattern Recognition, pp. 203–225. Springer, Berlin (2018). MR3895981
Chapter 4
What Nerve Complexes Tell Us About
Image Shapes
Abstract This chapter explores what nerve structures in cell complexes tell us in
approximating shapes revealed by light reflected from curved surfaces. Recall that
a nerve, in its simplest form, is a collection of nonempty sets that overlap. That is,
the parts of a nerve have nonempty intersection. This simplest form of a nerve was
introduced by H. Edelsbrunner and J. L. Harer in their monograph on Computational
Topology [1, Sect. 3.2, p. 59]. Of great interest here are the two forms of nerve
complexes introduced by Alexandroff [2], namely.
Alexandroff nerve Collections of triangles with vertices that are various forms of
seed points and with a common vertex in a cell complex on a triangulated bounded
surface region.
Alexandroff star nerve Collections of triangles with vertices that are barycenters
and with a common vertex in a cell complex on a triangulated bounded surface
region (also called a barycentric star nerve).
4.1 Introduction
Here, the focus is on nerve complexes composed of skeletons that have either a
common vertex or a common edge.
Apart from their very interesting structures in and of themselves, nerve complexes
help reveal the geometry of surface shapes that are part of every visual scene. For
example, in Fig. 4.1, a nerve is displayed as a collection of filled triangles (shaded
green) covering surface shapes such as vehicles, overhead intersection street lights,
buildings and trees that are part of a triangulated drone video frame.1
Physical shapes have signatures defined by the distribution and character of the
holes in surface shape interiors. Recall that a hole in a physical surface is that part
of the surface recognized by its dark interior and which absorbs flows of photons
bombarding the hole. Holes in visual scenes appear as dark blobs surrounded by
1 Many thanks to Enze Cui for the drone video containing this video frame.
surface regions that reflect light. What we normally think of as a surface shape with
a recognizable contour, is actually a very complex surface subregion with a nonempty
interior filled with holes. The holes (dark regions) in the interior of a shape serve
as a precise means of distinguishing between surface shapes. By deriving nerve
complexes derived from centroids of the holes of physical shape interiors, we arrive
at a means of distinguishing between surface shapes (Table 4.1).
Barycentric star nerves were introduced by Alexandroff [2, Sect.33, p. 39]. Let K
be a cell complex derived from a triangulation of a set of seed points S on a finite,
bounded, planar region. An Alexandroff nerve A (denoted by Nrv Alex A) is defined by
4.2 Alexandroff Barycentric Star Nerves 187
Alexandroff Nerve
⎧ ⎫
⎨ ⎬
Nrv Alex A = ( p, q, r ) ∈ Nrv Alex A : ( p, q, r ) = ∅ .
⎩ ⎭
p,q,r ∈S
(a) (b)
(see Fig. 4.2b). This leads to the introduction of a barycentric star nerve Nrv star B
shown in detail in Fig. 4.3. Notice that the nucleus p of Nrv Alex A is not included as
a vertex in Nrv star B, since p is not a barycenter. The nucleus of Nrv star B is vertex b5 ,
since this vertex is common to the barycentric triangles in Nrv star B. “
The barycentric star nerve in Fig. 4.3 is a picture proof of the following theorem.
This section introduces what are known as Pham polytopes on video frames proposed
by Pham [3]. A Pham polytope P (denoted by tope Pham P) is a convex hull of a finite
set of picture points in the interior and on boundary of an Alexandroff nerve on a
triangulated video frame. Recall that an Alexandroff nerve is a collection of filled
triangles with a common vertex in a cell complex K . In the case where the filled
triangles in cell complex K are on a video frame I mg, the interior of each is a set
of picture points inside the boundary of . Let Nrv Alex A be an Alexandroff nerve,
bdy(Nrv Alex A) the nerve boundary and let int() denote the interior of a triangle in
Nrv Alex A. Then
Pham polytope
tope Pham P = bdy(Nrv Alex A) ∪ int() .
∈Nrv Alex A
Fig. 4.4 Alexandroff nerve maps to pham polytope on a drone video frame
190 4 What Nerve Complexes Tell Us About Image Shapes
Problem 4.4 K
Implement Application 4.3. Give two sample videos containing highlighted Pham
polytopes with similar areas. “
Problem 4.5 ®
Prove that every oriented filament skeleton derived form the boundary of a Pham
polytope has a Betti number equal to 1. “
This section carries forward the Pham polytope structure to construct skeletal nerves
on a CW cell complex K on a finite, bounded planar region. Notice that the cell
complex K is a collection of skeletons. Let sk( Ai ), 1 < i ≤ n be a filament skeleton
on K . From Sect. 2.6, recall that a collection of skeletons in a CW complex defines
a skeletal nerve A (denoted by skNrvA) is defined by
skNrvA = sk(Ai ) ∈ K : sk(Ai ) = ∅ (Skeletalnerve).
1<i≤n
Notice that a filament skeleton skA can be extracted from the boundary of a Pham
polytope tope Pham A. That is, we obtain an oriented filament skeleton A (denoted
by skA) from tope Pham A in the following way.
For simplicity, we write skA instead of skA, assuming that we are working with
an oriented filament skeleton extracted from the boundary of a Pham polytope.
4.4 Skeletal Nerves Derived from Intersecting Polytopes 191
Fig. 4.5 Skeletal nerve derived from pham polytope boundaries on a drone video frame
Example 4.6 A skeletal nerve skNrvA derived from a collection of oriented filament
skeletons on the boundaries of Pham polytopes on a triangulated drone video frame
is shown in Fig. 4.5. A cell complex K results from the triangulation of a video
frame. The cells in K are 2-cells (filled triangles). In this case,
Skeletal nerve skNrv A on skA1 , skA2 , skA3
skNrvA = skAi ∈ K : skAi = p .
1<i≤3
That is, from the triangulated video frame in Fig. 4.5, we obtain a skeletal nerve that
is a collection of three oriented filament skeletons skA1 , skA2 , skA3 having vertex
p in common. “
This section explores free Abelian group representations of skeletal nerves that appear
in triangulations of video frames. A group G(◦) with binary operation ◦ is Abelian,
provided the ◦ commutes for each pair of group elements g, g , i.e.,
Abelian property
g ◦ g = g ◦ g.
In other words, the order in which we apply the group operation does not make a
difference.
Munkres [4, Sect. 1.4, p. 21] observes that an additive Abelian group G(+) is
free, provided every element g of G can be written as a linear combination of its
generators a1 , . . . , ai . . . , ak , i.e. (Fig. 4.6),
g ∈ G,
positive integers or 0
m1, . . . , mk ∈ Z0,+ .
g = m 1 a1 + · · · + m i ai + · · · + m k ak .
x =m·a
m copies of length of generator a
= a + a + ··· + a
In other words, filament a is a generator of the other members of the skeleton skA,
since each filament of skA is a multiple of filament a. This means that the skeleton
skA has an additive cyclic group representation. We can carry this idea a step further
by considering a skeletal nerve defined on a set of intersecting filament skeletons .
Example 4.7 Consider what happens when we have a skeletal nerve skNrvE con-
taining a pair of filament skeletons skA, skB with generators a , b, respectively.
That is, every segment of skA is a multiple of segment a and every segment of skB
is a multiple of segment b. These two filament skeletons have an edge in common,
the nucleus of a nerve skNrvE. Then
Skeletal nerve
>
skNrvE = skA, skB : skA ∩ skB = segmentb .
Let g be a filament of one of the skeletons in nerve skNrvE. Then we know that we
can write
Linear combination of generators a , b
g= m 1 a + m 2 b.
For example, in skA in Fig. 4.7, the length of every segment in the skeleton is a
multiple of the length of the segment labeled a. Similarly, the length of the every
segment in skB in Fig. 4.7 is a multiple of the length of the segment labeled b. Then,
for instance, let g be the length of the path between vertices p and q in Fig. 4.7. Then
In other words, the length of the path between any pair of vertices in the nerve can
be represented as a linear combination of the pair of generators. “
Life gets easier, if we map the coefficients in a linear combination of nerve gener-
ators to numbers modulo 2 (or your favorite modulus such 3 or 5). Recall the clock
arithmetic for k mod 2, which is the remainder after division by 2. For example,
From Example 4.7, let p be a vertex in a filament skeleton skA and let q be a
vertex in filament skeleton skB. Also from Example 4.7, let skA, skB have a common
edge, forming a skeletal nerve. We can then represent the length of the path g between
vertices p and q very simply in the following way.
g = 2a + 3b
maps to the sum of the coefficients mod 2
−→ 0mod 2 + 1mod 2
= (0 + 1)mod 2
= 1.
That is, the length of every path is now either 0 or 1. In that case, 0 is the identity
element. Also, every linear combination of the coefficients modulo 2 has an additive
inverse, namely, itself. To see this, let x be a linear combination of the coefficients
modulo 2. Then
(x + x)mod 2 = (2x)mod 2 = 0.
In effect, each of the skeletons skA, skB can be represented as a cyclic group on the
linear combination of the coefficients modulo 2. This also means that every skeletal
nerve skNrvA can be represented as a free Abelian group, since every member of
skNrvA can be written as a linear combinations of its cyclic group generators.
So what do we gain by viewing a skeletal nerve on a triangulated video frame as
a free Abelian group?
4.5 Free Abelian Group Representation of a Video Frame Skeletal … 195
To answer this question, notice that every skeletal nerve on a triangulated video
frame has a shape, namely, the boundary plus the interior defined by the collection
of oriented filament skeletons of the nerve. In other words, the shape of a skeletal
nerve is determined by the shape of its intersecting filament skeletons. Variations
in the number and shape of the filament skeletons in a nerve will usually result in
variations in the shape of a nerve.
From an application perspective, we now have a simple means of tracking changes
in nerve shapes in video frames, namely, the Betti number associated with every skele-
tal nerve. Each Betti number for a skeletal nerve tells us the number of connected
filament skeletons (read cyclic group generators) in the nerve. A skeletal nerve
that appears in the interior of one video frame and later moves close to the border
of another video frame will lose one or more filament skeletons that are cut off by
the later frame border. Recall that every centroid-based maximal nucleus cluster or
MNC (i.e., maximal number of triangles in a collection of triangles with a common
vertex) defines a skeletal nerve covering a dominant part of a video frame. This is so,
since a centroidal MNC is constructed from the highest number of centroids among
all of the nucleus clusters in a triangulated video frame. Put another way, a centroidal
MNC has the highest number of holes among all of the nucleus clusters in a video
frame. To derive a barycentric skeletal nerve from an MNC, we find the barycenters
of the triangles in the MNC and in the triangles of the nucleus clusters surrounding
the MNC. By attaching filaments to the barycenters, for example, we derive fila-
ment skeleton between each pair of barycenters. We repeat the same construction on
barycenters of the triangles along the border of an MNC to obtain a second filament
skeleton. Notice that the interior of each filament skeleton is nonempty. This means
that the innermost filament skeleton (in an MNC) is common to (intersects) each
of the outer filament skeletons. Hence, the collection of filament skeletons spiraling
outwards has the MNC skeleton in common, deriving a skeletal nerve.
Importantly, see also Giblin and Janeczko [9] on families of reflections of light
colliding with curved surfaces and on projections of bounded surfaces by Bruce and
Giblin [10] (see, for example, the geometry of lips and beaks in Fig. 5 and boundary
cusp in Fig. 7 [10, p. 411 and p. 413]). From this work on surface reflections by J.F.
Nye and P. Giblin and others, the natural affinity between reflected light waves and
filament skeletons comes as no surprise. Spiraling filament skeletons also provide
an analogue of an electromagnetic vortex formed by photons that possess some
net angular momentum along a longitudinal axis. This is an observation by I.V.
Dzedolik [11].
Example 4.8 In Example 4.7, the Betti number of skNrvE equals 2, since the free
Abelian group representation of this nerve is defined by the direct sum of a pair of
cyclic groups, namely, the cyclic groups representing skA, skB in skNrvE. “
Example 4.9 In Example 4.6, the Betti number of skNrv A equals 3, since the
free Abelian group representation of this nerve is defined by the direct sum of a
pair of cyclic groups, namely, the cyclic groups representing skA1 , skA2 , skA3 in
skNrvA. “
Example 4.10 (Two Systems of Nerve Complexes) Two sample systems of nerve
complexes is shown in Fig. 4.8, namely,
Fig. 4.8 Two systems of intersecting nerves Nrv sys A, Nrv sys B
“
198 4 What Nerve Complexes Tell Us About Image Shapes
If this covering condition fails, then we concatenate Nrv sys A to an adjacent nerve
system Nrv sys B to derive a galaxy of nerve systems Nrv galaxy G that possibly satisfies
the covering condition, namely,
In the case of an MNC derived from surface region centroids (centers of mass of
surface holes), an MNC has the highest concentration of centroids as its triangle
vertices. For this reason, a centroid-based MNC will usually overlap a dominant
surface shape. Gathering together intersecting nerve systems provide a more inclu-
sive surface shape cover. The goal is to satisfy one of these surface shape cover
conditions with minimal overlap between a shape boundary and surface region exte-
rior to a particular surface shape of interest. “
Also in this section, we must pay the doctor’s fee (so to speak) before we can
solve the problem of covering a shape on a triangulated curved surface in a minimal
fashion. The fee in our case is the analogue of the fee whimsically mentioned by
Punch [12, p. 41] in reporting an exchange between a patient and a doctor. Such
triangles obtained from the centroids of holes in shape interiors tend to spread across
surface shape boundaries. In other words, centroidal-based triangles usually overlap
shape boundaries and the regions exterior to shape boundaries. This leads to what is
known as the shape boundary overlap problem. When we rely on Alexandroff nerve
complexes to cover surface shapes, the bulging effects of the spread of triangles in
4.7 Galaxy of Systems of Nerve Complexes 199
Example 4.12 A sample barycentric filament skeleton (call it skA) derived from
the barycenters of the triangles of a maximal nucleus cluster (MNC, which is an
200 4 What Nerve Complexes Tell Us About Image Shapes
Alexandroff nerve) is shown in Fig. 4.9. Notice that skA covers a smaller surface
of the visual scene than the triangles in its parent MNC. This is a beginning of the
solution of the shape boundary overlap problem. “
Fig. 4.10 System of intersecting skeletal nerves skNrv sys A, skNrv sys B
When it is clear from the context that the skeletal nerves are derived from barycen-
ters, we write skNrv sys A instead of the more cumbersome skNrvsys barycentric A,
Example
4.13 (System of Intersecting Skeletal Nerve Complexes) Let skE = e A1 ,
e A2 , a filament skeleton containing edges common to the barycentric skeletal nerve
complexes skNrvA, skNrvB in Fig. 4.8, namely,
This section represents a return to the problem of finding a way to satisfy the shape
cover condition introduced in Motivation 4.7 in Sect. 4.6. We want to do is to tease
a surface shape cover by consider nesting, non-concentric vortex filament skeletons
that have one more spokes connected between them (see, e.g., Fig. 4.11).
Pairs of spokes connected between filament skeleton vertices (close to each other)
are added successively. With each new pair of neighbouring spokes attached between
filament skeletons, a shape E is born. This new shape E is called a filament spoke
4.9 Filament Spoke Shapes and the Importance of Closure 203
shape (denoted by skShapeE). Filament spoke shapes have a boundary derived from
a pair of spokes connected to a pair of filaments and a nonempty interior that partially
covers a surface shape.
Let skA, skB be barycentric filament skeletons on the barycenters of an MNC
X and on the barycenters of triangles on the border of MNC X , respectively. Let
skE 1 , . . . , skE i , skE i+1 , . . . , skE k be a collection of barycentric filament spokes.
Spokes skE i , skE i+1 are next to each other. Also, let b M N C , bM N C be a pair of
>
neighbouring barycenters on a pair of border triangles of MNC X and let b M N C bM N C
>
be a barycentric filament. In addition, let bskB bskB be a filament on the barycenters
on the triangles on the MNC X border. Recall that the closure of a set includes both
the boundary and the interior of the set. Then skShapeE is defined by
Example 4.14 A sample filament spoke shape skShapeE is shown in Fig. 4.9. Let
filament A, filamentB be filament spokes connected between vertices in opposite
filaments in skeleton skE and skeleton skB. This filament spoke shape sk ShapeE
is then defined by
The closure of {e6 ∪ filament A ∪ filamentB ∪ e5 } gives us both the boundary and
the interior of this filament spoke shape. “
Each new filament spoke shape partially covers a region on a triangulated surface.
Filament spoke shapes are a lot like pieces of a puzzle. Initially, we just put things
together using only two pairs of barycenters as the vertices of 1-cells (line segments)
attached to filament skeletons that are opposite each other. Doing this just once results
in two important structures, namely,
204 4 What Nerve Complexes Tell Us About Image Shapes
Proof By definition, the closure of skShapeE includes both the boundary and the
interior of a surface region. Hence, by definition, skShapeE is a planar shape.
Proof
1o From Lemma 4.15, filament spoke shape skShapeE includes the boundary
bdy(skShapeE). By definition of a filament spoke shape, the boundary is
> > > >
a sequence filaments
bb , bb , b b , b b attached to pairs of neighbouring
barycenters in b, b , b , b . Hence, bdy(skShapeE) is a filament skeleton.
2o Immediate from Part 4.9, since there is a sequence of filaments between
any pair of vertices in the barycentric filament skeleton on the boundary of
skShapeE.
An analogue of the closure of a set of cells attached to each other is the closure
of a tree leaf T (denoted by clT ). The boundary of T (bdyT ) is the set of leaf
cells along the edge of the leaf. And the interior of T (intT ) is set of veins
attached to a central leaf vein as well as the leaf cells in between the leaf veins
and in that part of a leaf bounded by the leaf edges. “
4.9 Filament Spoke Shapes and the Importance of Closure 205
Then
Pair of 0-cells
bdy(filament A) = ◦v, ◦v .
1-cell (edge)
int(filament A) = –.
Closure of a filament
cl(filament A) = ◦v, ◦v ∪ –.
In other words, the closure of filament A is all encompassing. That is, cl(filament A)
encompasses the boundary of filament A as well as all of the entire edge in the interior
of filament A.
In this section, we take a closer look at filament skeletons. The focus is on filament
skeletons that are 1-cycles. Recall from Sect. 1.12 that a 1-cycle is a finite, collection
of vertices (0-cells) connected by oriented edges (1-cells) that define a simple, closed
path so that there is a path between any pair of vertices in the collection.
Let v1 , v2 , . . . , vi−1 , vi , . . . , vn−1 , vn be n path-connected vertices (0-cells). A
cyclic filament skeleton E (denoted by sk cyclic E) is a 1-cycle on n vertices defined
by
In the sequel, the vertices of each cyclic filament skeleton are barycenters. So we
usually write sk cyclic E, instead of skCyc barycentric E.
Proof Let sk cyclic E be a cyclic filament skeleton. Let a be a filament in sk cyclic E that
has the shortest length. Assume that the length of each filament equals ma, a multiple
of the length of a. Summing on the coefficients modulo 2, we can map ma + m a
to (m + m )mod2. This gives us additive cyclic group (a , +2 ) representation of
sk cyclic E.
Now that we have cyclic filament skeletons working for us, we can begin to
consider a new class of shapes useful in approximating surface shapes. Let A cyclic
filament skeletal shape A (denoted by sk cyclic Shape A) on a cyclic filament skeleton
sk cyclic E, defined by
Closure of sk cyclic E
sk cyclic Shape A = cl(sk cyclic E)
sk cyclic E boundary unioned with the sk cyclic E interior
= bdy(sk cyclic E) ∪ int(sk cyclic E).
In the search for good approximations of surface shapes, it helps to imitate the
envelope created by a light caustic that is reflected off a shiny surface. After all, we
want to cover an unknown shape on a curved surface displayed in a digital image
with a known geometric shape. The closer a geometric shape covering is to a surface
shape, the more effective the geometric shape will be in approximating the surface
shape.
Recall from Observation 4.5 that a light caustic results from light reflected from
a curved surface. By way of illustration, Nye [6, Sect. 2.1, p. 9] introduces a coffee
cup caustic. Starting with parallel rays of light striking the inner curved surface of a
208 4 What Nerve Complexes Tell Us About Image Shapes
(a) (b)
(c)
filled coffee cup and, after reflection, envelop (cover) in three dimensions a curved
surface, which is the caustic. This caustic can be seen on the coffee surface during a
bright sunny day (see, for example, the sample coffee cup caustic2 in Fig. 4.14a). A
coffee cup caustic has two parts, namely,
cusp: Each of the pointed ends of the crescent of a coffee cup caustic, formed by
light rays making smaller angles on a coffee surface.
fold caustic: The tails of a coffee cup caustic, formed by light rays making larger
angles on a coffee surface.
In general, Wright [13] observes that a caustic is the envelope along which geo-
metrical light rays coalesce. The coffee cup caustic is an example of what is known
as photonic catastrophe. Photonic catastrophe theory comes into play here in the
2 Many thanks to S. Ramanna for capturing this caustic during a sunny afternoon in Manitoba.
4.11 Nye Coffee Cup Caustics in Optical Vortex Nerves 209
study of sudden and dramatic shifts in the behavior of reflected light. For a recent
study of photonic catastrophe, see Longhi [14].
Example 4.20 A Nye coffee cup caustic is shown in Fig. 4.14a. A geometric rendition
of this caustic is shown in Fig. 4.14b. Each arrow in Fig. 4.14b represents a ray of
light (flow of photons) colliding with and reflecting off the inner curved surface of the
coffee cup. Parallel rays of light bombarding the inner cup curvature above a coffee
surface and reflecting off the curved surface have their behaviour changed abruptly
as reflected in the folds and cusp of the resulting light caustic. In other words, an
example of a photonic catastrophe is displayed in Fig. 4.14a.
A repetition of the filament structures inspired by this caustic is found in the nested,
non-concentric cyclic filament skeletons (skeletal vortexes) attached to connecting
1-cells (edges or spokes), which is partially represented by the intersecting yellow
lines climbing up one of the spokes in Fig. 4.14c. The upper end of each of the spokes
in Fig. 4.14c models the cusp of a light caustic that sits on the curved shape of a
petal on a morning glory. This is known as a filament cusp. In Fig. 4.14c, the fold
of a light caustic is modelled by a pair of filament skeletons that intersect the base
of the filament cusp. This pair of filament skeletons is known as a filament skeleton
fold. “
All the while, notice that we are constructing tiny compartments (we call them
skShapeE shapes) covering some portion of an underlying surface shape. Taken
together, we obtain an instance of an optical vortex nerve sk cyclic NrvE defined by
∪ sk cyclic E 1 .
210 4 What Nerve Complexes Tell Us About Image Shapes
Example 4.21 A sample optical vortex nerve sk cyclic NrvE covering a surface shape
in a video frame is shown in Fig. 4.15. In this sample frame, the nerve sk cyclic NrvE
is constructed from the barycenters the triangles of an MNC on the front of a freight
train engine. The center of the MNC is a centroid of a surface hole found on the front
the train engine. Notice that the MNC itself is much larger than the surface shape
covered by the filament skeleton (call it sk cyclic E 0 ) in the interior of the MNC. That
is, sk cyclic E 0 overlaps a great deal less of the surface outside the engine front than
the MNC. The red spokes ◦–◦ attached to vertices in sk cyclic E 0 and the neighbouring
vertices on the outermost filament skeleton sk cyclic E 1 represent the analogue of the
cusp of coffee cup caustic. The filaments on sk cyclic E 0 radiating outward from the
cusps in Fig. 4.15 are the analogues of fold caustics make the larger angles of the
rays of sunlight on a coffee cup surface .
Notice that this analogue of a coffee cup caustic is repeated by each of the skeletal
shapes reaching outward along the border of sk cyclic E 0 , the innermost filament skeleton
on the optical vortex nerve. Each of the skShapeE shapes bulges outward to a varying
degree from sk cyclic E 0 . Each skShapeE shape isolates and covers a portion of the
region surrounding sk cyclic E 0 . “
Fig. 4.15 Sample barycentric optical vortex nerve on a train video frame
4.12 Cusp Filaments as Pathways of Reflected Light 211
In this work, the basic approach is to find the barycenters (intersection of triangle
median lines) of centroid-based triangles. By connecting such barycenters with 1-
cycles between triangles along the boundaries of Alexandroff nerves, we identify
pathways for reflected light from surface shapes in visual scenes. The steps to derive
a cusp filament are given in Algorithm 11.
For the geometry of cusp filaments, see Fig. 4.16a. The main thing to notice
in this geometry is that an edge is attached to barycenters of intersecting triangles
(i.e., triangles with a common edge). The end results is a 1-cell that is a cusp fil-
ament, which identifies a pathway for reflected light between surface shape holes.
Vertexes p, q, r, s in Fig. 4.16a represent centroids on holes in a visual scene surface
shape. Edge qr > is common to triangles ( pqr ), (qr s). Vertexes b, b are triangle
barycenters. By attaching an edge to barycenters b, b , we construct a cusp filament.
Example 4.22 (Cusp Filament on a Morning Glory) Vertex p in Fig. 4.16 is a hole
centroid and is also the vertex on one of the triangles in an Alexandroff nerve. One
of the nerve triangles is ( pqr ), represented geometrically in Fig. 4.16a. Each of
the 4 vertexes p, q, r, s is a centroid on a hole on the surfaces of the morning glory
in Fig. 4.16. Vertexes b, b are barycenters on triangles ( pqr ), (qr s). And the
>
cusp filament bb lies on that part of the morning glory surface that is between holes.
Hence, this cusp filament is an example of a pathway for reflected light. “
212 4 What Nerve Complexes Tell Us About Image Shapes
(a) (b)
This section introduces a basic result for optical vortex nerves using the Nye coffee
cup caustic paradigm. In this work, each filament (or spoke) attached between a vertex
on the innermost and outermost skeletons models the cusp of a coffee cup caustic.
Thinking of each spoke as a cusp caustic is motivated by a minimalist approach to
approximating visual scene shapes whose reflected light results in a digital image.
Recall from Observation 4.11 that an optical vortex nerve sk cyclic NrvE contains a
collection of coffee cup caustic cusp filaments (briefly, cusp filaments) attached to
a pair of nesting, non-concentric cyclic filament skeletons. Let sk cyclic E 0 , sk cyclic E 1 be
cyclic filament skeletons in sk cyclic NrvE. From Lemma 4.18, sk cyclic E 0 , sk cyclic E 1 have
cyclic group representations.
Lemma 4.23 A cusp filament in an optical vortex nerve has a cyclic group repre-
sentation.
Theorem 4.25 An optical vortex nerve has a free Abelian group representation.
Proof From Lemma 4.18 and Lemma 4.23, an optical vortex nerve sk cyclic NrvE can
be represented by a collection of cyclic groups, each with its own generator. Hence,
sk cyclic NrvE has a free Abelian group representation.
the structure of nerve sk cyclic NrvE, we have a means of deriving the Betti number of
such a nerve. These numbers are named after Enrico Betti (1823–1892) in Pisa (see
Betti’s portrait3 in Fig. 4.17). Hence, we arrive at a fundamental theorem for optical
vortex nerves.
Theorem 4.26 (Coffee Cup Caustic Cusp Filament Theorem) The Betti number of
a free Abelian group representation of an optical vortex nerve sk cyclic NrvE defined by
a pair of nesting cyclic filament skeletons attached to each other by k cusp filaments,
is k + 2.
Proof Let B(sk cyclic NrvE) be the Betti number of sk cyclic NrvE. From Theorem 4.25,
the free Abelian group representation of sk cyclic NrvE includes k generators of cusp
filament cyclic groups and two generators of the pair of the pair of cyclic groups
representing the nesting, non-concentric nerve cyclic filament skeletons. Hence,
B(sk cyclic NrvE) = k + 2.
Theorem 4.27 (Ahmad Betti Number Measure) Let the innermost cyclic filament
skeleton be an Alexandroff nerve NrvA and let |NrvA| be number of 2-cells in the
nerve. The Betti number of a free Abelian group representation of an optical vortex
nerve sk cyclic NrvE defined by a pair of nesting cyclic filament skeletons attached to
each other by |NrvA| cusp filaments, is |NrvA| + 2.
Proof Immediate from Theorem 4.26 and the fact that each cusp filament is attached
to a barycenter of a triangle in NrvA.
Example 4.28 (Optical Vortex Nerve Betti Numbers) Several different optical vor-
tex nerves on train video frames are shown in Fig. 4.18, namely, sk cyclic NrvE 1 in
3 From http://www-groups.dcs.st-and.ac.uk/history/Biographies/Betti.html.
4 Observed by M. Z. Ahmad.
214 4 What Nerve Complexes Tell Us About Image Shapes
(a) (b)
(c)
Fig. 4.18a, sk cyclic NrvE 2 in Fig. 4.18b and sk cyclic NrvE 3 in Fig. 4.18c. Each optical
vortex nerve has an innermost cyclic filament skeleton sk cyclic E 1 derived from the
barycenters of the triangles on a maximal nucleus cluster (MNC) on a triangulated
video frame. And each of these vortex nerves has a second cyclic filament skeleton
sk cyclic E 2 derived from the barycenters of the triangles along the border of the MNC.
A collection of cusp filaments are attached between the vertices on sk cyclic E 1 and
sk cyclic E 2 .
1o The optical vortex nerve sk cyclic NrvE 1 in Fig. 4.18a contains a pair of cyclic
filament skeletons attached to 8 cusp filaments. From Theorem 4.26,
2o The optical vortex nerve sk cyclic NrvE 2 in Fig. 4.18b contains a pair of cyclic
filament skeletons attached to 9 cusp filaments. From Theorem 4.26,
3o The optical vortex nerve sk cyclic NrvE 3 in Fig. 4.18c contains a pair of cyclic
filament skeletons attached to 10 cusp filaments. From Theorem 4.26,
Theorem 4.26 tells us that by counting the number of cusp filaments in video
frame optical vortex nerve, we arrive at a quick means of determining when there is
a change in the visual scenes recorded in a video. In other words, each video frame
has a shape signature represented by the Betti number of the MNC optical vortex
nerve on the frame.
Let sk cyclic NrvE be an optical vortex nerve of a particular video frame Betti number
B(sk cyclic NrvE). Track the persistence of sk cyclic NrvE (appearance and disappearance
of the nerve shape over a sequence of frames) by tracking the persistence of the Betti
number of interest.
Problem 4.29 Do the following.
1o Triangulate the frames in a video using centroids of seed points.
2o Find the barycenters of the frame triangles.
3o Find the barycentric maximal nucleus cluster (MNC) for each video frame. The
MNC is repeated in some frames. To avoid this problem, select an MNC with
a distinguishing feature. For example, pick the MNC with the largest area. If
there is a tie (more than one MNC with the largest area), use random choice as
a tie-breaker.
4o Derive an optical vortex nerve sk cyclic NrvE on each frame MNC.
5o Pick sk cyclic NrvE for an optical vortex nerve of a particular video frame with
Betti number B(sk cyclic NrvE).
6o Construct a chart showing persistence (number of consecutive frames) of
B(sk cyclic NrvE). This can be done with either a histogram or a dendogram.
The height of each spike represents the number of frames having the same Betti
216 4 What Nerve Complexes Tell Us About Image Shapes
number. Each point on the horizontal axis represents a particular Betti number
(the Betti number of interest plus other frame Betti numbers).
7o Give the triangulated video showing highlighted optical vortex nerves like the
ones in Fig. 4.18 and the constructed chart.
8o Repeat Step 1o for two different videos. “
This section identifies some of the sources and pointers to further reading on nerve
complexes.
Centroidal Vortices
For an introduction to maximal centroidal vortices on triangulated digital images,
see Ahmad and Peters [15]. For example, in Fig. 4.19a–c, we have
Closure of a set
Set Interior-Set Boundary Closure. For an introduction to the interior-boundary-
based closure of a nonempty set A (denoted by cl A), see Krantz [16, Sect. 1.2, pp.
7–8, especially p. 8]. Briefly, let bdy A be the boundary of the set A and let intA
be the interior of A. Then clA is defined by
It is this version of the closure of a nonempty set that we use in this chapter.
Proximity-Based Closure. For an introduction to the proximity-based closure
of a set A (denoted by clδ A) defined in terms of the closeness (proximity) of sets
of points to the nonempty set A, see Peters [17, Sect. 1.4, p. 15]. Briefly, let X δ A
4.14 Sources and Further Reading 217
When it is clear from the context what is meant by clδ A, we write cl A, instead.
Closure nerve
For an introduction to closure nerve structures, see Peters [17, Sect. 1.10, p. 31].
Briefly, let F denote a nonempty collection of sets and let clX be the closure of
a subset X in F. Then a closure nerve A (denoted by Nrvcl F) is defined by
Closure nerve
Nrvcl F = X ∈ F : X = ∅ .
1o NrvE
is a collection of convex sets, and
2o is convex.
∈NrvE
(a) (b)
In other words, let boutsideM N C be an edge vertex outside the MNC and nearest to
the MNC boundary. “
References 221
References
Abstract This chapter introduces of two basic types of proximities in the study of
relationships between sub-complexes in cell complexes, namely, spatial and descrip-
tive proximities. These proximities are useful in clustering and separating subcom-
plexes in triangulated finite, bounded surface regions such as those found in visual
scenes. This chapter introduces a number of connectedness proximities useful in
probing, analyzing, comparing and classifying cell complexes on triangulated sur-
face regions.
5.1 Introduction
Example 5.1 (Skeletal overlap with triangular regions) A filament skeleton skE
on a triangulated visual scene of a collection of harvested peppers is shown in
Fig. 5.1. Recall that a filament skeleton is a shape with both nonempty boundary
and a nonempty interior. In this triangulation, notice that
⩕
1o skE overlaps filled triangle A, i.e., skE δ A.
⩕
2o skE overlaps filled triangle B, i.e., skE δ B.
There are many other instances of strong proximity between skE and regions in
Fig. 5.1. What are they? “
A pair of nonempty complexes G, H are far from each other and do not have
spatial proximity (denoted by G δ H ), provided the complexes do not intersect. In
that case, we say that complexes G, H are remote from each other. i.e., the complexes
have no points or edges in common and they do not overlap.
Example 5.2 (Filament Skeleton remote from triangular regions) Two instances of
the remoteness of a skeletal shape skE from filled triangular regions are shown in
the triangulated visual scene in Fig. 5.1, namely,
1o skE is remote from the filled triangle G, i.e., skE δ G.
2o skE is remote from the filled triangle H , i.e., skE δ H .
There are many other instances of remoteness between skE and regions in Fig. 5.1.
What are they? “
A δΦ B). By description, we mean that one or more region features such as shape,
area, colour. For example, the description of complex A is denoted by Φ(A), for a
particular selection of features of complex A. When is a pair of complexes A and U
do not have descriptive proximity (denoted by A δ Φ U ), the description of complex
A does not match the description of complex U .
Example 5.3 (Descriptive proximity between triangular regions) Labelled filled tri-
angles A, B, G, H and skeletal shape sk E are shown in a triangulated visual scene
of a collection of harvested peppers is shown in Fig. 5.1. Let the description Φ of
each triangle be given in terms of shape, e.g., Φ(A) = shape of A. In this triangula-
tion, there are a number of descriptive proximities as well a instances where pairs of
regions lack descriptive proximity, namely,
1o A is descriptively near X for each X in {A, B, G, H } (A and X have the same
shape), i.e., A δΦ X .
2o skE is descriptively far from each X in {A, B, G, H } (skE and X do not have
the same shape), i.e., skE δ Φ X .
There are many other instances of the absence of descriptive proximity between skE
and regions in Fig. 5.1. What are they? “
An overview of the proximity landscape is given in the life and work of Naimpally
(Som) [1]. This is a remarkable story of a mathematician who began studying prox-
imity space theory after he completed his Ph.D. as a result of a chance meeting at the
University of Michigan between Som and a visitor from Cambridge University Press,
who invited him to write a monograph on proximity. This he did together with his
graduate student B.D. Warrack, leading to a complete overview of proximity space
theory until 1970 [2] (Fig. 5.3).
Origins of Proximities.
K The study of the nearness of sets now spans more than 100 years, starting
with the address by F. Riesz at the International Congress of Mathematicians
in Rome in 1908 [3], recently commented on by Naimpally [4, 5] and Di Con-
cilio [6–8]. One of the earliest introductions to nearness (proximity) relations
was given by E. Čech during a 1936–1939 Brno seminar, published in 1966 [9,
Sect. 25.A.1]. Čech used the symbol p to denote a proximity relation defined on
a nonempty set X , which Čech axiomatized. Čech’s work on proximity spaces
started two years after V.A. Efremovič’s work (in 1933), who introduced a
widely considered axiomatization of proximity, which was not published until
1951 [10]. For a detailed presentation of Efremovič’s proximity axioms, see,
e.g., [8, 11] and for applications, see, e.g., [12–16]. “
1 Many thanks to Braden Cross for the webcam image in Fig. 5.2, captured using the Matlab Com-
puter Vision System toolbox and Matlab implementation of the Canny edge detection algorithm.
5.3 What Is a Proximity Space? 227
Before we start considering the axioms for various forms of proximity spaces, it
helps to step back and ask the question What is a proximity space? Recall that any
nonempty set with its distinguishing properties is what we call a space. So we might
wonder when a nonempty set merits the name proximity space.
What is a proximity space?. Detectable Closeness of any two sets in the space .
K A finite nonempty set P is a proximity space, provided the closeness or
farness (remoteness, i.e., non-intersection) of any two subsets in P can be
determined. A proximity space P is sometimes called a δ-space by Smirnov
[17]. “
This section introduces the simplest of the proximity spaces called the C̆ech prox-
imity space. A proximity space P is sometimes called a δ-space [17], provided P is
equipped with a relation δ that satisfies, for example, the following C̆ech axioms for
sets A, B, C ∈ 2 P [9, Sect. 2.5, p. 439].
δ C̆ech Proximity.
C̆ech axioms
P1 All subsets in P are far from the empty set.
P2 A δ B =⇒ B δ A, i.e., A close to B implies B is close to A.
P3 A δ (B ∪ C) ⇔ A δ B or A δ C.
P4 A ∩ B = ∅ =⇒ A δ B (Closeness Axiom).
228 5 Surface Shapes and Their Proximities
A space P equipped with the C̆ech proximity (denoted by (P, δ)) is called a C̆ech
proximity space.
Let 2 P denote the collection of all subsets of a nonempty set P. We adopt the
convention for a proximity metric δ : 2 P ×2 P −→ {0, 1} introduced by Smirnov [17,
Sect. 1, p. 8]. We write δ(A, B) = 0, provided subsets A, B ∈ 2 P are close and
δ(A, B) = 1, provided subsets A, B ∈ 2 P are not close, i.e., there is a non-zero
distance between E and H . Let A, B, C ∈ 2 P . Then a proximity space satisfies the
following properties.
Smirnov Proximity Space Properties
Q1 If A ⊆ B, then for any C, δ(A, C) ≥ δ(B, C).
Q2 Any sets which intersect are close.
Q3 No set is close to the empty set.
In a C̆ech proximity space, the Smirnov proximity space property Q3 is satisfied
by axiom P1 and property Q2 is satisfied by axioms P2–P4, i.e., any subsets of P
are close, provided the subsets have nonempty intersection. That is, A close to B
implies B is close to A (axiom P2). Similarly, A close to B ∪ C implies A is close to
B or A is close to C (axiom P3) or A is close to B ∩ C (axiom P4). Let A ∩ C = ∅.
Then δ(A, C) = 1, since A has no points in common with C. Similarly, assume
B ∩ C = ∅. Then, δ(B, C) = 1, since B and C have no points in common. Hence,
property Q1 is satisfied, since
Proof. The proof is given by Willard [18, Sect. 26.4, p. 193]. For a new kind of
connectedness in which nonempty intersection is replaced by strong nearness, see
Guadagni [19, p. 72] and in Peters [20, Sect. 1.16].
Example 5.9 (Connected Nerve Space) Let NrvA be an Alexandroff nerve complex
on a triangulated planer surface region. Each collection of subsets in nerve Nrv A is
connected, since there are no disjoint open sets E, B in NrvA such that E ∪ B =
NrvA. Recall that NrvA is a collection of filled triangles with a common vertex
with the following properties:
230 5 Surface Shapes and Their Proximities
Problem 5.11 K
Prove Theorem 5.10. Hint: Let sk cyclic N be the innermost cyclic skeleton in an optical
vortex nerve sk cyclic NrvE and let sk cyclic A be a cyclic skeleton with an edge filament
in common with sk cyclic N . Define a vortex nerve spoke spokeN A to be
Use the fact that every pair of spokes in sk cyclic NrvE has nonempty intersection and
that sk cyclic NrvA is a connected space and that the union and intersection properties
are satisfied. “
Problem 5.13 ®
Prove Theorem 5.12. That is, for each optical vortex nerve sk cyclic NrvA, prove that
we can always find a path (sequence of filaments) between p, q for every pair of
vertices in sk cyclic NrvA. “
5.5 Connectedness Proximity Space 231
Problem 5.14 K
For each optical vortex nerve sk cyclic NrvA, invent a vertex proximity measure δ( p, q)
for pairs of vertices p, q in sk cyclic NrvA similar to the Smirnov proximity measure.
That is, define δ( p, q) so that this vertex proximity measure returns a value that is a
measure of the closeness or remoteness of vertices in sk cyclic NrvA. A refined version
of δ( p, q) would compute the lengths of the filaments between p, q in sk cyclic NrvA,
not just count the number of filaments between p, q in sk cyclic NrvA. “
conn
In this work, connectedness is defined in terms of the connectedness proximity δ
⩕
conn
and overlap connectedness δ . In both cases, nonempty intersection is replaced by
a connectedness proximity in the study of connected cell complex spaces populated
conn
by connected skeletons. For connected sets A, B ⊂ K , we write A δ B. In effect,
conn
for each pair of skeletons A, B in K , A δ B, provided there is a path between
at least one vertex in A and one or more vertices in B. A path is sequence of edges
conn
between a pair of vertices. Equivalently, A ∩ B = ∅ implies A δ B. If the sets of
skeletons A, B ∈ K are separated (i.e., A, B have no vertices in common), we write
conn
A δ B. With this view of connectedness, the C̆ech axiom P4 is an emperor with
new clothes, namely,
conn
δ replaces δ in C̆ech axiom P4.
conn
P4conn A ∩ B = ∅ ⇒ A δ B.
conn
By replacing δ with δ in the remaining C̆ech axioms, we obtain a complete picture
of C̆ech connectedness proximity.
conn
δ C̆ech Connectedness Proximity.
Connectedness proximity axioms.
conn
P1conn A ∩ B = ∅ ⇔ A δ B, i.e., non-overlapping skeletons are not connected.
conn conn
P2conn A δ B =⇒ B δ A, i.e., A close to B implies B is close to A.
conn
Proof. A δ B, provided there is a path between any pair of vertices in skeletons
A and B, i.e., A, B are connected, provided there is a vertex common to A and B.
conn
That is, if the skeletons A, B have a common vertex, then A δ B (from Axiom
P4conn). Hence, A ∩ B = ∅.
conn
Smirnov property Q1 is satisfied. Hence, (K , δ ) is a proximity space.
conn
v6 ∈ vcyc A ∩ skelE = ∅ ⇔ vcyc A δ skelE.
Skeletons that are not close have no vertices in common. For example, in Fig. 5.4,
conn
skelE δ skelH,
in Fig. 5.7
cycH2 ∈ vNrvH ∈ vNrvB.
in Fig. 5.6
cycE 2 ∈ vNrvE 1 ∈ vNrvE.
in Fig. 5.6
cyc A2 ∈ vNrvA1 ∈ vNrvA.
For simplicity, the filled interiors of the 1-cycles in Fig. 5.6 are often hidden
(not shaded). Let the collection of vortex nerves K be equipped with the prox-
conn
imity δ . Vortex nerves are close, provided the nerves have nonempty intersection.
conn
For example, vNrvA δ vNrvE,
δ(vNrvA, vNrvE) = 0. Hence, Smirnov
i.e.,
conn
property Q2 is satisfied by K , δ . Vortex nerves are far (not close), provided
conn
the vortex nerves have empty intersection. For example, vNrvA δ vNrvE, i.e.,
δ(vNrvA, vNrvE) = 1 (Smirnov property Q3). We also have, for example,
Fig. 5.8 cycA overlaps cycle cycB in skeletal vortex nerve skNrvE
236 5 Surface Shapes and Their Proximities
Fig. 5.9 Cusp filaments skA, skB, skE on a connectedness proximity space
Problem 5.23 K
Show that each each of the axioms for a connectedness proximity space are satisfied
conn
by K equipped with δ in Example 5.22 are satisfied. “
Example 5.25 (Spacetime Vortex Nerves Proximity Space) Spacetime vortex nerves
(overlapping vortex cycles) have been observed in recent studies of ground vortex
5.6 Vortex Nerves Proximity Space 237
aerodynamics by Murphy and MacManus [21] and in the vortex flows of overlapping
jet streams in ground proximity by Barata, Bernardo, Santos and Silva [22] and by
Silva, Durão, Barata, Santos, Ribeiro [23]. Physical vortex nerves can be observed
in the representation of the contours of overlapping turbulence velocity vortices in,
for example, Fig. 6 in [23, p. 8] and systems of vortex in Fig. 7 in Spalart, Strelets,
Travin and Slur [24]. “
The presence of holes in the interiors of vortex nerves in a cell complex equipped
conn
with the proximity δ gives us the following result.
Proof. Immediate from Theorem 5.20, since the relationships between vortex nerves
in K are unaffected by the presence of holes in the interiors of the nerves.
Example 5.27 A pair of disjoint vortex nerves skNrvE, skNrvG containing skeletal
cycles with holes in their interiors is represented in Fig. 5.10. Let skNrvE, skNrvG
conn
both equipped with the proximity δ . From Corollary 5.26, the connectedness prox-
imities of these two nerves is unaffected by the presence of holes in the interiors of
the nerve cycles. Hence,
conn
Vortex nerves in Fig. 5.10 equipped with δ
conn
conn
skNrvE, δ , skNrvG, δ
Example 5.28 (Cycle Holes in a Vortex Nerve with Connectedness Proximities) This
is a continuation of Example 5.22. Let K be a collection of 1-cycles cycA, cycB
with holes H1 , H2 in a skeletal vortex nerve skNrvE equipped with the proximity
conn
δ , represented in the video frame in Fig. 5.8. Cycles cyc A, cycB both have filled
interiors. In this case, int(cyc A) ⊂ int(cycB) (i.e., the interior of cyc A is entirely
inside the interior of cyc A). For this reason, skA ∩ skB = ∅. From axiom P4conn,
we have
conn
skA ∩ skB = ∅ ⇔ skA δ skB.
Problem 5.29 Let K be a collection of vortex nerves so that the boundary of each
of the holes has more than one vertex that is in the intersection 1-cycles in each of the
nerves in a planar cell complex. For an example of vortex cycles that overlap vertices
on the boundary of a hole, see Fig. 5.11. Prove that a vortex nerve is destroyed by a
hole whose boundary overlaps the nerve cycles in more than one vertex.
Problem 5.30 Let K be a collection of vortex nerves so that the boundary of each
of the holes has a single vertex that is in the intersection of the 1-cycles in each of
conn
the nerves in a planar cell complex. Also let K be equipped the proximity δ . Prove
that K is a connectedness proximity space.
In this section, weak and strong connectedness proximities of skeletons arise when we
consider pairs of vortex cycles with overlapping interiors or pairs of cyclic skeletons
with a common edge. For strong connectedness between a pair of skeletons, the
notion of the interior of a skeleton needs to be understood. There are two cases
to consider, namely, interior of a 1-cycle (edge attached to a pair of vertices) and
interior (the surface region inside the boundary of a cyclic skeleton). The boundary
5.7 Strong [Overlap] Connectedness Proximity Space 239
of a cyclic skeleton sk cyclic E (denoted by bdy(sk cyclic E)) is defined by the sequence
of edges that enclose a bounded surface region.
Interior of a 1-cycle Let > pq be a 1-cycle (represented by p •——• q), defined by
a pair of vertices p,
q with edge —— attached between p and q. The interior of
>
pq is denoted by > pq is the edge between p and q.
Interior of a cyclic skeleton Let sk cyclic E be a cyclic skeleton. We also write skE,
when it is understood that skE is a cyclic skeleton. The interior of sk cyclic E is
denoted by int(sk cyclic E), which is the bounded surface region of bdy(sk cyclic E)
(boundary of the cyclic skeleton sk cyclic E).
We can begin considering the strong connectedness of cyclic skeletons on a trian-
gulated bounded surface region K by equipping the resulting cell complex K with
⩕
conn
the strong connectedness proximity δ , which consider informally in the following
example.
That is, since the pair of cyclic skeletons skA, skB in Fig. 5.12 have an edge in
common, that pair of cyclic skeletons is strongly connected, i.e., the interior of a
1-cycle is common to both skeletons. If this pair of cyclic skeletons had only a vertex
conn
in common, then we write skA δ skB. Notice that this pair skeletons also has a
conn
pair of vertices in common. So it is also the case that skA δ skB. “
⩕
conn
Proximity δ leads to new axioms wrt skeletal interiors.
⩕
conn
Let K be a collection of skeletons equipped with the proximity δ , which is a
⩕
form of the strong proximity δ [20, Sect. 1.9, pp. 28–30]. Also let skA, skB be a pair
⩕
conn
of skeletons in K . The weak and strong forms of δ satisfy the following axioms.
240 5 Surface Shapes and Their Proximities
⩕
conn
P4intConn [weak overlap axiom] int(skA) ∩ int(skB) = ∅ ⇒ skA δ skB.
⩕
conn
P5intConn [strong overlap axiom] skA δ skB ⇒ skA ∩ skB = ∅.
1-cells [edges], and 2-cells [filled triangles]) in every triangulated finite, bounded
planar region is a CW complex.
Let A, B, E ∈ K be sets of skeletons in a CW complex space K equipped with
⩕
conn
the proximity δ , which satisfies the following axioms.
⩕
conn
δ C̆ech Strong Connectedness Proximity.
Overlap Connectedness proximity axioms.
⩕
conn
P0intConn [empty set axiom] ∅ δ A, i.e., the empty set is not connected to
any skeleton A in K .
⩕
conn
P1intConn [disconnectedness axiom] A ∩ B = ∅ ⇔ A δ B, i.e., the skele-
tons A and B are not close ( A and B are far from each other).
⩕ ⩕
conn conn
P2intConn [commutativity axiom] A δ B =⇒ B δ A, i.e., A overlaps
(is close to) B implies B overlaps (is close to) A.
⩕ ⩕ ⩕
conn conn conn
P3intConn [union axiom] A δ (B ∪ C) =⇒ A δ B or A δ C.
⩕
conn
P4intConn [weak overlap axiom] intA ∩ intB = ∅ ⇒ A δ B.
⩕
conn
P5intConn [strong overlap axiom] A δ B ⇒ A ∩ B = ∅. “
⩕
conn
An overlap connectedness space is denoted by K, δ . Skeletons A, B in K
are close, provided the interior intA has nonempty intersection with the interior intA.
⩕
conn
δ , Closeness of Interiors Within Boundaries of Shapes.
⩕
K The proximity
conn
δ nudges us to take a closer look at the closeness of the
shape interiors. This is a very different view of the usual C̆ech view of the
closeness of shapes and of what is sometimes hidden or ignored inside the
boundaries of shapes themselves. “
Theorem 5.32 Let K be a collection of vortex nerves in a planar cell complex. The
⩕
conn
space K equipped with the relation δ is a proximity space.
Proof. The result follows from Lemma 5.16, since K is also a collection of skeletons
conn
equipped with the proximity δ .
242 5 Surface Shapes and Their Proximities
Example 5.33 (Overlapping Vortex Nerves) Two pairs of overlapping vortex nerves
⩕ ⩕
conn conn
are represented in Fig. 5.13, namely, vNrvA δ vNrvE and vNrvB δ vNrvH .
In the case of the pair of vortex nerves v Nr v A, v Nr v E, the gray region for these
nerves in Fig. 5.13 represents the nonempty intersection of the interior of the 1-cycle
intcyc A2 ∈ vNrvA and the interior of the 1-cycle intcycE 2 ∈ vNrvE. From axiom
P4intConn, we have
⩕
conn
intcyc A2 ∩ intcycE 2 = ∅ ⇒ cyc A2 δ cycE 2
⩕
conn
⇒ vNrvA δ vNrvE, (Ax. P5intConn)
⩕
conn
vNrvA δ vNrvE ⇒ intcyc A2 ∩ intcycE 2 = ∅.
Concentric vortex nerves vNrvB, vNrvH are also represented in Fig. 5.13, The inte-
rior I ntcycH2 is represented in Fig. 5.7 in the vortex nerve vNrvH , which is in the
interior of vortex nerve vNrvB. Again, from axiom P4intConn, we have (Table 5.2)
⩕
conn
intvNrv B ∩ int vNrv H = ∅ ⇒ vNrv B δ vNrv H,
& from Axiom P5intConn, we have
⩕
conn
vNrvB δ vNrvH ⇒ intvNrv B ∩ int vNrv H = ∅.
5.7 Strong [Overlap] Connectedness Proximity Space 243
accuracy of the descriptions of the objects. Every physical object has a description.
Physical objects of interest appear in triangulated video frames. Recall that each
triangulated video frame contains is a cell complex containing skeletons.
In this work, a description is a feature vector in Euclidean space Rn . In other
words, the description of an object is a point in Rn . Let K be cell complex and let
A be a skeleton in K . The ith feature of A is represented by a real-valued probe
function Φi : A −→ R (e.g., shape contour length Φi (A) = 30.5 mm). A feature
vector that describes A (denoted by Φ(A)) is defined by
Let A, B ∈ 2 K , the collection of all skeletons in cell complex K . Then Φ(K ) is the
set of feature vectors for all skeletons in K , i.e.,
Φ(K ) = {Φ(A) : A ∈ K } .
A δΦ B ⇔ Φ(A) ∩ Φ(B) = ∅.
Notice that Axiom PδΦ 3 is written in terms of ordinary set union ∪. This axiom
can be rewritten using one of the following four different forms of descriptive union,
5.9 Ahmad Descriptive Union 245
introduced by Ahmad in [29, Def. 5, p. 9]. The symbolic forms of Ahmad descriptive
unions are given in Table 5.3, briefly explained as follows.
• restrictive descriptive union: all the elements in A ∪ B are considered.
• non-restrictive descriptive union: only the elements in A ∩ B are considered.
• descriptive nondiscriminatory union: we consider element with any value of
description.
• descriptive discriminatory union: we consider elements with specific values of
description. “
Let A, B, C be nonempty sub-complexes in a cell complex K and let φ : 2 K → Rn
maps to an n-dimensional real-valued feature vector that describes set of cells in
cellular complex K . Then, for instance, we can then rewrite Axiom PδΦ 3 in the
following way.
K δΦ 3uni on : A δΦ B C ⇔ A δΦ B or A δΦ C.
Φ
For more about this, see Appendix A.1.
Problem 5.34 ®
Prove that Axiom K δΦ 3uni on and the original Axiom PδΦ 3 are equivalent. Hint: see
Appendix A.1. “
Description of conn(A)
Φ(conn(A)) = number of vertexes in conn(A).
5.10 Clusters of Sub-complexes 247
conn
skE in the interior of shape δ E
conn conn
skE ∈ int( δ A) and skE ∈ int( δ B) , such that
Φ(skE) = Φ(skE ), then
⩕
conn
skE δΦ skE .
Notice that the complexes K and K can either be different, spatially separated
complexes or K = K .
248 5 Surface Shapes and Their Proximities
Description of skE
Φ(skE) = number of vertexes in skE.
⩕
For this reason, the δΦ descriptive proximity is considered a strong proximity. Here
⩕
are the axioms for δΦ .
Definition 5.37 [20, Sect. 1.9, pp. 28–29] Let K be a cell complex, A, B, C ⊂ K
⩕
and x ∈ K . The relation δΦ on the collection of complexes 2 K is a Lodato strong
descriptive proximity, provided it satisfies the following axioms.
⩕
δΦ Lodato Strong Descriptive Proximity
δ ⩕
(dsnN0) ∅ ⩔Φ A, ∀A ⊂ X , and X δΦ A, ∀A ⊂ X .
⩕ ⩕
(dsnN1) A δΦ B ⇔ B δΦ A.
⩕
(dsnN2) A δΦ B ⇒ A ∩ B = ∅.
Φ
⩕
(dsnN3) If {Bi }i∈I is an arbitrary family of subsets of X and A δΦ Bi ∗ for some
⩕
i ∗ ∈ I such that int(Bi ∗ ) = ∅, then A δΦ ( i∈I Bi ).
⩕
(dsnN4) intA ∩ intB = ∅ ⇒ A δΦ B.
Φ
⩕
Overlapping Interiors in Lodato Strong Descriptive Proximity δΦ .
⩕
K For the strong [overlap] descriptive proximity δ Φ
, we require more machin-
⩕
ery than we did for the C̆ech proximity δΦ . With δΦ , we take into account the
requirements for strong descriptive proximity in the case where the interiors
of a pair of disjoint complexes overlap descriptively, i.e., the description of the
interior of one complex matches the interior another complex. Notice that such
complexes such as visual scenes recorded in video frames, can be separated in
⩕
spacetime. With the Lodato δΦ , we explicitly treat the case where the description
of interiors overlap. “
250 5 Surface Shapes and Their Proximities
⩕
When we write A δΦ B, we read A is descriptively strongly near B. The notation
δ
A ⩔Φ B reads A is not descriptively strongly near B. For each descriptive strong
proximity, we assume the following relations:
⩕
(dsnN5) Φ(x) ∈ Φ(int(A)) ⇒ x δΦ A.
⩕
(dsnN6) {x} δΦ {y} ⇔ Φ(x) = Φ(y).
Example 5.38 (Descriptively close shapes) The pair of shapes shown in Fig. 5.16
are descriptively close, provided this pair of shapes has matching descriptions. “
Example 5.39 (Strong Descriptive closeness of separated shapes) The pair of shapes
shown in Fig. 5.17 are descriptively close, provided this pair of shapes has matching
descriptions. “
⩕
Fig. 5.17 E ∈ K δΦ int(NrvB) ∈ K
In other words, even though each of these skeletons is on a different cell complex,
they are descriptively close, i.e., the angle between each pair of vertexes p, q on skA
equals the angle of each pair of vertexes p , q on skB. “
252 5 Surface Shapes and Their Proximities
The situation with the strong descriptive proximity is quite different from the
⩕
δΦ proximity. That is, δΦ is used to detect the closeness of a skeleton skA on a
complex K to another skeleton skB that resides in the interior of a shape shE (i.e.,
skB ∈ int(shE)) on a complex K . Complexes K , K can either be the same or
different.
In other words, this pair of skeletons has strong descriptive closeness, even though
one of these skeletons resides in the interior of a shape on the cell complex K , since
the angle between each pair of vertexes p, q on skA equals the angle of each pair of
vertexes p , q on skB. “
5.11 Descriptive Connectedness Proximity 253
In effect, this is a repeated application of Axiom Pconn 4 in the search for path-
δΦ
connected sub-complexes skB on complex K with descriptions that match the
254 5 Surface Shapes and Their Proximities
Du
ρ = −∇
P +
Dt
Net pressure force
Acceleration of a fluid particle
μΔu.
Net viscous force
Notice that every vortex field can be represented by vortex nerve sk cyclic NrvE, which is
a collection of vortex lines with a common center vortex vector. This line of reasoning
conn
leads to a straightforward application of the descriptive connected proximity δΦ in
finding clusters of vortexes that are descriptively close to a particular vortex.
Problem 5.42 K
Vorticity dynamics in describing a cusp filament .
The basic approach in this problem is to view the vertex vector q at the tail end of a
cusp filament in an optical vortex nerve as a photon in motion over a sequence of tri-
angulated video frames. The vortex
vector q is on boundary of an optical
vortex nerve
sk cyclic NrvE (denoted by bdy sk cyclic NrvE ). The boundary bdy sk cyclicNrvE is an
example of a connected vortex line, i.e., the set of vertex vectors on bdy sk cyclic NrvE
are path-connected.
This means that there is a path between any pair of vortex vectors
on bdy sk cyclic NrvE . Notice that Using the Lagrangian description of the accelera-
tion of a fluid particle in a fluid vortex in Cottet and Koumoutsakos [30, Sect. 1.2,
pp. 5-7], do the following:
1o Select a sequence of video frames V in a video.
2o Triangulate each of the video frames in V .
3o Identify an optical vortex nerve in each of the video frames in V .
4o Let each cusp filament filamentE represent the path of reflected light in the
cusp of a light caustic, which is represented in each sk cyclic NrvE in video frame
in V . And let p, q be the end vertexes in filamentE. Let ρ be the average
wavelength of the hues on filamentE. Introduce a mathematical representation
of the average rate of change of the acceleration ρ Dq
Dt
of the vortex vector q (tail
of the cusp filament), i.e., the rate of change of the expansion (or contraction)
of the length of the maximal cusp filament filamentE in a succession of video
frames.
5o Select a threshold th > 0 for the difference between descriptions of cusp fila-
ments.
Give an example of a pair ofvideo frames in V in which the description
Φ(sk cyclic NrvE) − Φ(sk cyclic NrvE) < th.
6o Repeat Step 5.42 for a pair of videos. “
256 5 Surface Shapes and Their Proximities
Problem 5.43 K
Optical vorticity in describing optical vortex nerves .
An optical vortex nerve on a CW complex on a triangulated bounded planar surface
region is a representation of a very simple optical vortex. Using features of optical
vortexes in Nye [32, Sect. 1, p. 2–3] and Dudley, Dias, Erkintalo and Gentry [33],
do the following:
1o Select a sequence of video frames V in a video.
2o Identify an optical vortex nerve in each of the video frames in V .
3o Let each cusp filament filamentE represent the path of reflected light in the
cusp of a light caustic, which is represented in each sk cyclic NrvE in video frame
in V . And let p, q be the end vertexes in filamentE. Let ρ be the average
wavelength of the hues on filamentE. Introduce a mathematical representa-
tion of the average
rate of change of the optical wavefield represented by the
boundary bdy sk cyclic NrvE of sk cyclic NrvE in video frame in V . Recall that an
optical vortex nerve straddles a maximal barycentric Alexandroff nerve com-
plex (MNC) on a triangulated set of centroids of surface holes. That is, each
vertex in sk cyclic NrvE is a barycenter of a triangle either on an MNC or on a
triangle on the boundary of the MNC. This means that bdy sk cyclic NrvE rep-
resents a path for reflected light (between surface holes) (Figs. 5.20 and 5.21).
The focus here is on representing the rate of change of bdy sk cyclic NrvE in a
succession of video frames.
4o Select a threshold th > 0 for the difference between descriptions of cusp fila-
ments.
Give an example
ofa pair of video
frames in V in which the description
bdy sk cyclic NrvE − bdy sk cyclic NrvE < th.
5o Repeat Step 1 for a pair of videos. Hint: Try using the representation of a
monochromatic optical wavefield in Nye [32, Sect. 1, p. 2] in terms of a plane
wave ψ, instead of considering multiple plane waves given in Nye’s represen-
tation of a wave field. “
The descriptive closure of A (denoted by clΦ A) [20, Sect. 1.4, p. 16] is defined
by
⎧ ⎫
⩕
⎨ conn ⎬
clΦ A = x ∈ K : x δΦ A .
⎩ ⎭
⩕
conn
The weak and strong forms of δΦ satisfy the following axioms.
⩕
conn
PΦ 4 [weak option] intA ∩ intB = ∅ ⇒ A δΦ B.
Φ
⩕
conn
PΦ 5 [option] A δΦ B ⇒ A ∩ B = ∅.
Φ
“
5.12 Strong Descriptive Connectedness Proximity 259
⎛ ⩕
⎞
conn
A descriptive overlap connectedness space is denoted by ⎝ K , δΦ ⎠. Skeletons
A, B in K are close descriptively, provided the interior int A has nonempty descriptive
intersection with the interior int A. This form of proximity has many applications,
since we often want to compare objects such as 1-cycles by themselves or vortex
cycles or the more complex vortex nerves that do not overlap spatially or at the same
time.
Example 5.44 (Descriptive Connectedness Overlap of Disjoint Vortex Cycles in
Spacetime) Let vcyc A, vcycB be a pair of vortex cycles in a collection of vortex
⩕ ⩕
conn conn
cycles equipped with the proximities δ and δΦ . Assume these vortices represent
non-overlapping electromagnetic vortexes that have matching descriptions in space-
time, e.g., Φ(vcyc A) = Φ(vcycB) = (persistence duration). That is, the length of
⩕
conn
time that vcyc A persists equals the duration of vcycB. In that case, vcyc A δΦ
vcycB. “
Example 5.45 (Descriptive Connectedness Overlap of Cell Complexes) The bar
graph2 in Fig. 5.22 compares feature values for a pair of cell complexes, namely,
vertex count, hole count, maximum vortex cycle area, nerve cycle count and nerve
⩕
conn
count. From the bar graph, K 1 δΦ K 2 , since
This is the case, even though the hole count and nerve cycle count are far apart. “
2 Many thanks to M.Z. Ahmad for the LATEX script used to display this bar graph, which does not
depend on an external file.
260 5 Surface Shapes and Their Proximities
Proof. The result follows from Lemma 5.16, since each vortex cycle in K is also a
⩕
conn
collection of skeletons equipped with the proximity δΦ .
Corollary 5.48 Let K be a collection of vortex nerves in a planar cell complex. The
⩕
conn
space K equipped with the relation δΦ is a proximity space.
Proof. The result follows from Theorem 5.47, since each vortex nerve in K is also
⩕
conn
a collection of intersecting vortex cycles equipped with the proximity δΦ .
5.12 Strong Descriptive Connectedness Proximity 261
conn
cyc A1 δ cycB2 (Cycles cyc A1 , cycB2 do not overlap) ,
⩕
conn
cyc A1 δΦ cycB2 since Φ(cyc A1 ) = Φ(cycB2 ) = (6). “
⩕
conn
δΦ -based Zero Shot Recognition.
Descriptive connectedness, in its various forms, provides a basis for a form of
zero-shot recognition explored by Lu, Li, Yan and Zhang [36]. Zero-shot recog-
nition identifies an unseen class that each new image belongs to. Let K be a cell
complex on a finite bounded region of a triangulated image, optical vortex nerve
⩕
conn
sk cyclic NrvE ⊂ K , equipped with strong descriptive connectedness proximity δΦ .
And let Φ(sk cyclic Ai ), 1 ≤ i ≤ k be a description of sk cyclic Ai , a skeletal cycle in
sk cyclic NrvE. Using Φ(sk cyclic Ai ), we can separate the optical vortex nerve regions
in each member of a collection of triangulated images into classes based on the
description of the vortex cycle interiors in each of the nerves.
Application: Strong Descriptive Connectedness-Based Zero Shot Recogni-
tion.
K Zero-shot classification of images with no training data is highly attrac-
tive, since it is less rigid than traditional classification techniques that rely on
training data and, hence, build into the learning process unwanted à priori
assumptions implicit in the training data. For more about this, see Molina and
Sánchez [37] and an introduction to low-shot visual recognition by Hariharan
and Girshick [38]. Unlike zero-shot recognition, low-shot visual recognition
tunes the learning process based on a few sample examples per class. Then each
new visual scene A is classified based on the closest similarity of A to a sample
in one of class examples. In our case, this would entail starting a visual scene
recognition process with sample optical vortex nerves (derived from triangu-
lated visual scenes) that fit into one of the classes of nerves that we consider
interesting. “
Example 5.51
⎛ (Proximal
⎧ Relator
⎫⎞ Space) Example 5.49 introduces a proximal rela-
⩕
⎨conn conn ⎬
tor space ⎝ K vNrv , δ , δΦ ⎠, useful in measuring, comparing, and classifying
⎩ ⎭
collections of vortex nerves that either have or do not have
⎛ matching
⎧ descriptions.
⎫⎞
⩕
⎨conn conn ⎬
Similarly, Example 5.50 introduces a proximal relator ⎝ K cyc , δ , δΦ ⎠, use-
⎩ ⎭
ful in the study of collections of 1-cycles that either have or do not have matching
descriptions. “
⩕
The connection between δ and δ is summarized in Lemma 5.52.
⎛ ⎧ ⩕ ⎫⎞
⎨conn conn
⩕
conn
⎬
Lemma 5.52 Let ⎝ K , δΦ , δ , δ ⎠ be a proximal relator space K , A, B ⊂
⎩ ⎭
K . Then
⩕
conn conn
1o
A δ B⇒A δ B.
⩕ ⩕
conn conn
2o A δ B ⇒ A δΦ B.
⩕
conn
Proof. 1o : From Axiom P5conn, A δ B implies A ∩ B = ∅, which implies
conn conn
A δ B. From Lemma 5.15, A δ B implies A ∩ B = ∅, which implies A δ B
(from C̆ech Axiom P4).
2o : From (1), there are cyc x ∈ A, cyc y ∈ B common to A and B. Hence, Φ(cyc x) =
Φ(cyc y), which implies A ∩ B = ∅. Then, from the descriptive connectedness
Φ
⩕
conn
Axiom PΦ 4conn, A ∩ B = ∅ ⇒ A δΦ B. This gives the desired result.
Φ
3 Many thanks to Ron Enns for this picture of an Ontario, Canada maple tree, captured with a cell
phone during the Thanksgiving holiday in October, 2018.
264 5 Surface Shapes and Their Proximities
⩕
conn
Fig. 5.24 skA overlaps skeleton skB and skA δΦ skB
conn
From δ Axiom P4conn, Sect. 5.5
conn
int(MNCA) ∩ int(MNCB) = ∅ ⇒ MNCA δ MNCB.
⩕
conn
From δ Axiom P4overlap, Sect. 5.7
⩕
conn
int(MNCA) ∩ int(MNCB) = ∅ ⇔ MNC δ MNCB.
5.14 Vortex Cycle Spaces Equipped with Proximal Relators 265
Proximities up to 1970:
Naimpally and Warrack, [2], a good overview of proximity space theory up to
1970.
Beer, Di Concilio, Di Maio, Naimpally, Pareek and Peters [1] on Som Naimpally’s
life and work.
Proximities since 1970:
Di Concilio [8] gives an excellent overview of proximities and their utility in
extension theory, functions spaces, hyperspaces, boolean algebras and point-free
geometry.
Local Proximity Spaces:
Guadagni [19], on recent research on local proximity spaces.
Leader [41], a seminal work on local proximity.
Descriptive proximity:
Di Concilio, Guadagni, Peters and Ramanna [34], a fairly comprehensive view of
both the theory and application of descriptive proximity.
Proximal relators:
Peters [42], an introduction to relators that are collections of proximities.
Computational proximity:
Peters [43], an introduction to computational proximity that combines various
proximites and algorithms useful in implementing proximities.
Shapes, Optical Vortex Nerve Structures and their Proximities:
Peters [44] introduces proximal vortex cycles and vortex nerve structures, includ-
ing the introduction to non-concentric, nesting, possibly overlapping homology
cell complexes. This paper was inspired by recent work by Worsley and Peters [45]
on electron magnetic moment anomaly from the electron charge using geometric
principles.
References
1. Beer, G., Di Concilio, A., Di Maio, G., Naimpally, S., Pareek, C., Peters, J.: Somashekhar
Naimpally, 1931–2014. Topol. Its Appl. 188, 97–109 (2015). https://doi.org/10.1016/j.topol.
2015.03.010, MR3339114
2. Naimpally, S., Warrack, B.: Proximity Spaces. Cambridge Tract in Mathematics, vol. 59. Cam-
bridge University Press, Cambridge (1970). X+128 pp. Paperback (2008), MR0278261
3. Riesz, F.: Stetigkeitsbegriff und abstrakte mengenlehre. Atti del IV Congresso Internazionale
dei Matematici II, 182–109 (1908)
4. Naimpally, S.: Near and far. A centennial tribute to Frigyes Riesz. Sib. Electron. Math. Rep. 2,
144–153 (2009)
5. Naimpally, S.: Proximity Approach to Problems in Topology and Analysis. Oldenbourg Verlag,
Munich, Germany (2009). 73 pp. ISBN 978-3-486-58917-7, MR2526304
6. Concilio, A.D.: Proximal set-open topologies on partial maps. Acta Math. Hungar. 88(3), 227–
237 (2000). MR1767801
7. Concilio, A.D.: Topologizing homeomorphism groups of rim-compact spaces. Topol. Its Appl.
153(11), 1867–1885 (2006)
8. Concilio, A.D.: Proximity: A powerful tool in extension theory, functions spaces, hyperspaces,
Boolean algebras and point-free geometry. In: Mynard, F. Pearl, E. (eds.) Beyond Topology.
AMS Contemporary Mathematics 486, pp. 89–114. American Mathematical Society (2009)
9. C̆ech, E.: Topological Spaces. Wiley Ltd., London (1966); Fr seminar, Brno, 1936–1939; rev.
ed. Z. Frolik, M. Katĕtov
10. Efremovič, V.: The geometry of proximity I (in Russian). Mat. Sb. (N.S.) 31(73)(1), 189–200
(1952)
11. Naimpally, S.: Proximity Spaces. Cambridge University Press, Cambridge (1970). X+128 pp.
ISBN 978-0-521-09183-1
12. Peters, J., Ramanna, S.: Pattern discovery with local near sets. In: R. Alarcón, P. Barceló (eds.)
Proceedings of Jornadas Chilenas de Computación 2012 workshop on pattern recognition, pp.
1–4. The Chilean Computing Society, Valparaiso (2012)
13. Peters, J., Naimpally, S.: Applications of near sets. Notices of the Am. Math. Soc. 59(4),
536–542 (2012). https://doi.org/10.1090/noti817, MR2951956
14. Naimpally, S., Peters, J.: Topology with Applications. Topological Spaces via Near and Far.
World Scientific, Singapore (2013). Xv+277 pp. Am. Math. Soc. MR3075111
15. Di Maio, G., S.A. Naimpally, E.: Theory and Applications of Proximity, Nearness and Unifor-
mity. Seconda Università di Napoli, Napoli, Italy (2009). 264 pp. MR1269778
268 5 Surface Shapes and Their Proximities
16. Naimpally, S., Peters, J., Wolski, M.: Foreword [near set theory and applications]. Math. Com-
put. Sci. 7(1), 1–2 (2013)
17. Smirnov, J.M.: On proximity spaces. Math. Sb. (N.S.) 31(73), 543–574 (1952). English trans-
lation: Am. Math. Soc. Trans. Ser. 2, 38, 1964, 5-35
18. Willard, S.: General Topology. Dover Publications, Inc., Mineola (1970). Xii+369 pp. ISBN:0-
486-43479-6 54-02, MR0264581
19. Guadagni, C.: Bornological convergences on local proximity spaces and ωμ -metric spaces.
Ph.D. thesis, Università degli Studi di Salerno, Salerno, Italy (2015). Supervisor: A. Di Concilio,
79pp
20. Peters, J.: Computational Proximity. Excursions in the Topology of Digital Images. Intelligent
Systems Reference Library, vol. 102 (2016). Xxviii+433 pp. https://doi.org/10.1007/978-3-
319-30262-1, MR3727129 and Zbl 1382.68008
21. Murphy, J., MacManus, D.: Ground vortex aerodynamics under crosswind conditions. Exper.
Fluids 50(1), 109–124 (2011)
22. Barata, J., N. Bernardo, P.S., Silva, A.: Experimental study of a ground vortex: the effect of
the crossflow velocity. In: 49th AIAA Aerospace Sciences Meeting, pp. 1–9. AIAA (2011)
23. Silva, A., ao, D.D., Barata, J., Santos, P., Ribeiro, S.: Laser-doppler analysis of the separation
zone of a ground vortex flow. In: 14th Symposium on Applications of Laser Techniques to
Fluid Mechanics, Lisbon, Portugal, pp. 7–10. Universidade Beira Interior (2008)
24. P.R. Spalart M. Kh. Strelets, A.T., Slur, M.: Modeling the interaction of a vortex pair with the
ground. Fluid Dyn. 36(6), 899–908 (1999)
25. Dzedolik, I.: Vortex properties of a photon flux in a dielectic waveguide. Tech. Phys. 75(1),
137–140 (2005)
26. van Leunen, H.: The hilbert book model project. Technical report, Deparment of Applied
Physics, Technische Universiteit Eindhoven (2018). https://www.researchgate.net/project/
The-Hilbert-Book-Model-Project
27. Litchinitser, N.: Structured light meets structured matter. Sci. New Ser. 337(6098), 1054–1055
(2012)
28. Hance, M.: Algebraic structures on nearness approximation spaces. Ph.D. thesis, University
of Pennsylvania, Department of Physics and Astronomy (2015). Supervisor: H.H. Williams,
vii+113 pp
29. Ahmad, M., Peters, J.: Descriptive unions. A fibre bundle characterization of the union of
descriptively near sets. 1–19 (2018). arXiv:1811.11129v1
30. Cottet, G.H., Koumoutsakos, P.: Vortex methods. Theory and Practice. Cambridge University
Press, Cambridge (2000). xiv+313 pp. ISBN:0-521-62186-0, MR1755095
31. Tian, S., Gao, Y., Dong, X., Liu, C.: A definition of vortex vector and vortex. 1–26 (2017).
arXiv:1712.03887
32. Nye, J.: Events in fields of optical vortices: rings and reconnection. J. Opt. 18, 1–11 (2016).
https://doi.org/10.1088/2040-8978/18/10/105602
33. Dudley, J., Dias, F., Erkintalo, M., Gentry, G.: Instabilities, breathers and rogue waves in optics.
Nat. Photon. 8, 755–764 (2014). www.nature.com/naturephotonics
34. Concilio, A.D., Guadagni, C., Peters, J., Ramanna, S.: Descriptive proximities. properties and
interplay between classical proximities and overlap. Math. Comput. Sci. 12(1), 91–106 (2018).
MR3767897, Zbl 06972895
35. Peters, J.: Local near sets: pattern discovery in proximity spaces. Math. Comput. Sci. 7(1),
87–106 (2013). https://doi.org/10.1007/s11786-013-0143-z, MR3043920, ZBL06156991
36. Lu, J., Li, J., Yan, Z., Zhang, C.: Zero-shot learning by generating pseudo feature representa-
tions. 1–18 (2017). arXiv:1703.06389v1
37. Molina, M., Sánchez, J.: Zero-shot learning with partial attributes. In: Brito-Loeza,
C., Espinosa-Romero, A. (eds.) Intelligent Computing Systems. ISICS 2018. Communications
in Computer and Information Science, vol. 820, pp. 147–158. Springer Nature, Switzerland
AG (2018). https://doi-org.uml.idm.oclc.org/10.1007/978-3-319-76261-6_12
38. Hariharan, B., Girshick, R.: Low-shot visual recognition by shrinking and hallucinating fea-
tures. 1–10 (2016). arXiv:1606.02819v4
References 269
39. Peters, J.: Proximal relator spaces. Filomat 30(2), 469–472 (2016). https://doi.org/10.2298/
FIL1602469P
40. Száz, A.: Basic tools and mild continuities in relator spaces. Acta Math. Hungar. 50(3–4),
177–201 (1987). MR0918156
41. Leader, S.: Local proximity spaces. Math. Ann. 169, 275–281 (1967)
42. Peters, J.: Proximal relator spaces. Filomat 30(2), 469–472 (2016). MR3497927
43. Peters, J.: Computational Proximity. Excursions in the Topology of Digital Images. Intelligent
Systems Reference Library, vol. 102. Springer, Berlin (2016). viii+445 pp. https://doi.org/10.
1007/978-3-319-30262-1
44. Peters, J.: Proximal vortex cycles and vortex nerve structures. non-concentric, nesting, possibly
overlapping homology cell complexes. J. Math. Sci. Model. 1(2), 56–72 (2018). ISSN 2636-
8692, www.dergipark.gov.tr/jmsm, See, also, https://arxiv.org/abs/1805.03998
45. Worsley, A., Peters, J.: Enhanced derivation of the electron magnetic moment anomaly from
the electron charge using geometric principles. Appl. Phys. Res. 10(6), 1–14 (2018). http://apr.
ccsenet.org
46. Dennis, M., King, R., Jack, B., Holleran, K., Padgett, M.: Isolated optical vortex knots. Nat.
Phys. 6(2), 118–121 (2010). https://doi.org/10.1103/PhysRevD.81.066004
47. Pike, O., Mackenroth, F., Hill, E., Rose, S.: A photon-photon collider in a vacuum hohlraum.
Nat. Photon. 8(6), 434–436 (2014). https://doi.org/10.1038/nphoton.2014.95
48. Worsley, A.: The formulation of harmonic quintessence and a fundamental energy equivalence
equation. Phys. Essays 23(2), 311–319 (2010). https://doi.org/10.4006/1.3392799
49. Worsley, A.: Harmonic quintessence and the derivation of the charge and mass of the electron
and the proton and quark masses. Phys. Essays 24(2), 240–253 (2011). https://doi.org/10.4006/
1.3567418
50. Worsley, A.: The formulation of harmonic quintessence and a fundamental energy equivalence
equation. Phys. Essays 23(2), 311–319 (2010). https://doi.org/10.4006/1.3392799, ISSN 0836-
1398
51. Nye, J.: Natural Focusing and Fine Structure of Light. Caustics and Dislocations. Institute of
Physics Publishing, Bristol (1999). xii+328 pp. MR1684422
Chapter 6
Leader Clusters and Shape Classes
Abstract This chapter introduces a number of basic types of shape classes com-
monly found in CW complexes. These shape classes are useful in clustering and
separating subcomplexes in triangulated finite, bounded surface regions such as
those found in visual scenes. Spatial shape classes derived from spatial proximi-
⩕
⩕ conn conn
ties δ, δ , δ , δ are examples of what Leader [1] called clusters. A Leader cluster
is a collection of near sets, derived from a given member A of a proximity space X ,
by finding all subsets E of X that are near A. Each spatial shape class is a Leader
cluster. Four types of spatial shape classes are considered in this chapter.
6.1 Introduction
A spatial shape class is a collection of shapes E that are spatially near a particular
⩕
shape A by considering, for example, the strong spatial proximity E δ A for a given
⩕ ⩕
δ
shape A in a proximity space X equipped with δ (denoted by clsshape (A)), defined
by
⩕
δ
Strongly near shape class clsshape
⩕ ⩕
δ
clsshape (A) = E⊂X:E δA .
⩕
δ
Examples of instances of clsshape (sk cyclic NrvA), can be found, for example, in tri-
angulated video frames containing optical vortex nerves sk cyclic NrvA in a cellular
⩕
⩕ conn conn
complex K equipped with the proximities δ, δ , δ , δ . This chapter also intro-
clsδshape
Φ
(A) = {E ⊂ X : E δΦ A} .
This section briefly revisits what it means to say that one set is descriptively close to
another set. Let X be a nonempty set and let 2 X denote the collection of subsets in
the set X . In the case where E is a nonempty set in the space X , we can also write 2 E
to denote the collection of nonempty subsets of the set E. Descriptive proximity is
defined in terms of feature vectors in an n-dimensional space of real-valued vectors
(denoted by Rn ). Let x1 , . . . , xi , . . . , xn be n real numbers. Each point p in Rn has
the form
vector in Rn
p = (x1 , . . . , xi , . . . , xn ) ∈ Rn .
members of subset A
A = {A1 , . . . , Ai , . . . , An } .
feature vector that describes subset A
Φ(A) = (Φ(A1 ), . . . , Φ(Ai ), . . . , Φ(An )) ∈ Rn .
feature vectors that describes space X
Φ(X ) = A ∈ 2 X : Φ(A) ∈ Rn .
In other words, A ∩ E is the collection of all members of A and E that have the
Φ
same description. The descriptive proximity relation is denoted by δΦ .
The expression A δΦ E reads A is descriptively near E, i.e., Φ(A) (description
of A) matches Φ(E) (description of E). In practice, a weakened form of descriptive
proximity is used. Let th > 0 be a threshold on a descriptive proximity. Then we
have
274 6 Leader Clusters and Shape Classes
feature vector
Φcolour,sidesCount (x) ∈ R × R = (colour, sidesCount).
for rooftop x in X .
6.2 Descriptive Closeness Revisited 275
Then we have
For the first example, rooftops x1 and x2 have the same colour, namely, coffee-
coloured. For the last example, rooftops x3 , x5 have the same colour and the same
number of sides as rooftop x9 . “
feature vector
Φcolour,sidesCount (r t) ∈ R × R = (colour, sidesCount).
for rooftop r t in X ∪ Y , the complete collection of the rooftops in Fig. 6.2a and b.
Then we have
In other words, even though the rooftops x3 , x5 , x9 in Fig. 6.2a and rooftop y in
Fig. 6.2b are spatially separated, they are descriptively close to each other. That is,
rooftops x3 , x5 , x9 and rooftop y have the same colour and the contour of these
rooftops has the same number of sides. “
276 6 Leader Clusters and Shape Classes
This gives the following result. For simplicity, let Φ := Φcolour,sidesCount . Then we
have
x1 ∩ y ∈ Φ(X ∪ Y ) = {Φ(x) : x ∈ X ∪ Y } .
Φ
Recall from complex Observation No. 1.27 in Sect. 1.24, the two Alexandroff–
Hopf requirements for a CW complex, namely,
TK(1) Alexandroff–Hopf Cell Complex Containment Condition. Each cell on
any cell in a complex K is also in K .
TK(2) Alexandroff–Hopf Cell Complex Intersection Condition. The intersec-
tion of two closed cells in K is a closed cell on both of them.
From we have observed in Observation No. 6.2, the sample descriptive proximity
satisfies the two Alexandroff–Hopf requirements for a CW complex. It is a straight-
forward task to verify that each of the descriptive proximities also satisfy the two
Alexandroff–Hopf requirements for a CW complex. For this reason, we conclude
that Φcolour,sidesCount (X ∪ Y ) in Example 6.2 is a form of CW complex. Hence,
Φcolour,sidesCount (X ∪ Y ) is an example of a what is known as a descriptive CW
6.2 Descriptive Closeness Revisited 277
Problem 6.3 Give the containment and intersection conditions for a descriptive CW
complex. “
2o For your example, show that the containment and intersection conditions for a
descriptive CW complex are satisfied. “
This section briefly looks at the angle between vectors representing the vertexes on a
cusp filament. Recall that a vector space is a collection of objects that can be added
together or multiplied by numbers. A highly recommended, detailed introduction
to vector spaces appears in Gellert, Küstner, Hellwich and Kästner [2, Sect. 17.3,
starting on p. 362]. Here the objects are 2-tuples that are the coordinates of points
in the Euclidean plane. For example, in the Euclidean plane, a cusp vertex is an
example of a vector p, represented by (x, y), the horizontal and vertical coordinates
of p. Interest in the angle between cusp filament vectors stems from a view of cusp
filaments in the pathways for light reflected from visual scene surfaces. For more
about cusp filaments as pathways for reflected light between surface shape holes
on cell complexes derived from the barycenters of triangles that are found in the
triangulation of seed points that are centroids of surface holes, see Sect. 4.12.
278 6 Leader Clusters and Shape Classes
dot product
p · q = (x, y) · (x
, y
) = x x
+ yy
).
The norm (specifically, the L 2 norm) of the vector p (denoted by p) is defined
by
norm of vector p
p = p2 = x 2 + y2.
norm of vector q
q = q2 = x
2 + y
2 .
Example 6.6 (Sample angle between cusp filament vectors) The angle θ between
cusp filament vectors p and q is represented in Fig. 6.3. In this case, we have
p · q = 3 ∗ 5 + 5 ∗ 3 = 30.
p × q = 5.83095 × 5.83095 = 34.
angle θ between cusp filament vectors p and q
p · q
θ= cos −1
p × q
30
= cos −1
34
6.3 Angle Between Cusp Filament Vectors 279
= 0.489957 (radians).
= 28.0725o (degrees).
Recall from Observation No. 4.11 in Sect. 4.11 that a cusp filament is a 1-cell, which
is an edge attached between a barycenter on a cyclic skeleton sk cyclic E 0 and the closest
barycenter on the cyclic skeleton sk cyclic E 1 found along the boundary of a maximal
Alexandroff nerve complex NrvE. Put another way, we attach an edge between a
barycenter b of a triangle A on NrvE and to the barycenter b
of the triangle A
Cusp filaments are important not only because they are analogues of light cusps
in visual scenes captured by a camera but also because cusp filaments are among
the simplest structures in characterizing complex structures such as optical vortex
nerves. In this section, we focus on the angle between cusp filament vertexes.
280 6 Leader Clusters and Shape Classes
This section briefly introduces a sample δΦ -based shape class (denoted by clsδshape
Φ
).
For example, let A, E be a pair of dice and let Φ(A) be the colour of die A.
Similarly, let Φ(E) be the colour of die E. Then A ∩ E is the collection of dice E
Φ
that have the same colour as die A.
Two nonempty sets are descriptively near if and only if their descriptive inter-
section is non-empty or, equivalently, if and only if their descriptions intersect. The
introduction of descriptive intersection led to new forms of proximity (see, e.g., [3,
Sect. 3, p. 90], [4, Sect. 3]). Let X be a non-empty set, A and B nonempty subsets
of X, and let Φ : X → Rn be a probe. Then
Example 6.9 (Two δΦ -Based Optical Nerve Classes on Video Frames) In this exam-
ple, we assume that each optical vortex nerve consists of a pair of nesting cyclic
filament skeletons attached to each other by cusp filaments so that each cusp fila-
ment is the analogue of a cusp in a coffee cup caustics (for the details about coffee
cup caustics, see Sect. 4.11). Recall from Theorem 4.26 that the Betti number of an
optical vortex nerve defined by a pair of nesting cyclic filament skeletons attached
to each other by k cusp filaments, is k + 2.
Let Φ(sk cyclic NrvA) equal the Betti number of the free Abelian group G represen-
tation of the optical vortex nerve sk cyclic NrvA. Let B(G) denote the Betti number of
the free Abelian group G. Recall from Sect. 4.11, Observation No. 4.11 that a cusp
filament is an edge (1-cycle) connected between a pair of nesting cyclic skeletons
in an optical vortex nerve. Also, recall that the Betti number of a free Abelian group
representation of an optical vortex nerve is a count of the cusp filaments in the nerve
+ 2. Two δΦ -based optical nerve classes are represented in the video frames shown
in Fig. 6.1, namely,
δΦ -based class, Betti number = 9 + 2
Optical vortex nerves with Betti number = 11.
282 6 Leader Clusters and Shape Classes
That is, each optical vortex nerve in this classes has 9 cusp filaments (represented
by •——•), leading to an abelian free group representation with a Betti number
equal to 9 + 2.
4 members of class
clsδshape
Φ
(sk cyclic NrvE 2 ) = {NrvE 2 , NrvE 3 , NrvE 4 , NrvE 5 }
2 other members of this class
∪ {NrvE 6 , NrvE 7 }
That is, each optical vortex nerve in this descriptive proximity class has 8 cusp
filaments (again represented by •——•), leading to an abelian free group repre-
sentation with a Betti number equal to 8 + 2. “
6.5 Descriptive Proximity-Based Shape Class 283
Other forms of descriptive shape classes are derived using the proximities
⩕
⩕ conn conn
δΦ , δΦ , δΦ .
⩕
In this section, we consider an approach to the construction of δΦ -based classes,
which pigeonhole shapes into classes based on the affinities of the interiors of the
⩕
⩕ δ
shapes. In general, a class of δΦ -based shapes (denoted by clsshape
Φ
) is a collection
of shapes with interiors that have matching features. The focus in this work is on
⩕
strong descriptive proximity shape δΦ class construction, which reduces to checking
whether the features of the interior of a particular object shape match the features of
the interior of a representative of a known class of shapes. The importance of shape
⩕
interiors is pinpointed by the δΦ proximity.
284 6 Leader Clusters and Shape Classes
In this section, the steps to construct a class of maximal Alexandroff nerve shapes
found in triangulated video frame shapes, are given in Algorithm 12. This class of
shapes is important, since it isolates those maximal nucleus cluster (MNC) shapes
that have interiors with matching descriptions. This form of Alexandroff nerve shape
underlies a number of forms of nerves such as optical vortex nerve shapes and optical
cusp nerve shapes.
⩕
When we write A δΦ B, we read A is snd B, i.e., A and B are strongly near,
⩕
descriptively. The notation A
δ B reads A is not strongly near B, descriptively. For
Φ
each strong descriptive proximity, we make two additional assumptions.
⩕
(sndN5) Φ(x) ∈ Φ(int(A)) ⇒ x δΦ A
⩕
(sndN6) {x} δΦ {y} ⇔ Φ(x) = Φ(y)
K
For a collection of subsets 2 , it is axiom (sndN4) that we need to check, in
⩕
δ
considering whether a subset A in 2 K is a member of a class clsshape
Φ
E, i.e., we need
to check if
Example 6.11 Cyclic Skeletons in the interiors of a pair of triangulations are shown
in Fig. 6.5. The triangulation of an autumn Ontario, Canada maple tree is shown in
286 6 Leader Clusters and Shape Classes
Fig. 6.5a (call it T ). The triangulation of the LG& TBQ conference announcement1
is shown in Fig. 6.5b (call it T
). Let Φ(T ) equal the number of vertices in at least
one cyclic skeleton drawn on T . Triangulation T contains a pair of cyclic skeletons,
namely, skA and skB. Similarly, let Φ(T
) equal the number of vertices in at least
one cyclic skeleton drawn on T
, which contains a cyclic skeleton skE. Then we
have
Φ(intT ) = Φ(intT ) = 6.
1 Manythanks to Talia Fernos [11] for posting this LG& TBQ (conference in Geometry, Topology,
and Dynamics) announcement.
6.6 Importance of Shape Interiors Pinpointed by Strongly Descriptive Shape Classes 287
⩕
Hence, this is the beginning of a class of δΦ -based class of triangulation interiors,
namely,
⩕
δΦ
clsshape (T ) = T, T
. “
⩕
This section introduces a pair of δΦ -based optical vortex nerve classes (denoted
⩕
δ
Φ
by cls Nr vShape ). To get things started, we need to select the features of an optical
vortex nerve we want to consider. Let sk cyclic NrvE be an optical vortex nerve and let
filament A be a cusp filament on sk cyclic NrvE. For simplicity, we consider only two
features, namely,
Betti number: Φ1 (sk cyclic NrvE) = Betti number of the nerve.
Nerve Cusp feature:
1, if there is a cusp filament is on a green blob.
Φ2 (sk cyclic NrvE) =
0, otherwise.
describing an optical vortex nerve and providing a basis for a strong descriptive
proximity nerve class defined by
(b) Optical vortex nerve on girl’s forehead (c) Optical vortex nerve on a pair of faces
⩕
Example 6.12 (Two δΦ -based Classes of Optical Vortex Nerves) Let K be a trian-
gulated Granata painting2 such as the one shown shown in Fig. 6.6a equipped with
⩕
the proximity δΦ . Let sk cyclic NrvE, sk cyclic NrvE
be optical vortex nerves shown in
Fig. 6.6a and c, respectively.
In addition, for an optical vortex nerve E, let
⩕
2 Many thanks to Alessandro Granata for his permission to use his paintings in this study of δ -based
Φ
optical vortex nerve classes. Also, many thanks to M. Z. Ahmad for supplying the Matlab script
used to find optical vortex nerves on triangulated digital images.
6.7 Optical Vortex Nerve Shape Class 289
Recall from Theorem 4.26 in Sect. 4.13 that the Betti number of an optical vortex
nerve equals the number nerve cusp filaments plus 2. Then observe
⩕
A second δΦ -based optical vortex nerve class is present in the collection of Granata
paintings in Fig. 6.6. To see this, observe
290 6 Leader Clusters and Shape Classes
It is possible to equip a complex with more than one proximity, i.e., a collection of
proximities called a proximal relator.
conn
Example 6.13 ( δ -Based Class) Let K be ⎧a complex on⎫a triangulated finite
⩕
⎨conn conn
⩕
conn ⎬
bounded region equipped with the proximities δ , δ , δΦ . Let NrvA, skNrvB
⎩ ⎭
be Alexandroff nerve complex and skeletal nerve complex on K . Then the following
classes emerge from K .
Alexandroff class The Alexandroff nerve complex NrvA has a nucleus, which is
a vertex p common to the triangles in Nrv A. Let A, A
be filled triangles in
NrvA. Then
conn
From δ Axiom P4conn, Sect. 5.5
conn
A ∩ A
= p ⇒ A δ A
.
This holds true for each pair for each pair of triangles in Nrv A. Hence, NrvA is
an Alexandroff class containing proximal triangles. Since each vertex in K is the
nucleus of an Alexandroff nerve, the complex K is a collection of Alexandroff
clusters.
Class of skeletal nerves A skeletal nerve complex skNrvB is a collection of skele-
ton that have nonempty intersection. Let skB, skB
be a pair of skeletons in
skNrvB. Then
conn
From δ Axiom P4conn, Sect. 5.5
conn
skB ∩ skB
= p ⇒ skB δ skB
. “
6.8 Connectedness Proximity Classes Derived from Skeletal and Vortex Nerves 291
⩕
conn
δ Φ
To arrive at clsshape (A) classes, we rewrite the C̆ech closeness (δ) axioms to set up
⩕ ⩕
conn conn
a framework for δΦ spaces. In the run-up to the detection of classes in a δΦ space,
we first consider revisit the axioms of weak and strong descriptive connectedness
proximities of skeletons with matching description. A vortex cycle description is a
feature vector that contains features values extracted from vortices with what are
known as probe functions. Let K be a collection of vortex cycles equipped with the
⩕
conn ⩕
descriptive proximity δΦ , which is an extension of the descriptive proximity δΦ [4,
Sects. 3–4, pp. 95–98]. The mapping Φ : K −→ Rn yields an n-dimensional feature
vector in Euclidean space Rn either a vortex cyc A ∈ K (denoted by Φ(cyc A)) or
a vortex cycle vcycE in K (denoted by Φ(vcycE)) or a vortex nerve vNrvH in
K (denoted by Φ(vNrvH )). Let 2 K be a collection of cell complexes on space K
⩕
conn
equipped with the strong descriptive proximity δΦ . Also let Φ(K ) be a description
of space K defined by
A ∈ 2K .
Φ(A) ∈ Rn = real-valued feature vector describing A.
Φ(K ) = (Φ(A1 ), . . . , Φ(Ai ), . . . , Φ(An )) : Φ(Ai ) ∈ Rn .
Feature vector that describes the complex K
Each term of this massive feature vector Φ(K ) (description of the space K ) is itself
a feature vector, which is a description of a sub-complex A in the collections of
complexes 2 K . For the axioms for a descriptive proximity on space K , the usual set
intersection is replaced by descriptive intersection [4, Sect. 3] (denoted by ∩) defined
Φ
by
A ∩ B = {x ∈ A ∪ B ∈ 2 K : Φ(x) ∈ Φ(A) and Φ(x) ∈ Φ(B)}.
Φ
The descriptive closure of A (denoted by clΦ A) [10, Sect. 1.4, p. 16] is defined
by
⎧ ⎫
⩕
⎨ conn ⎬
clΦ A = x ∈ K : x δΦ A .
⎩ ⎭
⩕
conn
The weak and strong forms of δΦ satisfy the following axioms.
⩕
conn
⩕ 4 Conn [weak option]
Pconn intA ∩ intB
= ∅ ⇒ A δΦ B.
Φ
δΦ
6.9 Descriptive CW Complexes and Strong Descriptive … 293
⩕
conn
Pconn
⩕ 5 Conn [strong option] A δΦ B ⇒ A ∩ B
= ∅
Φ
δΦ
Axiom Pconn
⩕ 4 Conn is a rewrite of the C̆ech axiom P4 and axiom P ⩕ 5 Conn is an
conn
δΦ δΦ
⩕
conn
addition to the usual C̆ech axioms. It is easy to see that the δΦ proximity satisfies the
⩕
conn
remaining C̆ech axioms after replacing δ with δΦ . Let A, B, C ∈ K , skeletons in a
⩕
conn
cell complex space K equipped with the proximity δΦ , which satisfies the following
axioms.
⩕
conn
δΦ , another look.
δΦ
⩕
conn
A ∩ B = ∅ ⇔ A
δΦ B,
Φ
i.e., the sets of skeletons A and B that do not have matching descriptions are not
descriptively close (i.e., A and B are far from each other).
⩕ 2Conn
Pconn
δΦ
⩕ ⩕
conn conn
A δ B ⇒ B δΦ A,
Φ
⩕
conn
δΦ is Abelian (commutative)
⩕ ⩕ ⩕
conn conn conn
A δ (B ∪ C) implies A δΦ B or A δΦ C.
Φ
⩕
conn
δΦ detects closeness of shape interiors in unions
294 6 Leader Clusters and Shape Classes
Pconn
⩕ 4Conn
δΦ
⩕
conn
intA ∩ intB
= ∅ ⇒ A δΦ B.
Φ
interiors with matching descriptions
⩕
conn
A δΦ B ⇒ A ∩ B
= ∅ .
Φ
Tells us interiors have matching descriptions
This is called the Strong Descriptive Connectedness Axiom, since we require not
just descriptive connectedness (connected skeletons that have matching descrip-
tion) but also that the interiors of the connected skeletons have matching descrip-
⩕
conn
tions. In other words, with δΦ , we gain yet another way to compare surface
shapes. “
⩕
conn
δΦ descriptively connected interiors in spacetime.
⎛ ⩕
⎞
conn
A descriptive overlap connectedness space is denoted by ⎝ K , δΦ ⎠. Skeletons
skA, skB in a cell complex K are strongly close descriptively, provided the descrip-
tion of the skeletal interior int(skA) matches the description of the skeletal interior
int(skB). This form of proximity has many spacetime applications, since we often
want to compare objects such as 1-cycles by themselves or vortex cycles or the more
complex vortex nerves that do not overlap spatially or at the same time but yet have
descriptively matching interior shapes.
6.9 Descriptive CW Complexes and Strong Descriptive … 295
Assume nerves in Figs. 6.7a and 6.8b represent nesting, non-overlapping vortexes
(each with its own innermost, nucleus skeletal vortex with a collection of light caustic
skeletons along their boundaries) that have matching descriptions in spacetime.
Betti numbers were introduced by Ghrist [12, Sects. 2.1–2.2, pp. 65–69], [13,
Sect. 5.13, pp. 104–106] as stepping stones towards barcodes that provide a graphical
representation of the changing character of topological data. The goal is to detect
296 6 Leader Clusters and Shape Classes
⩕
conn
instances where δΦ proximities occur between optical vortex nerves over time in
terms of their descriptions using Betti numbers.
For example, Φ(sk cyclic Nrv1) = Φ(sk cyclic Nrv4) = 9, the Betti number of the corre-
sponding optical vortex nerves sk cyclic Nrv1, sk cyclic Nrv4 containing skeletal vortexes
sk1, sk4, respectively. In other words, the structure of these optical vortex nerves
persists across different video frames. That is, the Betti number for the sk cyclic Nrv1
optical vortex nerve equals the Betti number of the counterpart of sk cyclic Nrv1, namely,
sk cyclic Nrv4. In other words, the structures of these nerves persists for the snapshots
⩕
conn
shown in Figs. 6.7a and 6.8b, respectively. In that case, sk cyclic Nrv1 δΦ sk cyclic Nrv4.
6.9 Descriptive CW Complexes and Strong Descriptive … 297
Similarly, let sk cyclic Nrv2, sk cyclic Nrv3 be a pair of optical vortex nerves shown in
Figs. 6.7b and 6.8a, respectively. In this case, Φ(sk cyclic Nrv2) = 8 and Φ(sk cyclic Nrv3)
= 7. In other words, the Betti number of the corresponding optical vortex nerves
sk cyclic Nrv2, sk cyclic Nrv3 containing skeletal vortexes sk2, sk3, respectively, also do
not persist across different video frames. The structures of these optical vortex
⩕
conn
nerves change over time. In that case, sk cyclic Nrv2
δΦ sk cyclic Nrv3, i.e., the structures
of nerve sk cyclic Nrv2 and its counterpart change over time and this change is reflected
across a sequence of video frames. This is not surprising, considering the changes in
the movements and composition of the express train recorded in the video represented
in Figs. 6.7 and 6.8. “
298 6 Leader Clusters and Shape Classes
Recent work on optical vortexes: Nye [14, Sect. 3, p. 6] offers insights concern-
ing light wave vortexes, introducing a quasi-monochromatic wave function
References
1. Leader, S.: On clusters in proximity spaces. Fundam. Math. 47, 205–213 (1959)
2. Gellert, W., Küstner, H., Hellwich, M., H. Kästner, E.: The VNR Concise Encyclopedia of
Mathematics, 760 p (56 plates). Van Nostrand Reinhold Co., New York, London (1977). ISBN:
0-442-22646-2, MR0644488; see Mathematics at a glance, A compendium. Translated from
the German under the editorship of Hirsch, K.A. and with the collaboration of Pretzel, O.,
Primrose, E.J.F., Reuter, G.E.H., Stefan, A., Tropper, A.M., Walker, A., MR0371551
3. Peters, J.: Local near sets: pattern discovery in proximity spaces. Math. Comp. Sci. 7(1), 87–106
(2013). https://doi.org/10.1007/s11786-013-0143-z, MR3043920, ZBL06156991
4. Concilio, A.D., Guadagni, C., Peters, J., Ramanna, S.: Descriptive proximities. properties and
interplay between classical proximities and overlap. Math. Comput. Sci. 12(1), 91–106 (2018).
MR3767897, Zbl 06972895
5. Dareau, A., Levy, E., Aguilera, M., Bouganne, R., Akkermans, E., Gerbier, F., Beugnon, J.:
Revealing the topology of quasicrystals with a diffraction experiment. Phys. Rev. Lett., arXiv
1607(00901v2), 1–7 (2017). https://doi.org/10.1103/PhysRevLett.119.215304
6. Fermi, M.: Why topology for machine learning and knowledge extraction. Mach. Learn. Knowl.
Extr. 1(6), 1–6 (2018). https://doi.org/10.3390/make1010006
7. Fermi, M.: Persistent topology for natural data analysis - a survey. arXiv 1706(00411v2), 1–18
(2017)
8. Baikov, V., Gilmanov, R., Taimanov, I., Yakovlev, A.: Topological characteristics of oil and
gass reservoirs and their applications. In: A.H. et. al. (ed.) Integrative Machine Learning, LNAI
10344, 182–193 pp. Springer, Berlin (2017)
9. Pellikka, M., Suuriniemi, S., Kettunen, L.: Homology in electromagnetic boundary value prob-
lems. Bound. Value Probl. 2010(381953), 1–18 (2010). https://doi.org/10.1155/2010/381953
10. Peters, J.: Computational proximity. Excursions in the topology of digital images. Intell. Syst.
Ref. Libr. 102, Xxviii + 433 (2016). https://doi.org/10.1007/978-3-319-30262-1, MR3727129
and Zbl 1382.68008
300 6 Leader Clusters and Shape Classes
11. Fernos, T.: LGTBQ: a conference in geometry, topology, and dynamics on the work of LGTBQ+
mathematicians, 10–14 June 2019, at the University of Michigan. Technical report, Deparment
of Mathematics, University of Wisconsin (2018). http://www.math.wisc.edu/~kent/LG&TBQ.
html
12. Ghrist, R.: Barcodes: the persistent topology of data. Bull. Amer. Math. Soc. (N.S.) 45(1),
61–75 (2008). MR2358377
13. Ghrist, R.: Elementary Applied Topology, Vi+269 pp. University of Pennsylvania, Philadelphia
(2014). ISBN 978-1-5028-8085-7
14. Nye, J.: Events in fields of optical vortices: rings and reconnection. J. Opt. 18, 1–11 (2016).
https://doi.org/10.1088/2040-8978/18/10/105602
Chapter 7
Shapes and Their Approximate
Descriptive Proximities
7.1 Introduction
To get started, let K be a cell complex on a bounded region of a planar surface. Then
consider the following scheme that leads to two forms of approximate descriptively
connected proximities. To do this, we define a number of different approximate
descriptive proximities for cell complexes that are rewrites of the earlier descriptive
proximities (see Table 7.1). Let th > 0 be a threshold on the descriptive closeness
between a pair of shapes shA, shB on K .
• approximate descriptiveness unrestricted: With the δΦ proximity, all skele-
tons in K are considered such that
⩕
• descriptively restricted: With the δΦ proximity, all skeletons in Φ(int(A) ∪
int(B)) are considered.
© Springer Nature Switzerland AG 2020 301
J. F. Peters, Computational Geometry, Topology and Physics of Digital Images
with Applications, Intelligent Systems Reference Library 162,
https://doi.org/10.1007/978-3-030-22192-8_7
302 7 Shapes and Their Approximate Descriptive Proximities
conn
• descriptively connected nondiscriminatory: With the δΦ proximity, all skele-
conn
tons in Φ(A δ B) are considered approximately descriptively connected up to
some threshold on the difference between connectedness feature vectors on a pair
of skeletal shapes.
⩕
conn
• descriptively connected discriminatory: With the δΦ proximity, only the interior
skeletons skE in some shape shH are considered approximately strongly descrip-
tively connected up to some threshold on the difference between connectedness
feature vectors on the interiors of a pair of skeletal shapes, namely,
are considered.
For cell complexes A, B ∈ 2 K , the four different types of approximate descriptive
proximities from this categorization scheme are listed in Table 7.1.
Let K , K be a pair of cell complexes either covering the same video frame X or
covering a pair of different triangulated video frames X, X , respectively. In this case,
K ∩ K = ∅ (complexes K , K are disjoint), since X, X are different video frames.
Also, let shE, shE be a pair of shapes on K and let shK be a shape on K . Hence,
shE ∩ shE = ∅ (shapes shE, shE are disjoint), since K , K are on different video
frames. And let Φ(shE), Φ(shE ), Φ(E ) be feature vectors that describe shapes
shE, shE , shK , respectively. For example, let
m = shape mass.
v = shape velocity.
shape kinetic energy
1
Φ(energy(shE)) = m × v2 .
2
Φ(holeCount (shE)) = number of holes in the interior of shE
description of shape shE
Φ(shE) = (Φ(energy(shE)), Φ(holeCount (shE))) .
Recall that the kinetic energy of a body is the energy of that body as a result
of its motion. Each video frame shape has kinetic energy due to its motion across a
sequence of video frames, i.e., a shape shE that appears in frame frame f at time t and
reappears as shE in a later frame frame f at time t has a displacement Δf relative
304 7 Shapes and Their Approximate Descriptive Proximities
to a temporal interval represented by Δt. That is, shE ∈ frame f, shE ∈ framet
Hence, shape shE has a particle velocity vsh E , defined by
Let m sh E be the mass of a surface shape represented by shE in a video frame. Then
the energy E sh E of shape sh E is defined by
Assume that each of these nerve shapes has the same energy and almost equal
hole counts and assume that the norm of the difference between the feature vectors
is less that a threshold th, i.e.,
Φ(skcyclic NrvE) − Φ(skcyclic NrvE ) < th ⇒
skcyclic NrvE ∩ skcyclic NrvE = ∅
Φ
In other words, this example illustrates Axiom PδΦ 4 that you will find in
Sect. 7.4. “
7.2 Approximate Descriptive Intersection 305
(a) (b)
(a) (b)
are the only shapes on K and shE is the only shape on K . Assume that each of
these shapes has the same energy and almost equal diameters and hole counts and
the norm of the difference between the features is less that a threshold th, i.e.,
Φ(shE) − Φ(shE ) < th ⇒ shE δΦ shE .
Φ(shE) − Φ(shE ) < th ⇒ shE δΦ shE
shapes approx. descrip. intersection
⇒ K ∩ K = shE, shE , shE .
Φ
In all four cases in Table 7.1, we have the approximate form of descriptive intersec-
tion of shapes shE, shE on K (denoted by shE ∩ shE ) and the approximate form
Φ
of descriptive intersection of any shape shE on K and shape shE on K (denoted
by shE ∩ shE ) to consider, defined by
Φ
and
Descriptions Φ(sh A), Φ(shB) are close
(K ∪ K ) = shA, shB ∈ K ∪ K : Φ(shA) − Φ(shB) < th .
δΦ
7.2 Approximate Descriptive Intersection 307
In other words, we can say the following about a pair of cell complexes K , K
covering a pair of video frames:
This says that a pair of cell complexes K and K of video frames have approximate
descriptive closeness, provided we can find at least one shape shA in complex K that
has approximate descriptive closeness to at least one shape shB in complex K .
In Fig. 7.3, shapes shE, shE in cell complex K have approximate descrip-
tive closeness to shape shE in cell complex K . Notice, however, that shapes
shE, shE do not have approximate descriptive closeness to shape skB (head
of a tiger) in cell complex K . In other words, K δΦ K , even though the
head of a tiger (shape shB) does not have a description that is close to the
description of any shape in complex K . In other words, a pair complexes cov-
ering a pair of video frames can have approximate descriptive closeness, even
though not all shapes across the pair of complexes have approximate descriptive
closeness. “
This section introduces the basic steps in the approximate proximity approach to
comparing triangulated snapshots of surface shapes. We illustrate these steps in
terms of comparing the descriptions of optical vortex nerves that are typical structures
that appear in triangulated video frames. The proximity δΦ is the simplest of the
approximate proximities. Let skcyclic NrvE, skcyclic NrvE be a pair of optical vortex
nerves. Also let the description
Φ(skcyclic NrvE)
Select an approximation threshold th. Then skcyclic NrvE, skcyclic NrvE are close to
each other approximately descriptively, provided
Φ(skcyclic NrvE) − Φ(skcyclic NrvE ) < th.
308 7 Shapes and Their Approximate Descriptive Proximities
Example 7.3. (Nerves with approximate descriptive closeness) A pair of cell com-
plexes K , K covering a pair of different triangulated video frames is shown in
Fig. 7.3. Separated triangulated shapes shE (cylinder), shE (cylinder slice) on tri-
angulated cell complex K and shE (torus slice) on triangulated cell complex K
are also shown in Fig. 7.3. Also assume that shE, shE are the only shapes on K and
shE is the only shape on K . Assume that each of these shapes has the same energy
and almost equal diameters and hole counts and the norm of the difference between
the features is less that a threshold th, i.e.,
7.3 Steps in the Approximate Proximity Approach 309
shE − shE < th ⇒ shE δΦ shE .
shE − shE < th ⇒ shE δΦ shE
approximate descriptive intersection ∩
Φ
⇒ K ∩ K = shE, shE , shE .
Φ
Problem 7.4. K
Do the following:
1o Capture a 1/2–1 min video using your cell phone or digital camera, not videos
from the internet.
2o Use shape energy and number of shape holes as features of optical vortex nerves.
3o Choose an approximation threshold th.
4o Implement Algorithm 13 using Matlab. That is, write a Matlab script to search
through a pair of cell complexes K , K on a pair of frames in a video to find
examples of a shape shE on cell complex K .
5o Give an example of a shape shE on complex K that has approximate descriptive
closeness to a shape shE on complex K .
o
6 Give an example of a shape shE on complex K that does not have approximate
descriptive closeness to a shape shE on complex K .
7o Repeat steps 1–6 for two different videos.
8o Give a results Table with the following columns:
Shape shE, Esh E , hol eC ountsh E , Shape shE , Esh E , hol eC ountsh E , t h,
Energy, holeCount, Φ(skcyclic NrvE),
?
Φ(skcyclic NrvE ), Φ(skcyclic NrvE) − Φ(skcyclic NrvE ) < th, Y/N
For shape columns of the table, insert a tiny image showing the shape of the
optical vortex nerves E and E such as shape (from Fig. 7.2a) and
This section gives a brief look at the approximate descriptive proximity, which is an
extension of the C̆ech axioms for proximity.
Axioms for δΦ .
A proximity space K equipped with the proximity δΦ satisfies the C̆ech axioms
from Sect. 5.4 for shapes shA, shB, shC ∈ 2 K , rewritten with δΦ instead of the
usual proximity δ. In particular, we have a new version of axiom P4, dubbed axiom
PδΦ 4:
PδΦ 4 : shA ∩ shB = ∅ ⇒ shA δΦ shB.
Φ
(a)
(b)
7.4 Approximate Closeness Based on Cech Proximity 311
It is easy to verify the collection of such shapes with the hue wavelength description
satisfy the axioms of the C̆ech approximate descriptive proximity. In particular, the
norm of the difference between the average wavelengths of the hues of the pair of
shapes in Fig. 7.4 pigeonholes this pair of shapes in the space K , δΦ . “
Problem 7.6. ®
Do the following:
1o Give the four axioms for the C̆ech approximate descriptive proximity space.
2o Prove that each of the axioms from Step 1 is satisfied for all shapes like the ones
in Example 7.5. “
Problem 7.7. K
Using Matlab, do the following:
1o Find an Alexandroff nerve NrvE in each triangulated frame in a video K . Assume
that the video K is equipped with the approximate descriptive proximity δΦ .
2o Let Φ(NrvE) = wavelength of the nucleus vertex of the nerve NrvE in each
video frame. Recall that the nucleus of an Alexandroff nerve is a vertex common
to the triangles in the nerve.
3o Pick a threshold th > 0.
4o Display a pair of nerves NrvE, NrvE in the video frames from Step 1 such that
Φ(NrvE) − Φ(NrvE ) < th ⇒ NrvE δΦ NrvE .
⩕ ⩕
The δΦ approximate strong descriptive proximity is restrictive, since skA δΦ skB
only applies to those skeletons in the interior some shape shE on complex K . In
other words,
⩕
rior of another shape shE, then skA δΦ skB is considered an approximate strong
⩕
proximity. Here are the axioms for δΦ .
Definition 7.8. Let K be a cell complex, shapes A, B, C ⊂ K and sub-complex
⩕
x ∈ K . The relation δΦ on the collection of complexes 2 K is a strong descriptive
Lodato proximity, provided it satisfies the following axioms.
δ ⩕
(xdsnN0) ∅ ⩔Φ A, ∀A ⊂ K , and K δΦ A, ∀A ⊂ K
⩕ ⩕
(xdsnN1) A δΦ B ⇔ B δΦ A
⩕
(xdsnN2) A δΦ B ⇒ A ∩ B = ∅
Φ
⩕
(xdsnN3) If {Bi }i∈I is an arbitrary collection of complexes in K and A δΦ Bi ∗
⩕
for some i ∗ ∈ I such that int(Bi ∗ ) = ∅, then A δΦ ( i∈I Bi )
⩕
(xdsnN4) intA ∩ intB = ∅ ⇒ A δΦ B
Φ
⩕
When we write A δΦ B, we read shape A is approximately strongly descriptively
δ
near shape B. The notation A ⩔Φ B reads A is not approximately descriptively
strongly near B. For each approximate descriptive strong proximity, we assume the
following properties hold true.
⩕
(xdsnN5) Φ(x) ∈ Φ(int(A)) ⇒ x δΦ A
⩕
(xdsnN6) {x} δΦ {y} ⇔ Φ(x) − Φ(y) < th for some threshold th > 0
The focus here is on Axiom (xdsnN6) in checking if a pair of shapes have approx-
imate strong descriptive proximity to each other. Here is an example.
Example 7.9. (Close Shapes in a Lodato Approximate Strong Descriptive Proximity
Space) Let E, E be triangular shapes as shown in Fig. 7.5. Let Φ(E) = average
wavelength of the hues in E and let th > 0 be a sufficiently large threshold on the
difference between the average wavelengths of the hues of the two shapes E, E .
Hint: Again, choice of threshold th > 0 will vary, depending on the spread of the
numerical values of the norms of the feature vectors.
From what is displayed in Fig. 7.5 and an appropriate threshold th > 0, we have
Φ(E) − Φ(E ) < th. Consequently
Φ(E) ∈ Φ(int(E )) ⇒
⩕
E δΦ E (Axiom (xdsnN5))
⩕
E δΦ E ⇔
Φ(E) − Φ(E ) < th. (Axiom (xdsnN6)).
7.5 Approximate Strong Descriptive Proximity 313
(a) (b)
⩕
Fig. 7.5 E ∈ K δΦ E ∈ int(NrvB) ∈ K
It is straightforward task to verify the collection of such shapes with average hue
wavelength description satisfy the axioms of the Lodato approximate strong descrip-
tive proximity. “
This section carries forward the work on proximities between nerve shapes by creat-
ing an extension of Algorithm 13. This extension is given in Algorithm 14, which will
⩕
make it possible to check whether or not there is a δΦ (approximate strong descrip-
tive proximity) between a pair of optical vortex nerves skcyclic NrvE, skcyclic NrvE on
images cell complexes K , K covering the pair of images img, img .
In Algorithm 14, the focus shifts to a comparison between the possible approxi-
mate strong descriptive closeness of the interiors of a pair of optical vortex nerves.
314 7 Shapes and Their Approximate Descriptive Proximities
Problem 7.10. K
Do the following:
1o Capture a 1/2–1 min video using your cell phone or digital camera, not videos
from the internet.
2o Use shape energy and number of shape holes as features of optical vortex nerves.
3o Choose an approximation threshold th.
4o Implement Algorithm 14 using Matlab. That is, write a Matlab script to search
through a pair of cell complexes K , K on a pair of frames in a video to find
examples of a shape shE on cell complex K .
5o Give an example of a shape shE on complex K that has approximate strong
descriptive closeness to a shape shE on complex K , i.e., answer the question:
?
⩕
shE δΦ shE .
o
6 Give an example of a shape shE on complex K that does not have approximate
strong descriptive closeness to a shape shE on complex K .
7o Repeat steps 1–6 for two different videos.
⩕
8o Give a δΦ results Table with the following columns:
Shape shE, Esh E , hol eC ountsh E , Shape shE , Esh E , hol eC ountsh E , t h,
Energy, holeCount, Φ(skcyclic NrvE),
?
Φ(skcyclic NrvE ), Φ(skcyclic NrvE) − Φ(skcyclic NrvE ) < th, Y/N
For shape columns of the table, insert a tiny image showing the shape of the
7.6 Steps to Set Up Checks on Possible Approximate Strong … 315
optical vortex nerves E and E such as shape (from Fig. 7.2a) and
Problem 7.11. ®
Do the following:
⩕
1o Select shapes in a cell complex K equipped with the δΦ approximate strong
descriptive proximity.
⩕
2o Prove that each of the δΦ axioms are satisfied for all shapes from Step 1. “
Problem 7.12. K
Using Matlab, do the following:
1o Find an Alexandroff nerve NrvE that is maximal in each triangulated frame in a
video K , i.e., NrvE has the highest number of triangles with a common vertex
(the MNC of the triangulated video frame). Assume that the video K is equipped
⩕
with the approximate strong descriptive proximity δΦ . In the case where there
is more than one MNC in a video frame, randomly pick on the of MNCs.
2o Let Φ(NrvE) = with smallest area in the nerve NrvE in each video frame.
3o Pick a threshold th > 0.
4o Display a pair of nerves NrvE, NrvE in the video frames from Step 1 such that
Φ(NrvE) − Φ(NrvE ) < th ⇒ NrvE δΦ NrvE .
Table 7.2 Four different types of approximate descriptive proximity classes of shapes
which results from the triangulation of a finite, bounded planar region. Table 7.2
distinguishes between approximate descriptive proximity classes based on whether
members of a class reflect descriptive proximities that are restricted or not, or whether
members of a class reflect descriptive proximities that are discriminatory or not.
In the sequel, we introduce the construction of two of these approximate descrip-
tive proximity classes of shapes, namely,
Approximate Descriptive Optical Vortex Nerve Class:
δΦ
class clsshape E contains optical vortex nerve shapes that have approximate
descriptive closeness to the shape of a representative member of the class (see
Sect. 7.8) .
Example 7.13. (Sample Optical Vortex Nerve Contour) The contour of an optical
vortex nerve on a train traffic video frame is shown in Fig. 7.6 By contour, we mean
the cyclic skeleton that defines the boundary of an optical vortex nerve. Computing the
particle velocity of such a nerve is in terms of the contour vertexes (a sample contour
vertex is shown in Fig. 7.6). See Problem 7.14. The energy of an optical vortex nerve
is computed in terms of the particle velocity of the nerve and its relativistic mass. The
relativistic mass of an optical vortex nerve equals the sum of the nerve areas bounded
by the inner nerve vortex and outer (contour) nerve vortex. See Problem 7.16. “
Problem 7.14. K
Implement Algorithm 16 using the δΦ proximity to construct an optical vortex
nerve shape class for the frames in a video using Matlab. This will construct a class
δΦ
clsshape E containing optical vortex nerve shapes that have approximate descriptive
closeness to a representative optical vortex nerve shape for this class, defined by
Let bdy(skcyclic NrvE) be the vortex cycle along the boundary (contour) of the opti-
cal vortex nerve skcyclic NrvE. See Example 7.13 for a sample optical vortex nerve
contour, which is the boundary of the nerve. Use particle velocity vbdy(skcyclic NrvE)
as feature of a cusp nerve system shape in constructing a cusp nerve systems shape
class. That is, compute the particle velocity of a vertex on the boundary of the vortex
nerve containing N vertexes. The particle velocity vbdy(skcyclic NrvE) equals the dis-
placement Δf r between the occurrence of an optical vortex nerve in a video frame
and its subsequent occurrence in another video frame divided by Δt, i.e.,
Δf r
vbdy(skcyclic NrvE) = .
Δt
7.8 Steps to Construct Approximate Descriptive Optical Vortex Nerve Classes 319
Use your camera or cell phone, not videos from the internet to obtain videos used
δΦ
to construct optical vortex nerve shape classes. Give two sample clsshape E shape
classes found on the frames on one or more selected video. “
Problem 7.15. ®
Prove that the particle velocity of a single vertex on the contour (boundary) of
an optical vortex nerve equals the particle velocity of the entire optical vortex
nerve. “
Problem 7.16. K
Implement Algorithm 16 using the δΦ proximity to construct an optical vortex
nerve shape class for the frames in a video using Matlab. This will construct a class
δΦ
clsshape E containing optical vortex nerve shapes that have approximate descriptive
closeness to a representative optical vortex nerve shape for this class, defined by
In this problem, consider the particle velocity v par ticle and relativistic mass of an
optical vortex nerve to arrive at a precise view of the energy of an observed optical
vortex nerve skcyclic NrvE. That is, the energy E skcyclic Nrv of an optical vortex nerve
shape over a sequence of video frames is defined by
Use energy Φ(skcyclic NrvE) = E skcyclic Nrv to describe a cusp nerve system shape in
constructing a cusp nerve systems shape class. That is, each member skcyclic NrvG of
δΦ
the optical vortex nerve class clsshape E with energy that is approximately the same
as the energy of the representative optical vortex nerve skcyclic NrvE. Let th be a
threshold for the approximations for this class. Then skcyclic NrvG is a member of the
δΦ
class clsshape E, provided
skcyclic NrvG − skcyclic NrvE < th.
Use your camera or cell phone, not videos from the internet to obtain videos used
δΦ
to construct optical vortex nerve shape classes. Give two sample clsshape E shape
classes found on the frames on one or more selected video. “
Problem 7.17. ®
Redo Problem 7.16 by constructing an optical vortex nerve class containing nerves
that are approximately descriptive close to the class representative nerve using a
feature vector with 2 features to describe each member of the class, namely, particle
velocity and energy. “
In Problem 7.18, we use the energy E skcyclic Nrv of an optical vortex nerve in classify-
ing such nerves found on a sequence of video frames. Let k +1 be the number bound-
ary vertexes plus the nucleus vertex on an maximal nerve shape MNCshape G (i.e.,
maximal Alexandroff nerve) on a triangulated video frame. Also, let h be Planck’s
constant. a proportionality constant calculated by Planck [1, p. 563], in a measure of
the energy carried by a photon, given by
E = hn,
where n is the number of Planck quanta in a system per unit time. This result was found
by Worsley [2, Eq. (1), p. 312]. For us, the total energy of E MNCshape G (measured in
Joules J) of a MNC shape is estimated using
7.8 Steps to Construct Approximate Descriptive Optical Vortex Nerve Classes 321
E MNCshape G = h × (k + 1)J/s.
Let sk cyclic E, sk cyclic E be the inner and outer cyclic skeletons on an optical vor-
tex nerve skcyclic NrvG and let ksk cyclic E , ksk cyclic E be the number of vertexes on
skcyclic NrvG. For optical vortex nerve skcyclic NrvG, total nerve energy E skcyclic NrvG
is estimated using
E skcyclic NrvG = h × ksk cyclic E + ksk cyclic E J/s.
Problem 7.18. K
Implement Algorithm 16 using the δΦ proximity to construct an optical vortex nerve
shape class for the frames in a video using Matlab. This approach will construct a class
δΦ
clsshape E containing optical vortex nerve shapes that have approximate descriptive
closeness to a representative optical vortex nerve shape for this class, defined by
This class contains optical vortex nerves that are δΦ close, i.e., cusp nerve systems
that are approximately descriptively close to a representative optical vortex nerve
relative to 5 system features, namely, system particle velocity, system energy, sys-
tem diameter, system hole count and average system contour vertex wavelength.
Members of this class can be approximately descriptive close, provided each mem-
ber has a feature vector (description) that is close to the description of the class
δΦ
representative. A description of each member the class clsshape E is a feature vector
containing
particle velocity: vskcyclic Nrv for an optical vortex nerve contour (its outer boundary)
derived from particle velocity of the contour vertexes (particles).
energy: E skcyclic Nrv , which is based on the relativistic mass of an optical
vortex nerve.
diameter: diam skcyclic Nrv , which is the maximum distance between a pair
of optical vortex nerve contour vertexes.
hole count: holeskcyclic Nrv , which is the number of holes in the interior of an
optical vortex nerve, which is found by counting the number
centroids in the interior of an optical vortex nerve.
wavelength: λskcyclic Nrv , which is the average wavelength of the vertexes on
an optical vortex nerve contour.
The description Φ(skcyclic Nrv) of each member G of an optical vortex nerve system
class will be defined by the following feature vector:
322 7 Shapes and Their Approximate Descriptive Proximities
Φ(G skcyclic Nrv ) = (vskcyclic Nrv , E skcyclic Nrv , diam skcyclic Nrv ,
holeskcyclic Nrv , λskcyclic Nrv ).
In this approach to constructing n optical vortex nerve class, each member of the class
G will have a description that is close to the description of the class representative
G, relative to a chosen threshold th, i.e.,
Φ(G skcyclic Nrv ) − Φ(G skcyclic Nrv ) < th.
Use your camera or cell phone, not videos from the internet to obtain videos used
δΦ
to construct optical vortex nerve shape classes. Give two sample clsshape E shape
classes found on the frames on one or more selected videos. “
Problem 7.19. K
Redo Problem 7.16 by doing the following:
⩕
Φ
δ
1. Modify Algorithm 16 so that a class of shapes clsshape E (shapes that have approx-
imate strong descriptive closeness to the representative shape of the class) is con-
structed. The assumption made here is the shapes are found triangulated video
frames.
2. Implement the algorithm from Step 1 in Matlab. Your Matlab script will require
the introduction of an approximation threshold th > 0, an optical vortex nerve
shape shE found in a video frame, and a set of features used in a feature vector
Φ(shE) that describes shE. This nerve shape shE will serve as the representative
⩕
δΦ
of a clsshape class of nerve shapes. Based on the selection of the representative
⩕
Φ
δ
shape shE, your Matlab script will find and pigeonhole (add to a class clsshape E)
all of those video frame shapes shE that have approximate strong descriptive
closeness to shape shE, i.e..
3. Use a feature vector with 2 features to describe each member of the class, namely,
energy and hole count (number of holes in the interior an optical vortex nerve).
4. Give two examples of classes of shape constructed using your Matlab script on a
video.
Use a cell phone or a digital camera to capture videos no more than 1/2–1 min long.
Do not use videos from the internet. “
7.8 Steps to Construct Approximate Descriptive Optical Vortex Nerve Classes 323
Problem 7.20. K
Redo Problem 7.19 using the 5 features from Problem 7.18. That is use the following
⩕
Φ
δ
features to describe the members of a class of shapes clsshape E.
particle velocity: vskcyclic Nrv for an optical vortex nerve contour (its outer boundary)
derived from particle velocity of the contour vertexes (particles).
energy: E skcyclic Nrv , which is based on the relativistic mass of an optical vortex nerve.
diameter: diam skcyclic Nrv , which is the maximum distance between a pair of optical
vortex nerve contour vertexes.
hole count: holeskcyclic Nrv , which is the number of holes in the interior of an optical
vortex nerve, which is found by counting the number centroids in the interior of
an optical vortex nerve.
wavelength: λskcyclic Nrv , which is the average wavelength of the vertexes on an optical
vortex nerve contour.
The description Φ(skcyclic Nrv) of each member G of an optical vortex nerve system
class will be defined by the following feature vector:
Φ(G skcyclic Nrv ) = (vskcyclic Nrv , E skcyclic Nrv , diam skcyclic Nrv ,
holeskcyclic Nrv , λskcyclic Nrv ).
⩕
Φ
δ
This results in the construction of a class of optical vortex nerve shapes clsshape E
(shapes that have approximate strong descriptive closeness to the representative shape
of the class).
The assumption made here is that the shapes are found on triangulated video
frames. Use a cell phone or a digital camera to capture videos no more than 1/2–
1 min long. Do not use videos from the internet. “
With the introduction of a threshold th > 0 on the different between feature vectors
that describe shapes, we can look forward to witnessing separated classes of shapes
sometime merging into a single class of shapes. Let skcyclic NrvE be an optical vortex
324 7 Shapes and Their Approximate Descriptive Proximities
⩕
conn
δΦ
nerve in space K . The clsshape shape class containing this nerve is defined by
⩕ ⎧ ⎫
⩕
conn
δΦ
⎨ conn ⎬
clsshape (skcyclic NrvE) = skcyclic NrvE ∈ 2 K : skcyclic NrvE δΦ skcyclic NrvE .
⎩ ⎭
Example 7.21. (Merging Descriptive Classes) Recall from Example 6.16 in Sect. 6.9
the following descriptive proximity classes.
6.7b
6.8a
7.9 Approximate Strong Descriptive Connected Class of Shapes 325
Here, we are interested in deriving a class of shapes that have approximate descrip-
⩕
conn
tive proximity δΦ to each other. What happens next depends on the choice of the
threshold th > 0. Let th = 2. Then we obtain
From the basic approach in Example 7.21, we can begin deriving approxi-
mate descriptive
⎧ proximity classes⎫ in sequence of video frames. For each of the
⩕
⎨ ⩕ conn conn ⎬
proximites δΦ , δΦ , δΦ , δΦ coupled with the choice shape features and
⎩ ⎭
appropriate threshold th > 0, we can begin constructing classes of shapes. This
is important, since each class represents a collection of video frame shapes that
are descriptively similar. For small thresholds th > 0 and a small number of
shape features, we can expect the shapes in video frames to be chopped up into
small collections of similar shapes. This, in fact, would be desirable for cases
where the only shapes of interest are those shapes that are very close descrip-
tively. The Betti number of an optical vortex nerve shape is a good feature to
start with, since a Betti number for the free Abelian group representation of an
optical vortex nerve represents a significant amount of information about the
structure such nerve shapes. The choice of an approximate descriptive prox-
imity threshold th > 0 is analogous to the choice of the size of the holes in a
fishing net. The greater the shape proximity threshold, the greater the number
of shapes that drop into a class of shapes that have approximate descriptive
proximities close to a target shape. “
Problem 7.22. K
Do the following:
1o Select a video that is a record of visual scenes illuminated with natural sunlight
on filled cups of coffee shown at different angles in relation to the sunlight.
Each video frame is a record of a coffee cup light caustic for different cups of
coffee in different locations. Hint: Use a cell phone or an iPad to capture the
videos, which should be no more 3 min long.
2o Triangulate the centroids of holes in the video frames X in the selected video
to obtain a CW space K . Let K be equipped with the δΦ proximity.
3o Find a maximal Alexandroff nerve Nrv A, which is a collection of the maximal
number of triangles that have a common vertex (the MNC of complex K ). Each
vertex in NrvA is a centroid of a hole on video frame X . There may be more
than MNC on a video frame.
4o Find the barycenters of the triangles in the MNC NrvA.
5o Construct an optical vortex nerve skcyclic NrvE with vertexes that are the
barycenters of the triangles of the selected MNC and of the triangles along
the border
of the selected
MNC.
6o Let Φ skcyclic NrvE = the Betti number of skcyclic NrvE. See Sect. 4.11 on how
to compute the Betti number of an optical vortex nerve.
7o Select a threshold th > 0. Repeat Step 3 for each of the video frames in the
selected video.
7.9 Approximate Strong Descriptive Connected Class of Shapes 327
δΦ
This step constructs the class clsshape (skcyclic NrvE)
δ
Φ
10o Store and display the frames in the clsshape (skcyclic NrvE) class.
11o Repeat Step 1 for 3 different videos. “
Problem⩕ 7.23. K
δΦ
Let clsshape (shE) be a class of shapes that are approximately strongly descriptive
close to a representative shape shE. Repeat the steps in Problem 7.22 to construct a
⩕
δΦ
clsshape (shE) class of video frame shapes. “
Problem
conn
7.24. K
δΦ
Let clsshape (shE) be a class of shapes that are approximately descriptively connected
(close) to a representative shape shE. Repeat the steps in Problem 7.22 to construct
conn
δ Φ
a clsshape (shE) class of video frame shapes. “
Let sk cyclic CuspA, sk cyclic CuspB be a pair of cusp skeletons on an optical vortex
nerve skcyclic NrvE and let sk cyclic be the innermost skeleton on the nerve skcyclic NrvE.
Also assume that the cusp skeletons and the innermost skeleton have a common
vertex p. In that case, we obtain what is known as a tri-skeleton optical vortex nerve
G (denoted by triSk cyclic NrvG), defined by
triSk cyclic NrvG = sk ∈ skcyclic NrvE : sk = ∅ .
Example 7.25. (Tri-skeleton optical vortex nerve) Let a optical vortex nerve
skcyclic NrvE and a tri-skeleton optical vortex nerve triSk cyclic NrvG be represented
in Fig. 7.7. In this example, we have
triSk cyclic NrvG = sk ∈ skcyclic NrvE : sk = p .
That is, in Fig. 7.7b, the cusp skeletons cpA, cpB and the innermost skeleton skE on
the optical vortex nerve skcyclic NrvE have a common vertex, namely, p. Hence,
328 7 Shapes and Their Approximate Descriptive Proximities
(a) (b)
triSk cyclic NrvG is an optical vortex nerve is a child of the nerve skcyclic NrvE in
Fig. 7.7a.
Theorem 7.26. Every optical vortex nerve is a collection of tri-skeleton optical
vortex nerves.
Proof. Let sk cyclic be the innermost skeleton of an optical vortex nerve skcyclic NrvE
on a triangulated finite bounded surface shape in the plane. By definition, there is a
collection of cusp skeletons attached to the boundary of skcyclic NrvE. Each pair of
adjacent cusp skeletons cp A, cpB has a vertex p in common with skcyclic NrvE, since
each of the cusp filaments terminates in a vertex on sk cyclic . Consequently, we have
Then the set of skeletons cpA, cpB, sk cyclic is a tri-skeleton optical vortex nerve.
Every vertex on the innermost skeleton sk cyclic is attached to a cusp filament on the
nerve skcyclic NrvE. Hence, skcyclic NrvE is a collection of tri-skeleton optical vortex
nerves.
Problem 7.27. ®
What is the Betti number of a tri-skeleton optical vortex nerve? Is it always the
same? “
Problem 7.28. ®
What is the Betti number of a pair a tri-skeleton optical vortex nerves that have an
edge in common? Is it always the same? “
Problem 7.29. ®
What is the Betti number of a collection of tri-skeleton optical vortex nerves on an
optical vortex nerve? Is it always the same? “
7.9 Approximate Strong Descriptive Connected Class of Shapes 329
(a) (b)
Problem 7.30. K ⩕
conn
Repeat the steps in Problem 7.22 using the δΦ proximity to construct a
⩕
conn
δΦ
clsshape (triSk cyclic NrvG)
class derived from the optical vortex nerves on the triangu-
lated video frames of 3 different videos. Hint: Instead of comparing the description
of whole optical vortex nerves, compare the total area (instead of the Betti number) of
a selected tri-skeleton optical vortex nerve triSk cyclic NrvG on a video frame with the
tri-skeleton optical vortex nerves that on the remaining frames in the selected video.
This will result in a class of tri-skeleton optical vortex nerves that have approximately
the same total area. “
Cusp filaments represent polarized light relative to a visual scene shape captured in
a camera image or in a video frame. The polarization of light refers to the geometrical
orientation of an electromagnetic wave, which rotates either in a right direction or
in a left direction. Polarized light is a flow of photons, each of which as a particular
spin. Each cusp filament on an optical vortex nerve represents what is known a
photon qubit, which is a description of one of two possible spin states of a photon in
polarized light. A photon spin is either in the righthand or in the lefthand in direction
of travel. A cusp filament on an optical vortex nerve represents the direction of travel
of a photon along a cusp filament boundary. In our case, a light cusp filament vector
qubit records the left or right orientation of polarized light. A left orientation of a
cusp vertex vector yields an angle past 90o (see, e.g., Fig. 7.8a), a right orientation
of a cusp vertex vector is an angle less than 90o (see, e.g., Fig. 7.8b). For more about
this, see Appendix A.15.
Problem 7.31. K ⩕
conn
Repeat the steps in Problem 7.22 using the δΦ proximity to construct a
⩕
conn
δΦ
clsshape (triSk cyclic NrvG)
class derived from the optical vortex nerves on the triangu-
lated video frames of 3 different videos. Hint: Compare orientation angles of the cusp
330 7 Shapes and Their Approximate Descriptive Proximities
filaments (test case) in a particular tri-skeleton optical vortex nerve triSk cyclic NrvG on
a selected video frame with the orientation angles of the cusp filaments of each of the
tri-skeleton optical vortex nerves (sample cases) that are on the remaining frames in
the selected video. This will result in a class of tri-skeleton optical vortex nerves that
contain cusp filaments that have approximately the same orientations angles. Here,
great care is needed to set up the comparison of tri-skeleton optical vortex nerves.
Let θ L , θ R be the orientation angles of a pair of selected cusp filaments. Then let the
description (actual feature values) of these orientations be Φ(θ L ), Φ(θ R ), respec-
⩕
conn
δΦ
tively. This leads to the following set up in constructing a clsshape (triSk cyclic NrvG)
class.
This section carries forward the method used to construct approximate descriptive
classes of shapes by extending Algorithm 16 to take care of the construction of
⩕
conn conn
δΦ δΦ
clsshapeclasses of optical vortex nerve shapes. In a clsshape E
shape class, each shape
shA has approximate strong descriptive closeness to a shape shE (class representative
shape). That is, for a selected threshold th > 0, we have
⩕
conn
shE δΦ shA ⇔ shE − shA < th.
The first thing to notice about the strong descriptive connectedness proximity is the
focus on the interior of a connected space containing collections of vertexes that
7.11 Features of Approximate Strong Descriptively Connected Nerve Shapes 331
are path-connected. The features of the interior of a connected space determine the
membership of members of an optical vortex nerve shape class.
Notice that each vertex on an optical vortex nerve represents a single photon in
the reflected light that makes up the fabric of an optical vortex nerve skcyclic NrvE
shape. Also notice that the interior of nerve skcyclic NrvE is defined on the inner
cyclic skeleton sk cyclic ∈ int(skcyclic NrvE) and the cusp filaments attached to
sk cyclic . For this reason, in starting the construction of an approximate strong
descriptive connectedness shape class derived from optical vortex nerves, we
consider features of the interior int(skcyclic NrvE). “
The sum of the distances L sk cyclic E between pairs of vectors on sk cyclic E is defined by
Example 7.32. (Sample cusp filament attached to an inner cyclic skeleton on a nerve)
Sample cusp filament vertexes p, q on an optical vortex nerve skcyclic NrvE are shown
in Fig. 7.9. The outer boundary of the inner cyclic skeleton sk cyclic E are shown with
— segments. From what we have observed, the total length L L can
filament A ,L sk cyclic E
then be computed, providing a useful feature value for skcyclic NrvE. Be equipping the
⩕
conn
collection shapes on video frames with the proximity δΦ , we can begin constructing
⩕
conn
δΦ
an approximate strong descriptive connectedness class clsshape E using Algorithm 18,
⩕
conn
δ
using the single component feature vector Φ(clsshape
Φ
E) = (L L ).
filament A ,L sk cyclic E
334 7 Shapes and Their Approximate Descriptive Proximities
The construction of such a class can be refined by including some of the features for
the connected space defined by the interior of an optical cusp nerve. “
Since each cusp filament vertex represents the action of a particular photon from
the light reflected from a surface shape in a visual scene recorded by a video
camera or single shot digital camera, we compute the wavelength of a filament
vertex in terms of the wavelength of a photon. Recall from Appendix A.22 that
the wavelength of a photon (denoted by λ) is defined by
filament kinetic energy: Let filament A be a cusp filament on an optical vortex nerve
with mass m filament A and particle velocity vfilament A . Then the kinetic energy
E filament A of a cusp filament is defined by
that define an optical vortex nerve. Hence, let m := 1, since the mass m is
negligible. The energy of a cusp filament is an important example of a feature of
the interior of the connected space defined by an optical vortex nerve.
filament vertex energy: Let λ, , c be the wavelength of a photon measured in
nanometers, abbreviated nm, one thousand millionth of a meter (the wavelengths
of photons in visible light are in the intervals 620–750 nm (red), 495–570 nm
(green) and 380–400 nm (blue) with green in the middle of the visible spec-
trum), Planck’s constant and the speed of light in a vacuum (299,792, 458 m/s.
or 299,792 km/s, or 186,282 m/s), respectively. Recall from Sect. 2.7 that energy
of a photon E(λ) is defined by
Descriptive proximity:
For an introduction to computational proximity, see Peters [3].
Descriptive proximity:
For a recent overview of descriptive proximity, see Di Concilio, Guadagni, Peters,
Ramanna [4].
336 7 Shapes and Their Approximate Descriptive Proximities
Kinetic energy:
Gupta and Gupta [5] gives a good introduction to energy (capacity to do work)
and kinetic energy of a body (relative to the motion of the body).
Fundamental energy equivalence:
For an introduction to the formation of harmonic quintessence and a fundamental
energy equation, see Worsley [2].
Let h be Planck’s constant and let n be the number of Planck quanta present in a
quantum system per unit time. The total energy E of a system is defined by
E = hn.
Example 7.33. (Total energy of a quantum system [2, Appendix A, p. 317]) Let
References
Abstract This chapter takes a look at the Alexandroff version of the Brouwer–
Lebesgue tiling theorem and introduces systems of nerve complexes that have prox-
imity to each other and which are known shapes that cover all or part of the interior
of unknown surface shapes in visual scenes.
8.1 Introduction
The study of tilings extends from the time of H. Lebesgue (1875–1941) on space-
filling curves [1], approximating polygons for Lebesgue space-filling curves by
Sagan [2, Sect. 5.5, pp. 79–81] (this work moves, at least in the beginning, towards
triangulation to approximate space-filling curves), L. E. J. Brouwer (1881–1996) on
surface dimensions [3, 4] to current work on tiling by, for example, Adams, Morgan
and Sullivan [5] on colliding soap bubbles and Salepci and Welshinger [6] on tiling
skeletons on cell complexes. Brouwer introduced the simplicial approximation theo-
rem, using the Borel-Lebesgue covering theorem. Recall that a nonempty set G in the
Euclidean space Rn that includes its interior points but does not include its boundary
points is called an open set. By contrast, a nonempty set E in Rn that includes both
its interior points and its boundary points is called a closed set. Open sets and closed
sets are the basic building blocks in the Borel-Lebesgue covering theorem (Fig. 8.1
and Table 8.1).
This section briefly considers structures that are the basic building blocks in the
approximation of unknown shapes with known shapes.
(a) (b)
The focus in this work is on finite, bounded planar regions that are slices of a
3-dimensional shape such as discs from a cylinder or discs with a hole from a torus.
In other words, a plane surface (briefly, surface) is a flat slice of a 3-dimensional
shape. Every surface has a shape derived from both the boundary and the interior of
its parent 3D shape. A video frame is a flatland view of many surfaces spread across
a visual scene. Every physical surface shape is somewhat like the facet of a cut
gemstone with many holes in its interior. The trick is to gain knowledge of unknown
physical shapes represented as a collection of flat bounded surfaces in video frames.
A tile on a flat surface is a closed, bounded shape. Recall that a closed shape is a
surface subregion that includes both its shape boundary as well as its shape interior.
Examples of closed, bounded shapes are polytopes (each polytope is the intersection
of a collection of half planes, each of which includes its bounding edge) and 2-cells
(filled triangles ) in a cellular complex. Recall from Sect. 1.12 that a 2-cell can
have one or more holes in its interior. Similarly, the interior of any surface tile can
8.2 Surfaces, Shapes, Tiles and Tiling 339
either contain one or more holes or that interior itself can be a hole (see, for example,
the torus tile with a nonempty boundary and an interior that is a hole as shown in
Fig. 8.2d).
Example 8.1 (Sample Simple Tiles) The disc shA in Fig. 8.2b is a slice from a solid
tube shown in Fig. 8.2a. This is a tile, since it includes its nonempty interior int(shA)
with a boundary bdy(shA) from the slice of the tube wall, i.e.,
An ultra-thin tire tile is a disc with a hole in its interior. A source of tire tiles is a
slice of a pipe or tube. The irregular shape1 shB in Fig. 8.2d is a slice from the torus
shown in Fig. 8.2d. This is also an example of a tile that has a nonempty boundary
bdy(shB) with an interior int(shB) that is a hole. Shape shB is a tile, since
Both the disc and the torus slice are closed bounded sets, since both are defined by
their boundaries and interiors. “
Every finite bounded surface is, by definition, a shape shE, since the surface is
defined by its boundary together with its interior. A tiling of a flat surface region shE
is a placement of the tiles with no gaps between the tiles and no overlapping tiles so
that shE is a subset of the union of the tiles. Let shT1 , . . . , shTk be a collection of
tiles that cover shape shE. Then
k
G i covers the bounded closed set E
i=1
k
E⊂ Gi .
i=1
of the subsets covers E. For example, we can find triangles G i , i ≤ 3 inside the
boundaries of the polygon G so that the union of the triangles covers shape E. That
is, we have
3
G i covers the bounded closed set E
i=1
3
E⊂ G i . “
i=1
Notice that if the tiles in Theorem 8.5 are not sufficiently small relative to a
surface shape being covered, this theorem fails. Consider a 2D shape covered by a
large tile. For example, if we cover a planar (2D) shape sh A with a large triangular
shaped tile shE so that shA ⊂ shE, then all points in shA reside in 1 tile. Whenever
sufficiently small tiles cover a surface shape, Theorem 8.5 tells us that the tiles
342 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
covering the surface shape approximate that shape. What sufficiently small means
will depend on the surface shape being covered. This is a very simple idea that has
hugely important implications in the study of surface shapes.
Problem 8.6 ®
Give an example of a surface shape shE and a choice of a collection of triangles that
cover shE. “
Problem 8.7 K
For each frame in a video, pick sufficiently small triangles to cover a selected shape
in each video frame. “
In approximating surface shapes, the focus is on covering an unknown surface
shape shA with a known shape shE. A known shape is a convex hull of a finite set
of points in some Euclidean space Rn , n ≥ 1 (see Sect. 1.7 for more about convex
hulls). In other words, what we are calling a known shape is, for example, a polytope
such as a filled triangle or a filled polygon. The requirement that the known shape be
a filled triangle in the plane, for example, is important, since we are interested not
only in approximating the contour but also approximating the interior of an unknown
surface shape In this work, a known shape shE is a collection of non-overlapping
polygons with known geometry in cases where covered surface shapes have no known
representation with space-filling curves.
Recall from Sect. 1.7 the notion of covering. In terms of covering one surface
shape with another shape, shape shE covers shape shA, provided sh A ⊆ shE. In
other words, all parts of the shape shA are contained in the shape shE.
Observation 6 Approximating an Unknown Surface Shape
The basic goal is to approximate an unknown surface shape shE with a
known surface shape shE . The geometry of a known shape shE gives us a
means of measuring the limiting conditions of an unknown shape shE covered
by the known shape. Let the unknown shape shE be a finite, bounded closed
surface shape and choose the known shape shE be a collection of sufficiently
small surface shapes such as filled triangles on a triangulated surface. Then,
from the Brouwer–Lebesgue Theorem 8.5, we can always find a known shape
shE that approximates the unknown shape shE. “
Notice that a simple, filled, cyclic planar skeleton skA can be decomposed into
a collection of filled triangles that cover sk A. Recall that a planar cyclic skeleton
skA is defined by a collection of path-connected vertexes so that there is a path on
skA from any vertex p on skA that ends on p. A skeleton skA is simple, provided
the skeleton skA has no loops (self-intersections). Each edge on skeleton skA is the
bounding edge of a half plane that reaches either up or down or across, onto the
interior of skA. The end result is that skeleton skA is defined by the intersection of a
collection of half planes with bounding edges along the boundary of sk A. The area
of an unknown planar surface shape can be approximated by covering the shape with
non-overlapping triangles.
8.4 Brouwer–Lebesgue Tiling Theorem for Sufficiently Small Tiles 343
Recall that a tile on a finite, bounded flat surface region π is a shape, i.e., a nonempty
set that includes both its boundary and its interior. A familiar example of a plane tile
in a CW complex is a filled triangle. An example of a planar tile is an Alexandroff
nerve NrvE that results from the triangulation of π. This is the case, since NrvE
is a collection of triangles with a boundary defined by the outer edges of the filled
triangles with a common vertex p in NrvE and with a nonempty interior defined by
the interiors and edges attached to the nucleus p.
Proof Let K be a 2-cell complex that covers a shape shE with tiles that are Alexan-
droff nerve complexes NrvE. From the Brouwer–Lebesgue Theorem 8.5, there are
points in shE that are in at least 2 + 1 tiles, provided the tiles covering shE are
sufficiently small. A nerve tile NrvE contains a minimum of 3 vertices for a nerve
with a least one filled triangle A. By definition, each vertex in A in NrvE is the
344 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
(a) (b)
(c) (d)
Fig. 8.5 Cell Complex K, K , K approximations of the Olympus Mons Crater area
8.5 Alexandroff-Nerve Tiling Theorem 345
chain of Hawaiian mountains (from Kauai to Hawaii) would fit inside the Olympus
Mons. For more about this, see the NASA report [8].
From Theorem 8.9, we know that we can approximate the 2-dimensional crater
shape by covering the shape with sufficiently small Alexandroff nerves in a trian-
gulation of the crater shape. The trick is to choose the number of seed points in the
triangulation so that the maximum number of points in the crater shape are contained
in k, k ≥ (2+1) of the covering nerves. An approximation of the area of the Olympus
Mon crater is also obtained by covering the crater image with a filled cyclic skeleton
skA in a cell complex K containing non-equilateral 2-cells (filled triangles) and then
summing the areas of the triangles in skA.
Observation 7 Importance of Alexandroff-Nerve Tiling Theorem
The Alexandroff-Nerve Tiling Theorem 8.9 provides a lower bound on the ade-
quacy of a surface shape tiling with nerve complexes. Once we reach the point
where we have sufficiently small nerve tiles so that there are surface shape
points contained in 2 + 1 of the nerves in the covering of the shape, we also
have an indication of a starting point in improving the approximation of an
unknown surface shape with a known surface shape, namely, the shape of a
collection of contiguous Alexandroff nerves covering the unknown shape. The
term unknown surface shape means that the measurements of the interior and
the boundary of an observed surface shape are not known. “
For example, by triangulating 89 centroids and covering the crater with a filled
cyclic skeleton skA, we obtain a crude approximation of the area of Mars volcanic
crater as shown in Fig. 8.5b. Here, there is a skeleton skA filled with Alexandroff
nerves covering the crater. For instance, NrvE ∈ skA contains ( pqr ), where each
vertex is the nucleus of another nerve.
By increasing the number of centroids from 89 to 144, we obtain an improved
approximation of the area of the crater as shown in Fig. 8.5c. In that approximation,
the boundary of the filled skeleton skA is closer to the crater edges than the filled
skeleton skA in Fig. 8.5b. Another increase in the number of triangles from 144 to
233 leads to a rather good approximation of the covered crate shape as shown in
Fig. 8.5d. In this case, the boundary of the filled skeleton skA is very close to edge
of the Mars Mons crater. These three approximations have increasing accuracy with
almost concentric nerve tilings covering the triangulated Olympus Mons crater, i.e.,
Example 8.11 (Tiling Portrait on 20CA Dollar Shapes with Alexandroff nerves) One
side of a 20 CA dollar bill is shown in Fig. 8.6. Successive Tilings of the Queen
Elizabeth portrait are next (Fig. 8.7).
346 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
Figure 8.7a: Filled skeleton skA containing nerve tiles with vertexes in a subset
of 89 centroids on the face of the Queen Elizabeth portrait.
Figure 8.7b: Filled skeleton skA containing nerve tiles with vertexes in a subset
of 144 centroids on the face of the Queen Elizabeth portrait.
Figure 8.7c: Filled skeleton skA containing nerve tiles with vertexes in a subset
of 233 centroids on the face of the Queen Elizabeth portrait.
Figure 8.7d: 233 Segments covering Queen Elizabeth portrait.
Mathematical morphology is used to obtain a segmentation of the Queen Elizabeth
portrait. Each • is a centroid of a segment. Each segment has a different colour, rep-
resenting a region of the portrait where the portrait intensities are uniform. Notice the
shrinking size of the skeletons skA, skA , skA covering the face of the Queen Eliz-
abeth portrait, which is an indication of the increasing accuracy of the approximation
of the area of portrait face, i.e.,
We know that the Alexandroff nerves in this tiling are sufficiently small and that
Theorem 8.9 holds true, since the vertices of the triangle ( pqr ) in NrvE are nuclei
in the tiling of the face. In other words, there are points in the face surface that belong
to 2 + 1 nerves in the covering. “
This section introduces yet another form tiling of a plane surface in which each tile
is an optical cusp nerve that is extracted from an optical vortex nerve.
Definition 8.12 (Optical Cusp Nerve) Let skcyclic NrvG be an optical vortex nerve on
a CW complex K derived from the triangulation of barycenters of centroid-based
triangles on a finite, bounded surface and let skcyclic A be the central filled cyclic
skeleton and let skcyclic B be the outer skeleton on skcyclic NrvG. Also let skcyclic Q be a
filled cyclic cusp skeleton attached to the filled skeleton skcyclic A and let skcyclic Q be a
filled cyclic cusp skeleton attached to skcyclic A. A cusp skeleton is a cyclic skeleton
containing a pair of cusp filaments attached between vertexes on the central skeleton
skcyclic A and the outer skeleton skcyclic B in the optical vortex nerve skcyclic NrvG. There
8.6 Optical Cusp Nerve Tiles 347
(a) (b)
(c) (d)
Fig. 8.7 Successive K, K , K Alexandroff nerve tiling approximations of Queen Elizabeth Por-
trait
348 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
is a filament attached to the vertexes on cusp filaments on skcyclic A and the vertexes
on skcyclic B. An optical cusp nerve cuspNrvO is defined by
From Definition 8.12, we know what to expect when we look for optical cusp
nerves attached to an optical vortex nerve, namely, the cell common to three inter-
secting skeletons.
Lemma 8.13 The common cell in an optical cusp nerve is a vertex.
Proof Let cuspNrvE be an optical cusp nerve containing a central skeleton skcyclic A
and a pair of cusp skeletons skcyclic B, skcyclic C attached to skcyclic A. From Definition 8.12,
skcyclic B, skcyclic C have a common edge (call it filamentO), since those cusp skeletons
have a part in common with skcyclic A. Also by definition, filamentO has a vertex
attached to skcyclic A. Each cusp skeleton has an edge attached to skcyclic A (call the
edges filamentB, filamentC). Consequently,
By definition, the vertex that is a common cell in an optical cusp nerve is the
barycenter of a triangle.
From Lemma 8.13, each of the vertexes p of the central skeleton skcyclic A of an
optical vortex nerve is attached to a cusp filament, which is an edge common to a
pair of adjacent cusp skeletons attached to the central skeleton. Such a vertex p on
skcyclic A is the nucleus of an optical cusp nerve.
filament A ∩ filament A = p.
step 5: cuspNrvE = skcyclic Q 1 ∪ skcyclic Q 2 ∪ skcyclic A, such that
skcyclic Q 1 ∩ skcyclic Q 2 ∩ skcyclic A = p.
8.6 Optical Cusp Nerve Tiles 349
In other words, both filled skeletons skcyclic Q 1 , skcyclic Q 2 in Fig. 8.8 have an edge
attached to the central skeleton skcyclic A. namely, filament A, filament A , respectively.
Since these satellite skeletons are adjacent in the cusp nerve, then the attached edges
have a common vertex p, i.e., filament A ∩ filament A = p. This also means that all
three cyclic skeletons have vertex p in common, the nucleus of the cusp nerve. “
Notice that the each of the vertexes of the cusp nerve cuspNrvE in Fig. 8.8 is
the barycenter of a triangle. In the construction of an optical cusp nerve, we require
that each of the barycenters belong to a triangle whose vertexes are the centroids of
surface holes (i.e., surface regions that absorb light). For this reason, the interior of
a skeleton in an optical cusp nerve is a surface region that reflects light.
Example 8.15 (Optical cusp nerve on s triangulated photo of the Mars Olympus
Mons crater) A sample optical cusp nerve is shown on s triangulated photo of the
Mars Olympus Mons crater in Fig. 8.9. From Algorithm 19, there are thee cusp
> > >
filaments pp , qq , rr represented by red — line segments on Fig. 8.9. The area
labeled A covers most of the main volcanic crater. While the areas labeled B and C
cover smaller craters along the border of the main crater. This points to one of the
main advantages in constructing optical cusp nerves, namely, the capacity of a cusp
nerve to reach beyond portions of the border areas outside the main area covered
by the central vortex skeleton (this is the filled skeleton skcyclic A in Algorithm 19).
Recall that the central vortex skeleton is the innermost vortex in a pair of nesting,
non-concentric vortex skeletons in an optical vortex nerve. In this example, the arcs
> >
q p , p r are part of the outermost vortex skeleton (represented with orange segments
in Fig. 8.9) of the optical vortex nerve covering the Mons crater. “
350 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
Example 8.16 (Optical cusp nerve on a triangulated video frame) An optical cusp
nerve cuspNrvE is shown on s triangulated video frame2 of a museum lobby scene
in Fig. 8.10. This nerve contains 5 nesting, non-centric vortexes. Each pair of adjacent
(closest) vortexes have cusp filaments attached between them. From Algorithm 19,
there are many cusp filaments represented by dotted lines on Fig. 8.10. “
A skeletal vortex skVE is a filled vortex, provided the closure of skVE includes
a nonempty int(skV E), i.e., the interior of skVE is nonempty.
Example 8.17 (Filled Vortex in an Optical Vortex Nerve) Each vortex in the optical
vortex nerve in Fig. 8.10 is filled. For example, the shaded interior of the inner-most
vortex in Fig. 8.10 includes not only its boundary but also the shaded interior of this
vortex. “
Problem 8.19 K
Prove Theorem 8.18. “
Problem 8.20 ®
·········
For the pair of nesting, non-concentric vortexes (attached each other by cusp filaments
represented, for example, by blue dotted edges) in an optical
vortex nerve skcyclic NrvE shown in Fig. 8.10. Do the following:
1. Each vortex in Fig. 8.10 is filled. Justify the claim that skcyclic NrvE is an Alexan-
droff nerve. That is, verify that the entire collection of vortexes in Fig. 8.10 has
nonempty intersection.
2. Give the Betti number for the innermost pair of vortexes in skcyclic NrvE.
3. Give the Betti number for the outermost pair of vortexes in skcyclic NrvE.
4. Give the Betti number for the entire optical cusp nerve cuspNrvE. “
352 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
Fig. 8.11 Optical vortex nerve covering a Hyderabad campaign billboard, courtesy of S. Ramanna
Problem 8.21 ®
A pair of nesting, non-concentric vortexes (connected by red — cusp filaments) in
an optical vortex nerve skC ycNr v E covering the figures on a campaign billboard
in Hyderabad in 27 November 2018 in Hyderabad, India3 is shown in Fig. 8.11. Do
the following:
1. Give the Betti number for skC ycNr v E.
2. Give the Betti number for the optical cusp nerve cuspNr v E covering the face
of the central politician on the campaign billboard. “
Problem 8.22 K
Implement Algorithm 21 for single shot images using Mathematica. Give three sam-
ple images (using your camera or cell phone, not pictures from the internet) with
optical cusp nerves on them. “
Problem 8.23 K
Implement Algorithm 21 for single shot images using Matlab. Give three sample
images (using your camera or cell phone, not pictures from the internet) with optical
cusp nerves on them. “
Problem 8.24 K
Implement Algorithm 19 for the frames in a video using Matlab. Give three sample
video frames (using your camera or cell phone, not videos from the internet) with
optical cusp nerves on them. “
There can be more than one maximal nucleus cluster (MNC) in the triangulation of a
finite, bounded surface. For this reason, there can also be more than one barycentric
optical vortex nerve, each covering an MNC. In the case where there is more than
one optical vortex nerve, systems of optical cusp nerves appear when the cusp nerves
have an edge in common.
Example 8.26 (Sample optical cusp nerve system) A pair of optical vortex nerves
appear along the skyline of the Hyderabad street scene in Fig. 8.12. Each of these
skyline nerves has a central filled skeleton with vertexes marked with diamond
symbols. For example, in the upper lefthand corner of the skyline in Fig. 8.12, notice
the use of blue diamond symbols identify locations of the vertexes on the inner
vortex in an optical vortex nerve. Notice the inner vortex of a nerve that appears in
the center of skyline in Fig. 8.12 in which the locations of the vertexes are marked
with yellow diamonds. There is an optical cusp nerve system between that pair of
skyline nerves (see if you can identify them). There are a number of other optical
cusp nerve systems in Fig. 8.12 (see Problem 8.27). “
Problem 8.27 ®
Identify the optical vortex nerve systems in Fig. 8.12. “
Problem 8.28 K
Write a Matlab script that shades optical cusp nerve systems that appear in trian-
gulated frames in a video. Give 3 frames in a video containing shades optical cusp
nerve systems. “
A number of observations can be made about optical cusp nerves attached to the
inner vortex of an optical vortex nerve. Here are some of the observations.
354 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
Theorem 8.29 Let K be a cell complex on a triangulated surface. Let skcyclic NrvE,
cuspNrvG be an optical vortex nerve E and an optical cusp nerve G on complex K .
Then
1o The Betti number for an optical cusp nerve equals 2 + 1.
2o Every vertex p on the inner vortex of skcyclic NrvE is the nucleus of an optical
cusp nerve G on complex K .
3o The vertexes on cuspNrvG are path-connected.
4o Optical cusp nerves with an edge in common define an optical cusp nerve
system.
5o An optical vortex nerve is a collection of optical cusp nerves.
6o Each optical cusp nerve is attached to a vertex p on the inner vortex of an
optical vortex nerve.
7o Each vertex on cuspNrvG is path-connected to each vertex on skcyclic NrvE, i.e.,
every pair of vertexes on skcyclic NrvE is path-connected.
8o Let B(skcyclic NrvE) be the Betti number of the optical vortex nerve skcyclic NrvE.
The number of optical cusp nerves attached to skcyclic NrvE equals B(skcyclic NrvE)
− 2.
Proof
1o : Immediate from Definition 8.12 and Theorem 4.26 in Sect. 4.13.
2o : Immediate from Definition 8.12 for a cusp nerve and Lemma 8.13.
3o : see Problem 8.30.
4o : Let >
pq be an edge common to a collection of cusp nerves E. Then we can
write
8.7 Optical Cusp Nerve System 355
That is, the intersection of the cusp nerves in E is non-empty. Hence, from
Definition 8.25, E = cuspNrvSysE, E is a system of optical cusp nerves.
5o : From Lemma 8.13, every vertex on the inner vortex skcyclic A of an optical cusp
nerve skcyclic NrvE is attached to a cusp filament, which, by Definition 8.25,
belongs to an optical cusp nerve cuspNrvE attached to skcyclic NrvE. Conse-
quently, every vertex on skcyclic A is attached to a optical cusp nerve. Then, the
vortexes on skcyclic A belong to a collection of cusp nerves. Further, the inner
vortex skcyclic A of the nerve skcyclic NrvE is part of every cusp nerve attached
to skcyclic NrvE. Hence, an optical vortex nerve is, in fact, a collection of opti-
cal cups nerves. Further, each pair of cusp nerves cuspNrvG, cuspNrvG on
skcyclic NrvE has nonempty intersection, i.e.,
If a pair of optical vortex nerves have overlapping inner vortexes, then that pair
of nerves define a cusp nerve system (see Problem 8.31).
6o The fact that every vertex on the inner vortex of an optical vortex nerve is
the nucleus of an optical cusp nerve follows from Part 3 of this Theorem (see
Problem 8.32).
7o Each vertex on an optical cusp nerve cuspNrvG attached to an optical vortex
nerve skcyclic NrvE is path-connected to the nucleus p of that nerve. From Part 6
of this Theorem, the nucleus p is a vertex on the inner vortex of skcyclic NrvE.
By definition, the inner vortex skcyclic A on skcyclic NrvE is a cyclic skeleton.
Consequently, vertex p is path-connected to every vertex on skcyclic A. Hence,
via vertex p on skcyclic A, every vertex on cuspNrvG is path-connected to every
vertex on skcyclic A. This gives us the desired result. That is, every pair vortexes,
vertex q on cuspNrvG attached to skcyclic A and vertex r on skcyclic A is path-
connected. “
8o The k in the Betti number B(skcyclic NrvE) = k + 2 is a count of the number
of cusp filaments attached to the inner vortex of skcyclic A. From Part 2 of this
356 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
Theorem, each cusp filament is attached to a vertex p on the inner vortex skcyclic A
on skcyclic NrvE and p is the common part (nucleus) of a cusp nerve cuspNrvG
attached to vortex skcyclic A. The Betti number B(skcyclic NrvE) = k + 2 counts
not only the number of cusp filaments k attached to the inner vortex skcyclic A
but also includes a count of the number of vortexes in skcyclic NrvE, namely, 2,
which we do not consider in tabulating the number of cusp nerves attached to
skcyclic NrvE. Hence, B(skcyclic NrvE) − 2 is a count of the number of optical cusp
nerves attached to skcyclic NrvE. See Problems 8.33–8.35.
Problem 8.30 ®
Prove Part 3 in Theorem 8.29. “
Problem 8.31 K
Prove a pair of optical vortex nerves having overlapping inner vortexes define a cusp
nerve system. “
Problem 8.32 K
Prove Part 6 in Theorem 8.29. That is, walk through the steps that lead to the conclu-
sion that each vertex on the inner vortex of an optical vortex nerve is the nucleus of a
cusp nerve attached to the vortex nerve. Try a picture proof of this result for optical
cusp nerves. “
Problem 8.33 K
Write a Mathematica script that counts the number of cusp nerves attached to the
inner vortex of each optical vortex nerve derived from the barycenters of the triangles
in the maximal nucleus clusters (MNCs) in a triangulated digital image. Highlight
each optical vortex nerve found. Be sure to use a different color for each optical
vortex nerve found in the case where there is one occurrence of such a nerve in the
same image. Display three triangulated video images that includes the display of
the Betti number of the optical vortex nerves in each image and the corresponding
number of cusp nerves attached to the inner vortex of each optical vortex nerve
found. “
Problem 8.34 K
Write a Matlab script that counts the number of cusp nerves attached to the inner
vortex of each optical vortex nerve derived from the barycenters of the triangles in the
maximal nucleus clusters (MNCs) in each triangulated frame in a video. Highlight
each optical vortex nerve found. Be sure to use a different color for each optical
vortex nerve found in the case where there is one occurrence of such a nerve in the
same video frame. Display three triangulated video frames that includes the display
of the Betti number of the optical vortex nerves in each frame and the corresponding
number of cusp nerves attached to the inner vortex of each optical vortex nerve
found. “
8.7 Optical Cusp Nerve System 357
Problem 8.35 ®
What is the Betti number of a collection of intersecting optical cusp nerve
systems? “
In this section, the steps to construct a class E of barycentric cusp nerve shapes that are
⩕
Φ
δ
strongly descriptive close across video frames (denoted by clscuspNr vShape E). Recall
that cusp nerves are a collection of intersecting cyclic skeletons containing vertexes
that are barycenters found on maximal nucleus clusters (MNCs) on triangulated
video frames. A method of construction of a cusp nerve shape class is given in
Algorithm 21. This class of shapes is important, since it isolates those maximal
nucleus cluster (MNC) shapes that have interiors with matching descriptions. This
form of Alexandroff nerve shape underlies a number of forms of nerves such as
optical vortex nerve shapes and optical cusp nerve shapes. (Fig. 8.13)
Problem 8.36 K
Implement Algorithm 21 for single shot images using Mathematica. Your Mathemat-
⩕
Φ
δ
ica script constructs a class clscuspNr vShape E containing cusp nerve shapes that are
strongly descriptively close to a representative cusp nerve shape. Give 3 images
⩕
Φ δ
Fig. 8.13 Routines to build a cusp nerve shape class clscuspNr vShape E
containing shaded optical cusp nerve containing interiors that are descriptively
close. “
Problem 8.37 K
Implement Algorithm 21 for single shot images using Matlab. Your Mathematica
⩕
Φ
δ
script constructs a class clscuspNr vShape E containing cusp nerve shapes that are
strongly descriptively close to a representative cusp nerve shape. Give 3 images
8.9 Steps to Construct an Approximate Strong Descriptive … 359
containing shaded optical cusp nerve containing interiors that are descriptively
close. “
Problem 8.38 K
Implement Algorithm 19 for the frames in a video using Matlab. Give three sample
video frames (using your camera or cell phone, not videos from the internet) with
optical cusp nerves on them. “
Example 8.39 (Sample Cusp Nerve Contour Node Count) Starting with an optical
skeletal nerves E (this nerve contains an inner skeleton and outer skeleton joined
to each other by attaching a cusp filament between each node on the inner skeleton
and a node on the outer skeleton) on a triangulated video frame, select a cusp nerve
cuspNrvE on the skeletal nerves E.
A sample cusp nerve cuspNrvE is shown in Fig. 8.14. The contour of cuspNrvE
is the sequence of edges along the boundary of the cusp nerve. The length of the
contour is equated with the number of contour nodes. This is possible by making
360 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
the simplifying assumption that the length between each pair of neighbouring nodes
equals 1 unit length. In this example, the contour node count equals 13 × 1 = 13. “
Problem 8.40 K
conn ⩕
Implement Algorithm 21 using the δΦ instead of the δΦ proximity to construct a
cusp nerve systems shape class for the frames in a video using Matlab. This approach
⩕
δΦ
will construct a class clscuspNr vShapeSys E
contains cusp nerve system shapes contain-
ing skeletons that have approximate strong descriptive closeness to the skeleton a
representative cusp nerve system shape for this class, defined by
conn
sys. G δΦ close to sys. E
⩕
conn
δΦ conn
clsshape E = cuspNrvSysG ∈ K : cuspNrvSysG δΦ cuspNrvSysE .
conn
Unlike Problem 8.42, this class contains cusp nerve systems that are δΦ , i.e., cusp
nerve systems containing skeletons that are approximately descriptively close to the
skeletons a representative cusp nerve system. Let E be a representative shape in class
⩕
conn
δΦ
clsshape Eand let E be a possible member of this shape class. Let L be the contour
length of shape E and let L be the contour length of a shape E in a video frame.
Use contour node count to estimate contour shape length (see Example 8.39 to see
8.9 Steps to Construct an Approximate Strong Descriptive … 361
⩕
conn
δΦ
how to do this). Then determine if shape E is a member of clsshape E, using
th = approximation threshold.
Φinitial (E) = L .
Φnext (E ) = L .
Φinitial (E) − Φnext (E ) = L − L
⩕
δ
≤ th : accept E’ in clscuspNr
Φ
vShapeSys E, or
> th : reject E’.
Use your camera or cell phone, not videos from the internet, to obtain videos used to
⩕
conn
δΦ
construct cusp nerve system shape classes. Give two sample clsshape E shape classes
found on the frames on one or more selected videos. “
In this section, we take a look at the steps to construct a cusp nerve system shape class
⩕ ⩕
δ
Φ
δ
Φ
E (denoted by clscuspNr vShapeSys E). The class clscuspNr vShapeSys E contains cusp nerve
system shapes that have approximate strong descriptive closeness to a representative
cusp nerve system shape for this class, defined by
Recall from Definition 8.25 that a cusp nerve system is a collection of cusp nerves
that have a common part such as a common vertex or a common edge. For the details
about cusp nerve systems, see Sect. 8.7.
A method of construction of a cusp nerve system shape class is given in Algo-
rithm 23. The structure of the system to construct cusp nerve system shape classes is
depicted in the block diagram in Fig. 8.15. In this system, there is an Initialize Class
block sets of a system shape class containing a representative cusp nerve system
shape class. The Initialize Class block represents the steps in Algorithm 22. Each
additional member of this class must have a description that is approximately close to
the representative shape of the class. This notion of approximately close means that
the norm of the difference of the feature vectors describing the representative shape
362 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
⩕
δ
Φ
Fig. 8.15 Routines to build a cusp nerve system shape class clscuspNr vShapeSys E
and any other class shape must be less than some preset threshold. This approxi-
mately close requirement is enforced by the Finalize Class block in Fig. 8.15. The
Finalize Class block represents the steps in Algorithm 23.
⩕
In the construction of this shape class, the approximate proximity δΦ is utilized
(see Sect. 7.5 for details).
Particle Velocity Model for Cusp Nerve System Vertexes.
8.11 Shape Contour Particle Velocity 363
The vertexes on any cellular shape that appears and then reappears in a sequence of
video frames, are viewed as particles that have velocities. For simplicity, we consider
only the particle velocities in cusp nerve systems in this section.
The vertexes on the skeleton of an optical cusp nerve system can viewed as
particles moving along the contour of a cusp nerve system in spacetime. Vertexes on
a cusp nerve system contour like the one in Fig. 8.16 represents particles (photons)
that are in a flow of moving particles. Evidence of the movement of these particles of
light can be found in a sequence of video frames that record the changes in reflected
light from a visual scene such as traffic flow or movements of freight trains.
Recall that the velocity of an object equals the displacement covered by the object
covered in elapsed time t units. The directional distance between the initial and final
positions of a body is called displacement. In our case, we have
364 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
Δ f r = | f r1 − f r0 | .
elapsed time = number of seconds between the first occurrence and the next
occurrence of a particular cusp nerve system (denoted by Δt). Let t0 be the
elapsed time for the first occurrence of cusp nerve system in a video frame and
let t1 be the elapsed time the next occurrence of the nerve system in a later video
frame. Then we have
Δ f r = |t1 − t0 | sec.
th = approximation threshold.
G = representative of cusp nerve sys. shape class.
G = possible member of cusp nerve sys. shape class.
Φinitial (G) = vG .
Φnext (G ) = vG .
Φinitial (G) − Φnext (G ) = vG − vG .
⩕
δ
≤ th : accept G’ in clscuspNr
Φ
vShapeSys E, or
> th : reject G. “
Problem 8.42 K
Implement Algorithm 23 to construct a cusp nerve systems shape class
⩕
δΦ
clscuspNr vShapeSys E for the frames in a video using Matlab. Use particle velocity
vcuspNrv as a feature of cusp nerve system in constructing a cusp nerve systems shape
class. Then, let
Use your camera or cell phone, not videos from the internet, to obtain videos used
⩕
δΦ
to construct cusp nerve system shape classes. Give two sample clscuspNr vShapeSys E
shape classes found on the frames on one or more selected video. Hint: The same
video can be used to construct two different cusp nerve system shape classes. To do
this, select a cusp nerve system from a different video frame in the Class represen-
8.11 Shape Contour Particle Velocity 367
tative Step in Algorithm 23, i.e., choose a different initial cusp nerve system each
time you use Algorithm 23 on the same video. “
Problem 8.43 K
⩕
Implement Algorithm 23 using the δΦ instead of the δΦ proximity to construct a
cusp nerve systems shape class for the frames in a video using Matlab. This approach
⩕
δΦ
will construct a class clscuspNr vShapeSys E
contains cusp nerve system shapes that have
approximate strong descriptive closeness to a representative cusp nerve system shape
for this class, defined by
Unlike Problem 8.42, this class contains cusp nerve systems that are δΦ close, i.e.,
cusp nerve systems that are approximately descriptively close to a representative cusp
nerve system. In other words, we are not restricted to comparing the approximate
descriptive closeness of the interiors of a pair of cusp nerve systems. Members
of this class can be approximately descriptive close to either the boundary or the
interior or both boundary and interior of the representative cusp nerve system for a
particular class. Use particle velocity vcuspNrv as feature of a cusp nerve system shape
in constructing a cusp nerve systems shape class. Use your camera or cell phone, not
videos from the internet to obtain videos used to construct cusp nerve system shape
δΦ
classes. Give two sample clscuspNr vShapeSys E shape classes found on the frames on
one or more selected video. “
The relativistic mass of a cusp nerve system shape and the energy of a nerve system
observed in video frames are introduced in this section.
Let cuspNrvG, cuspNrvG be a pair of cusp nerves in a cusp nerve system
cuspNrvSysE and let A G , A G be the observed areas of cuspNrvG, cuspNrvG ,
respectively. Then the total area A E of the cusp nerve system shape is defined by
The mass m cuspNrvSys of a cusp nerve system shape is defined by its total area with
an idealized thickness equal to 1, i.e.,
368 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
The entire structure of any cellular shape containing nesting, non-concentric vor-
texes that appear and then reappear in a sequence of video frames, are viewed as a
collection of particles that have relativistic mass and velocities and that have energy.
By taking into account both the mass m cuspNrvSys and the particle velocity v par ticle
of a cusp nerve system observed in a sequence of video frames, we arrive at a means
of estimating the energy of an evolving nerve system. Lewis and Tolman [10, p.
782] observe that when a system acquires energy in any form, it acquires mass in
proportion.
In effect, for an evolving cusp nerve system on a triangulated video frame observed
in real-time, we have
By taking into account the role of the particle velocity v par ticle , we arrive at a precise
view of the energy of an observed cusp nerve system. That is, the energy E cuspNrvSys
of a cusp nerve system shape over a sequence of video frames is defined by
The issue of hole count on a cusp nerve system reduces to a count of the centroids
in the interior of the nerve system. Recall that each centroid is on a shape hole and
is source of seed points that are used to triangulate either a single image or a video
frame. For example, there are 3 holes in the interior of the cusp nerve (a system
containing one cusp nerve) in Fig. 8.10, 1 hole in the cusp nerve in Fig. 8.14 and 7
holes in the cusp nerve system in Fig. 8.16.
We still need to consider the energy of a cusp nerve system from a quantum
mechanics perspective. This is done by limiting our view of cusp nerve systems to
the flow of photons in the light reflected from visual scene surfaces and recorded
frame-by-frame in a video.
Problem 8.44 K
Recall from Sect. 2.7 that the energy of a photon with wavelength λ (denoted by
E photon (λ)) from a quantum mechanics perspective [11, Sect. 10.8, p. 344] is defined
by
2πc
E photon (λ) = (Energy of a single photon).
λ
Two formulas for cusp nerve system energy can be derived from a cusp nerve system
contour, namely,
Nerve contour vertexes: In this case, limit our observation of a cusp nerve system
shape on a triangulated visual scene to the N par ticles particles (vertexes) on the
system shape contour, give a formula for the energy of the cusp nerve system
contour. That is, give a formula for the energy of a cusp nerve system shape contour
in a video frame by considering the vertexes N par ticles particles (photons) and
ignoring the flow of photons represented by the line segments attached between
the vertexes on the nerve system shape contour.
370 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
Nerve contour vertexes plus edges: Give a formula for the energy of a cusp nerve
system shape contour in a video frame by considering both the vertexes as particles
(photons) and the flow of photons represented by the line segments attached
between the pairs of vertexes on the nerve system shape contour. For simplicity,
assume that an edge represents a flow of L photons, where L is the length of an
edge attached between a pair of vertexes. “
Problem 8.45 K
⩕
Implement Algorithm 23 using the δΦ instead of the δΦ proximity to construct a
cusp nerve systems shape class for the frames in a video using Matlab. This approach
⩕
δ Φ
will construct a class clscuspNr vShapeSys E contains cusp nerve system shapes that have
approximate strong descriptive closeness to a representative cusp nerve system shape
for this class, defined by
Unlike Problem 8.42, this class contains cusp nerve systems that are δΦ close, i.e.,
cusp nerve systems that are approximately descriptively close to a representative cusp
nerve system relative to the two system features, namely, system particle velocity
and system energy. Members of this class can be approximately descriptively close,
provided each member has a particle velocity vcuspNrvSys and system energy E cuspNrvSys
that are close to the particle velocity and energy of the class representative.
Use your camera or cell phone, not videos from the internet to obtain videos used
δΦ
to construct cusp nerve system shape classes. Give two sample clscuspNr vShapeSys E
shape classes found on the frames on one or more selected video. “
Problem 8.46 K
⩕
Implement Algorithm 23 using the δΦ instead of the δΦ proximity to construct a
cusp nerve systems shape class for the frames in a video using Matlab. This approach
δΦ
will construct a class clscuspNr vShapeSys E contains cusp nerve system shapes that have
approximate descriptive closeness to a representative cusp nerve system shape for
this class, defined by
Unlike Problem 8.42, this class contains cusp nerve systems that are δΦ close, i.e.,
cusp nerve systems that are approximately descriptively close to a representative cusp
nerve system relative to 5 system features, namely, system particle velocity, system
energy, system diameter, system hole count and average system contour vertex
8.12 Relativistic Mass of a Nerve Shape and Energy of a Nerve System 371
In this approach to constructing a cusp nerve system class, each member of the class
G will have a description that is close to the description of the class representative
G, relative to a chosen threshold th, i.e.,
Φ(G Nrvsys ) − Φ(G Nrvsys ) < th.
Use your camera or cell phone, not videos from the internet to obtain videos used
δΦ
to construct cusp nerve system shape classes. Give two sample clscuspNr vShapeSys E
shape classes found on the frames on one or more selected videos. “
Problem 8.48 uses contour node count as a feature in determining whether a cusp
nerve system in a video frame belongs to a shape class.
Example 8.47 (Sample Cusp Nerve System Contour Node Count) Starting with a
pair of optical skeletal nerves E, E (each with an inner skeleton and outer skeleton
joined to each by attaching a cusp filament between each node on the inner skeleton
and a node on the outer skeleton), select a cusp nerve on each of the skeletal nerves
E and E .
372 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
A pair of cusp nerves cuspNrvE and cuspNrvE with a common edge are shown
in Fig. reffig:cuspNrvSysContour. Hence, this pair of cusp nerves form a cusp nerve
system cuspNrvSysG. The contour of cuspNrvSysG is the sequence of edges along
the boundary of the nerve system. The length of the contour is equated with the
number of contour nodes. This is possible by making the simplifying assumption
that the length between each pair of neighbouring nodes equals 1 unit length. In this
example, the contour node count equals 27 × 1 = 27. “
Problem 8.48 K
conn ⩕
Implement Algorithm 23 using the δΦ instead of the δΦ proximity to construct a
cusp nerve systems shape class for the frames in a video using Matlab. This approach
⩕
δ
Φ
will construct a class clscuspNr vShapeSys E contains cusp nerve system shapes contain-
ing skeletons that have approximate strong descriptive closeness to the skeleton a
representative cusp nerve system shape for this class, defined by
conn
sys. G δΦ close to sys. E
⩕
conn
δΦ conn
clsshape E = cuspNrvSysG ∈ K : cuspNrvSysG δΦ cuspNrvSysE .
conn
Unlike Problem 8.42, this class contains cusp nerve systems that are δΦ , i.e., cusp
nerve systems containing skeletons that are approximately descriptively close to the
skeletons a representative cusp nerve system. Let E be a representative shape in class
⩕
conn
δΦ
clsshape Eand let E be a possible member of this shape class. Let L be the contour
length of shape E and let L be the contour length of a shape E in a video frame.
Use contour node count to estimate contour shape length (see Example 8.47 to see
⩕
conn
δΦ
how to do this). Then determine if shape E is a member of clsshape E, using
th = approximation threshold.
Φinitial (E) = L .
Φnext (E ) = L .
Φinitial (E) − Φnext (E ) = L − L
⩕
δΦ
≤ th : accept shape E’ in clscuspNr vShapeSys E, or
> th : reject shape E’.
8.13 Contour Node Count as a Feature of a Cusp Nerve System 373
Use your camera or cell phone, not videos from the internet, to obtain videos used to
⩕
conn
δΦ
construct cusp nerve system shape classes. Give two sample clsshape E shape classes
found on the frames on one or more selected videos. “
Problem 8.49 K
conn
This is a continuation of Problem 8.48. Implement Algorithm 23 using the δΦ instead
⩕
of the δΦ proximity to construct a cusp nerve systems shape class for the frames
⩕
Φ
δ
in a video using Matlab. This approach will construct a class clscuspNr vShapeSys E
contains cusp nerve system shapes containing skeletons that have approximate strong
descriptive closeness to the skeleton a representative cusp nerve system shape for
this class, defined by
conn
sys. G δΦ close to cuspNrvSysE
⩕
conn
δΦ conn
clsshape E = cuspNrvSysG ∈ K : cuspNrvSysG δΦ cuspNrvSysE .
conn
Unlike Problem 8.42, this class contains cusp nerve systems that are δΦ , i.e., cusp
nerve systems containing skeletons that are approximately descriptively close to the
skeletons a representative cusp nerve system. Let E be a representative shape in class
⩕
conn
δΦ
clsshape E and let E be a possible member of this shape class. Let L be the contour
length of shape E and let L be the contour length of a shape E in a video frame.
And let λ E be the average wavelength of the nodes on contour of shape E and let λ E
be the average wavelength of the nodes on contour of shape E . Use contour node
count to estimate contour shape length (see Example 8.39 to see how to do this).
⩕
conn
δΦ
Then determine if shape E is a member of clsshape E, using
th = approximation threshold.
feature vector for cuspNrvSysE
Φinitial (E) = (L , λ E ) .
feature vector for cuspNrvSysE
Φnext (E ) = L , λE .
Φinitial (E) − Φnext (E ) = (L , λ E ) − L , λ E
⩕
conn
δΦ
≤ th : accept shape E’ in clsshape E, or
> th : reject shape E’.
374 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
Use your camera or cell phone, not videos from the internet, to obtain videos used to
⩕
conn
δΦ
construct cusp nerve system shape classes. Give two sample clsshape E shape classes
found on the frames on one or more selected videos. “
One promising application of cusp nerve system shape classes is in the study of
vehicular traffic patterns on freeways. The proposed approach to discovering
similarities in cusp nerve systems covering important parts of frames in the indi-
vidual frames in traffic pattern videos typically collected by municipal planning
⩕
offices. The advent of δΦ -based cusp nerve system shape classes complements
earlier work on freeway traffic patterns by, for example, Małecki [12] (see. also,
Nagel and Schreckenberg [13]). Instead of the abstract view that results from
⩕
the study of traffic flow patterns using cellular automata, δΦ -based cusp nerve
system shape classes provides a visualization of traffic flows in terms of those
principal parts of triangulated video frames in which centroidal-based maximal
Alexandroff nerves (or MNCs) occur. In cases where there are multiple MNCs
that are close to each other in a video frame, we can expect to find cusp nerve
systems covering that part of a frame in which there is a high concentration
of barycenters of MNC triangles. Recall that each barycenter is on a trian-
gle with vertices that are centroids of image holes. Each hole is a dark (light
absorbing) region of a video frame. That high concentration of barycenters
highlights places where there is a high concentrations of holes (places where
centroids occur), which translates to places in frames where there are shapes
defined by their interior dark regions or holes. A cusp nerve system contains
⩕
limbs that spread across overlapping or close MNCs. Using the δΦ proxim-
ity to compare cusp nerve systems on different video frames with a particular
cusp nerve system of interest, leads to the construction of cusp nerve classes
that provide fine-grained comparisons of shapes with approximate descriptive
closeness across hundreds of frames in a traffic video. This is also an applica-
tion of Betti numbers, which provide a simple means of measuring the closeness
of cusp nerve system skeletons. Pairs of cusp nerve systems would be compared
in terms of their descriptions, which are Betti numbers. “
This section identifies open problems emerging from the study of proximal vortex
cycles and proximal vortex nerves. Vortex cycles can either be spatially close (over-
lapping vortex cycles have one or more common vertices) or descriptively close
8.14 Open Problems 375
(pairs of vortex cycles that intersect descriptively). For such cell complexes, we have
the following open problems.
Definition 8.50 (Leader Cluster) Let X be a nonempty set. For each given set A ∈
2 X , form a Leader cluster (denoted by CδsoFar (A))containing all subsets B ∈ 2 X
such that A ∩ B = ∅. Let δsoFar be any of the proximities. In effect,
Proof Each E ∈ τ is a Leader cluster of vortex cycles equipped with the proximity
⩕
conn
δΦ . Each closure cl(vcycH ) ∈ E intersects with a finite number of other vortex
cycles in E, since E is finite (closure finiteness property). Let cl(vcyc A), cl(vcycB) ∈
⩕
conn
E. For int(vcyc A) ∩ int(vcycB) = ∅ ⇒ cl(vcyc A) δ cl(vcycB), from Axiom
P4intConn (weak topology property). Hence, E has a CW topology.
Open Problems.
Here is a list open problems to consider.
open-1o Vortex photons can be spatially close (overlap). From Theorem 8.51, a
CW topology can be constructed on each cluster of vortex photons in a
uniform Leader topology on a collection of vortex photons. In that case,
the problem of considering the spatial closeness of vortex photons for
classification and analysis purposes, is simplified by considering a CW
topology on each cluster of intersecting vortex photons. This is a form
of problem reduction, which has not yet been attempted.
open-2o The space between the spiraling flux of vortex photons can be viewed
as holes. Modelling vortex photons with holes using a combination of
connectedness proximity and CW topology on clusters of such photons
for classification and analysis purposes, is an open problem. This is a
form of knowledge extraction.
open-3o It is well-known that real elementary particles can have the form of
knots [14], which have various forms in knot theory [15]. Vortex cycles
can be viewed as collections of intersecting knots. The collection of
376 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
Energy:
The kinds of energy (e.g., kinetic energy and potential energy) are considered in
Baldomir and Hammond [22, pp. 12–13, 53–55] with kinetic energy E defined
in terms of mass m and displacement ds (space interval) defined by
2
1 1 ds 1 ds 2
E = mv 2 = m = m .
2 2 dt 2 dt 2
Energy of a path:
The notion of the energy of a path comes from Milnor [23, Sect. III.12, pp. 70–
73]. This notion carries over in the study of differentiable paths in path-connected
skeletons on CW complexes. Let K be a cell complex covering a finite bounded
378 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
region of a flat surface. Assume that K has a closure finite weak topology on it.
In other words, let K be a CW complex. Let m : [0, 1] −→ K define a path on
skeleton skE on K from vertex p to vertex q with m(0) = p and m(1) = q. Also
assume that m is piecewise differentiable, which means that there is derivative
dm
dt
(particle velocity) for each vertex of m between p and q at time t. Let the set
of all such paths between p and q be denoted by
Let a, b be vertexes on path m. That is, for x, x ∈ [0, 1], we have m(x) = a and
m(x ) = b with m(x) < m(x ). Then, for 0 ≤ x ≤ x ≤ 1, the energy of m from
a to b (denoted by E ab (m)) is defined by
Problem 8.52
Do the following:
(1) Derive a formula for the energy of a path as a finite sum of particle velocities
(instead of an integral) from a to b on path m. This means that only the energy
of the vertexes (endpoints on the segments of a path-connected skeleton) would
be computed.
(2) Let K 1 , . . . , K n be cell complexes on n triangulated video frames. Let f a , f b
denote the occurrence of vertex a on skeleton skE in video frame f a at time ta
and the occurrence of vertex b on skeleton skE in video frame f b at time tb ,
respectively. Assume that skeleton skE is a replica of skeleton skE. In other
words, assume that skeleton skE reappears in a later video frame f b . Compute
particle velocity v = dm dt
using v = Δf
Δt
= ftbb − fa
−ta
. Also, assume the relativistic
mass of each vertex is 1. For a sequence K 1 , . . . , K n of triangulated video
frames, give a new version of the formula from Step 1(1).
(3) Use the formula from Step 2(2) to compute the path energy for a sequence of
triangulated video frames.
(4) Repeat Step 3(3) for two different sequences of triangulated video frames. “
Mappings between video frames:
Boxer, [24], an excellent paper on the properties of multivalued functions between
digital images, useful in the study of video frames.
Open Problems:
Peters [25, Sect. 3.3, p. 70] introduces 13 open problems related to research on
vortex cycles such as the construction of vortex cat cycles in spacetime (Problem
12) and the construction of a descriptive CW topology on a collection of C̆ech
8.14 Sources and Further Reading 379
nerve contours (Problem 13). From Chaps. 7 and 8 of this book, it is evident that
many different descriptive CW topologies on various forms of cell complexes are
possible. The fundamental Alexandroff-Whitenead closure finite and weak topol-
ogy properties can be refined and extended, depending on the choice of descriptive
proximity such as, for example, those in the following proximal relators.
Notice that, in each case, descriptive intersection will have a different form. See,
for example, the introduction to approximate descriptive intersection in Sect. 7.2.
The approximate form of descriptive proximity has been included in each of these
relators to pave the way for structures such as δΦ -based classes of vortex nerves
useful in a variety of applications that require detection, analysis and classification
of surface shapes such as those found in sequences of video frames (see, for
example, Sect. 7.8).
Photons:
Worsley and Peters [26] derives a threefold spherical model of the electron whose
radius is dictated by the speed of light.
Quantum dynamics:
Yurkin, Peters and Tozzi [27] gives a geometric view of the atom with implications
for the study of photons.
Rectangularity:
Hamrouni, Bensaci, Kherfi, Khaldi and Aiadi [28, Sect. 2.2, p. 599] extracts the
basic geometric properties of a leaf by covering the leaf with a rectangle and use
the measurements of the rectangle to approximate geometric properties of the
covered leaf. This is what is known as the rectangularity property.
380 8 Brouwer–Lebesgue Tiling Theorem and Nerve Complexes …
References
1. Lebesgue, H.: Sur les fonctions représentables analytiquement. J. de Math. 6(1), 139–216
(1905)
2. Sagan, H.: Universitext. Space-filling curves, p. xvi+193. Springer, New York (1994). ISBN:
0-387-94265-3, MR1299533
3. Brouwer, L.: Beweis der invarianz der dimensionenzahl (german). Math. Ann. 70, 161–165
(1911). Zbl JFM 42.0416.02, reviewer Prof. Bklaschke
4. Brouwer, L.: Über den natürlichen dimensionsbegriff (german). J. füar Math. 142, 146–152
(1913). Zbl JFM 44.0555.01, reviewer Prof. Bklaschke
5. Adams, C., Morgan, F., Sullivan, J.: When soap bubbles collide. arXiv 0412(020v3), 1–9 (2006)
6. Salepci, N., Welshinger, J.Y.: Tilings, packings and expected betti numbers in simplicial com-
plexes. arXiv 1806(05084v1), 1–28 (2018)
7. Grünbaum, B., Shephard, G.: Tilings and Patterns, pp. Xii+700. W.H. Freeman and Co, New
York (1987). MR0857454
8. NASA: Martian olympus mon volcano crater. Technical report, Jet Propulsion Labora-
tory/Caltech (2018). https://mars.jpl.nasa.gov/gallery/atlas/images/oly.jpg
9. Buslaev, A., Tatashev, A.: Exact results for discrete dynamical systems on a pair of contours.
Math. Methods Appl. Sci. 41(17), 1–12 (2018). https://doi.org/10.1002/mma/4822
10. Lewis, G., Tolman, R.: The principle of relativity, and non-Newtonian mechanics. Proc. Am.
Acad. Arts Sci. 44(25), 711–724 (1909). https://www.jstor.org/stable/20022495
11. Susskind, L., Friedman, A.: Quantum Mechanics. The Theoretical Minimum, xx+364 pp.
Penguin Books, UK (2014). ISBN: 978-0-141-977812
12. Malecki, K.: Graph cellular automata with relation-based neighbourhoods of cells for complex
systems modelling: A case of traffic simulation. Symmetry 9(12), 322 (2017). https://doi.org/
10.3390/sym9120322
13. Nagel, K., Schreckenberg, M.: A cellular automaton model for freeway traffic. J. Phys. I Francey
2(12), 2221–2229 (1992). https://doi.org/10.1051/jp1:1992277
14. Flammini, A., Stasiak, A.: Natural classification of knots. Proc. R. Soc. Lond. Ser. A Math.
Phys. Eng. Sci. 463(2078), 569–582 (2017). MR2288834
15. Toffoli, S.D., Giardino, V.: Forms and roles of diagrams in knot theory. Erkenntnis 79(4),
829–842 (2014). MR3260948
16. Peters, J.: Foundations of Computer Vision. Computational Geometry, Visual Image Structures
and Object Shape Detection, Intelligent Systems Reference Library 124. Springer International
Publishing, Switzerland (2017). i–xvii, 432 pp. https://doi.org/10.1007/978-3-319-52483-2,
Zbl 06882588 and MR3768717
17. Peters, J., Tozzi, A., Ramanna, S.: Brain tissue tessellation shows absence of canonical micro-
circuits. Neurosci. Lett. 626, 99–105 (2016). https://doi.org/10.1016/j.neulet.2016.03.052
18. Tozzi, A., Peters, J., Deli, E.: Towards plasma-like collisionless trajectories in the brain. Neu-
rosci. Lett. 662, 105–109 (2018)
19. Cui, E.: Video vortex cat cycles part 1. Technical report, University of Manitoba, Computa-
tional Intelligence Laboratory, Deparment of Electrical & Computer Engineering, U of MB,
Winnipeg, MB R3T 5V6, Canada (2018). https://youtu.be/rVGmkGTm4Oc
20. Cui, E.: Video vortex cat cycles part 2. Technical report, University of Manitoba, Computa-
tional Intelligence Laboratory, Deparment of Electrical & Computer Engineering, U of MB,
Winnipeg, MB R3T 5V6, Canada (2018). https://youtu.be/yJBCdLhgcqk
21. Ahmad, M., Peters, J.: Proximal C̆ech complexes in approximating digital image object shapes.
Theory and application. Theory Appl. Math. Comput. Sci. 7(2), 81–123 (2017). MR3769444
22. Baldomir, D., Hammond, P.: Geometry of Electromagnetic Systems, p. xi+239. Clarendon
Press, Oxford (1996). Zbl 0919.76001
23. Milnor, J.: Morse Theory. Based on Lecture Notes by M. Spivak and R. Wells, vi+153 pp.
Princeton University Press, Princeton (1963). MR0163331
24. Boxer, L.: Multivalued functions in digital topology. Note di Matematica 37(2), 61–76 (1909).
https://doi.org/10.1285/i15900932v37n2p61
References 381
25. Peters, J.: Proximal vortex cycles and vortex nerve structures. Non-concentric, nesting, possibly
overlapping homology cell complexes. J. Math. Sci. Modell. 1(2), 56–72 (2018). ISSN 2636-
8692, www.dergipark.gov.tr/jmsm, See, also, arXiv:1805.03998
26. Worsley, A., Peters, J.: Enhanced derivation of the electron magnetic moment anomaly from
the electron charge from geometric principles. Appl. Phys. Res. 10(6), 24–28 (2018). https://
doi.org/10.5539/apr.v10n6p24
27. Yurkin, Peters, J., Tozzi, A.: A novel belt model of the atom, compatible with quantum dynam-
ics. J. Sci. Eng. Res. 5(7), 413–419 (2018)
28. Hamrouni, L., Bensaci, R., Kherfi, M., Khaldi, B., Aiadi, O.: Automatic recognition of plant
leaves using parallel combination of classifiers. In: Amine, A., Mouhoub, M., Mohamed, O.A.,
Djebbar, B. (eds.) Computational Intelligence and Its Applications, pp. 597–606. Springer
International Publishing, Switzerland (2018). https://doi.org/10.1007/978-3-319-89743-1_51
Glossary
A.1 A
where ˜ is the spatially restricted and descriptively indiscriminant union. We can
Φ
represent this definition as the following diagram.
a b
A A∩B B
π φ π φ
c d
φ(A) φ(A) ∪ φ(B) φ(B)
A˜B
Φ
˜
A B⇔A B.
Φ
A.2 B
A, B ⊂ X.
R = {(a, b) ∈ X × X : a ∈ X and b ∈ X } .
2 K = collection of sub-complexes
in complex K .
δΦ = {(skcyclic NrvE, skcyclic NrvE ) ∈ 2 × 2 :
K K
skcyclic NrvE ∈ 2 K
Glossary 385
Binary relation on a Video: Let X video be a nonempty video, 2 X video the collection
. A binary relation Rvideo on video X video is a set of ordered
of frames in X video
pairs of X video 2 X video defined by
A.3 C
Cell: (a) In a cell complex K covering a finite, bounded region of the Euclidean
plane, a cell is a vertex, edge or filled triangle . (b) A cell is a finite planar
region with a boundary and nonempty interior. (c) Each boundary ∂q of a cell
is associated with an integer q ≥ 0, e.g., the number of edges on a hole boundary.
The p cells are generators of a free Abelian group B p , which we call a boundary
group: Eilenberg [4].
[Chain:] Let -complex X be a cell complex and let n (X ) be a free abelian group
with basis that is the collection of open n-simplexes en in X . Members of n (X )
and called n−chains. Also, elements of n (X ) are written as finite formal sums
386 Glossary
n i ein with coefficients n i ∈ Z (integers). The sum n i ein is viewed as a chain,
i i
which is a finite collection of n-simplexes in X . For about this, see Hatcher [5].
1-chain: A 1-chain is a formal sum of connected arcs (1-cells) in a -complex.
See, for example, the 1-chain defined by e1 + e2 in Fig. A.1. See 1-Cycle, Formal
sum, Path.
Chain of edges: Elements of C1 : Hatcher [5, p. 99].
Chain complex: A chain complex is a sequence C = C p , ∂ p , n = Z of
Abelian groups and their homomorphisms
∂n : Cn (K ) −→ Cn−1 (K )
is defined by
n
∂n [v0 , . . . , vn ] = (−1)i v0 , . . . , v̂i , . . . , vn .
i=0
The term v̂i is omitted from the sum. The alternating signs on the terms indicate
the simplexes are oriented, which means that for each positive term +v j , there
is a corresponding −v j , 0 ≤ j ≤ n. The maps ∂n are called chain maps (or
simplicial boundary maps). Each chain map ∂n is a homomorphism, e.g.. for
simplexes v, v , we have
2
∂n [v, v ] = (−1)i v, v , or,
i=0
∂n v + v = ∂n (v) + ∂n v , (homomorphic mapping).
∂n
v, v maps a sequence of simplexes to a sum with alternating
i.e., a chain map
signs or ∂n v + v maps a sum of simplexes to the sum of chain maps of the
individual simplexes.
Theorem A.4 (Fundamental Chain Map Theorem) For all n ≥ 0,
∂n−1 ∂n = 0.
Hence,
n
∂n x0 , . . . , x̂i , . . . , xn = (−1)i x0 , . . . , x̂i , . . . , xn .
i=0
Next, using a trick from Rotman [12, Proof of Prop. 1.1, p. 6], split this sum into
a pair of sums to obtain
388 Glossary
i−1
∂n x0 , . . . , x̂i , . . . , xn = (−1) j x0 , . . . , x̂ j . . . , x̂i , . . . , xn +
j=0
n
(−1)k−1 x0 , , . . . , x̂i . . . , x̂k , . . . , xn
k=i+1
twice in ∂n−1 ∂n, i.e.,
The term x0 , . . . , x̂i . . . , x̂ j , . . . , xn appears
in ∂n−1 x0 , . . . , x̂i , . . . , xn and in ∂n−1 x0 , . . . , x̂ j , . . . , xn . Hence, the first
terms has sign (−1)i+ j and the second terms has sign (−1)i+ j−1 . As a result, the
(n − 2) terms cancel in pairs, giving us the desired result, namely, ∂n−1 ∂n = 0.
Clock addition (aka Modular addition): Clock addition works just the way
we count on a clock, from Carter [13, Sect. 5.1.2, p. 65].
Connected Shape: A shape shA is connected, provided there is a edgewise path
between each pair of vertices in shA.
[Coset:] A coset is a set of products of elements of H each multiplied (or added)
by an element of H either on the right or on the left to form the right and left of
cosets. A subset of G of the form H x for x ∈ G is called the right coset of H
and a subset of G of the form x H is the left coset of H . For example, Let H be a
subgroup of the additive group G and let 0, a, b be in G and 0, b in H . Then, we
2 X covers E
E⊆ 2X .
Example A.5 (Shape Covered by a Nerve Complex) Let shE be a surface shape
and let NrvA be a nerve complex. Nerve Nrv A covers shape shE, provided
For more about this, see Weisstein [15]. See, also, Willard [16, Sect. 15.9, p. 104].
Example A.6 (Sample 1-cycle) Let C1 be a set of chains of edges, i.e., one-
dimensional 1-simplexes and let C0 be a set of linear combinations of vertices,
i.e., linear combinations of zero-dimensional 0-simplexes. Let σ be a path in
a simplicial
complex. Recall that a 1-chain c is a sum of paths, c = σi . Let
∂c = ∂ci be a boundary homomorphism on a 1-chain c. A 1-chain is a 1-cycle,
provided ∂c = 0. For example, in Fig. A.1, we have
[Cycles and boundaries:] The elements of Z p = ker ∂ p are called p-cycles and
the elements of B p = I m∂ p are called p-boundaries for a chain complex C.
For a chain complex, we have
cycles/boundaries = Z p /B p = ker ∂ p /I m∂ p .
Theorem A.7 ([6, Sect. 10.1, Prop. 10.1.3]) For any chain complex, B p = I m∂ p
is a subgroup of Z p = ker ∂ p ,
Proof The result follows from Theorem A.4, since ∂ p ◦ ∂ p+1 = 0, for all n.
n copies of a
b = a + a + · · · + a = na, where n is an integer.
Example A.8 Let Z0+ be the set of positive integers plus 0 with the binary oper-
ation +. The number 1 is the generator and Z0+ is cyclic group, since every
number x in Z0+ is a multiple of 1. “
Example A.9 Let Z p mod p be the set of integers plus {0, 1, 2, . . . , p − 1} with
the binary operation +. The number 1 is the generator and Z p mod p is cyclic
group, since every number x in Z p mod p is a multiple of 1. To see this, let x be
an member of Z4 mod5 = {0, 1, 2, . . . , x, . . . , 4} and recall that x mod5 is the
remainder after the division of x by 5. Then we have
390 Glossary
(0 + 5)mod5 = 0,
(1 + 5)mod5 = (6)mod5 = 1,
(2 + 5)mod5 = (1 + 1 + 5)mod5 = (7)mod5 = 2,
(3 + 5)mod5 = (1 + 1 + 1 + 5)mod5 = (8)mod5 = 3,
(4 + 5)mod5 = (1 + 1 + 1 + 1 + 5)mod5 = (9)mod5 = 4. “
A.4 D
A.5 E
[Edge:] 1-cell (line segment attached to a pair of vertexes (0-cells)). See Ver-
tex:Appendix A.21.
[Energy:] (a) Energy is shape shifter: Susskind [21, Sect. 7, p. 126].
(b) Various forms of energy from Quantum Mechanics as well as from traditional
Physics, offer a useful means of characterizing shapes. This is especially true of
shapes that appear on triangulated surfaces, especially the evolution of shape
contours: Appendix A.3. In this study of computational geometry and topology
of digital images, the following forms of energy are useful:
Milnor Energy of a path: Sect. 8.15, Item 8.15.
Kinetic energy: Sect. 8.15, Item 8.15. See, also, filament kinetic energy:
Sect. 7.11.
Nerve system energy: Sect. 8.12. See, also, Observation No. 8.12.
Photon energy: Sect. 2.7.
Shape kinetic energy: Sect. 7.2.
Susskind shape-shifting energy: Appendix A.12.
Euclidean plane: R2 .
Glossary 391
A.6 F
g = m 1 x1 + · · · m i xi + · · · + m n xn , m i ∈ Z(set of integers).
Lemma A.10 ([22, p. 764]) Every proper subgroup of a free group is free.
For more about this, see Alexandroff [Alexandrov] [27, Part III, Appendix 2, Sect.
2, p. 213].
392 Glossary
A.7 G
A.8 H
The notation inf {x, a : a ∈ A} reads greatest lower bound of the norm x, a.
It was this form of the distance between a point and a set that was introduced by
Hausdorff [30, Sect. 22, p. 128], a translation of the original German edition of
Hausdoff’s book [31]. For computational purposes, we use min {x, a : a ∈ A},
instead.
Let p = (x1 , a1 ), q = (x2 , a2 ) be points in the plane. The notation p − q is the
Euclidean distance between the points x and a, defined by
p − q = (x2 − x1 )2 + (a2 − a1 )2 (Euclidean distance).
dist ( p, A) = min { p − qi : qi ∈ A} “.
“
Glossary 393
Hole: (a) Empty space between edges: Hatcher [5, p. 101], (b) A surface region
that absorbs light, (c) An object that resists (prevents) shrinking of itself to a
single vertex, (d) A surface region with a boundary and an empty interior. Here is
a puzzler to consider: How many holes are there in the painting of the Mona
Lisa by Leonardo da Vinci? Krantz [32, Sect. 1.1, p. 1] asks: Is the hole in an
inflated basketball the same as the hole in the center of a donut? For a recent
look at the topology of black holes’ horizons and the negative mass that may exist
on the surface of a black hole, see Tozzi and Peters [33].
Homology: Study of chain complexes and chain maps that lead to Abelian groups
derived from the boundaries of holes in shapes and chain maps that are homo-
morphic [2].
[Homology Group:] (a) Harer and Edelsbrunner [34, Sect. IV.1, p. 79] observe
that a homology group provides a mathematical language for the holes in a
topological space. Homology groups focus on what surrounds the holes in a
space such as the holes in a collections of cells on a CW complex covering finite,
bounded flat regions like those found in video frames. (b) An homology group
is an Abelian group that counts the number of holes in a complex on a surface
(Munkres [35, Sect. 1.5, p. 26f]). (c) P. Giblin observes that every boundary is
a cycle [3, Sect. 4.8, p. 104]. Let (G, ◦), (H, +) be groups and let f : G −→ H
be a homomorphism (i.e., f (a ◦ b) = f (a) + f (b), a, b ∈ G). The kernel of f
(denoted by ker f ) is the set f −1 (e), where e is the identity element of G. The
image of f (denoted by im f ) is the subset f (G) in H . Let f : G −→ H be a
linear transformation between vector spaces G, H [34, Sect. IV.3, p. 93]. Then
kernel and image are defined by
Recall that G/H denotes a quotient group, which is the collection of right cosets
of H in G. For example, let H equal the subgroup {e, a}. The right cosets of H
in G are
H = {e, h} .
H a = {a, ha} ⊂ G.
Then we can write ker f /im f for the set of right cosets of im f in the kernel
ker f . Taking this a step further, let K be a cell complex on a finite, bounded flat
triangulated
surface with p cells (dimension of K ). Let σ be a cell on K . The
sum ai σi is called a p-chain (denoted by C p (K ), briefly, p ). A p-chain derived
i
from the cells on a skeletal cycle, where each cell σi is written as a multiple of
cycle cells modulo 2. The boundary of a p-cell complex is represented as a sum
u i (denoted by ∂ p ). For example, boundary of a p-chain c is the sum
i
∂c = ai ∂σi () .
i
cycles/boundaries = Z p /B p = ker ∂ p /I m∂ p .
For more about this, see Harer and Edelsbrunner [34, Sect. IV.1, p. 79-82] and
Appendix A.3 on cycles and boundaries. For more about quotient groups, see
Herstein [36, Sect. 2.6, starting on p. 41].
The quotient group Z p /B p is called a singular homology group on a cell complex
K . For more about this with lots of examples and applications, see Krantz [37,
Sect. 3.2, starting on p. 108]. See Quotient group A.16.
Homomorphism (aka Same Shape Mapping): Correspondence between two
groups, from Carter [13, Sect. 8.1, p. 157]. A homomorphism is a continuous
function [mapping] between two groups that mimics the structure of its domain in
its codomain, from Carter [13, Sect. 8.1, pp. 159–161]. For example, for a group
G and for a homomorphism ε : G → G, let g ∈ G under addition be a element
in a group G defined by g = a + b + a + a + b, we obtain
Put another way, let Ci , Ci−1 be groups. A homomorphism with respect to this
pair of groups is a continuous mapping ∂ : Ci −→ Ci−1 on Ci into Ci−1 , which
is a homomorphism, provided ∂(a + b) = ∂(a) + ∂(b), for all a, b ∈ Ci . See
Continuous, Mapping.
Historical Note 2 The term homomorphism comes from the Greek words oμo
(omo) meaning alike and μoρϕωκτ ις (morphosis), meaning to form or to shape:
Weisstein [38].
For more about this, see Klette and Rosenfeld [39, Sect. 6.4.6, p. 225].
Hue angle: Let img be a colour image in which each pixel has an R (red), G
(green) and B (blue) intensity. And let p ∈ img be a pixel with a hue angle θ p ,
which is estimated using
Example A.12 (Sample Correspondence between Colour Pixels and Their Wave-
lengths) The pixels is row 150 of the Alessandro Granata painting are false
coloured in green in Fig. A.3a. The 1-1 correspondence between colour pixel
hue angle and wavelength is shown in the plot in Fig. A.3b. “
A.9 I
Fig. A.3 Correspondence between hue angle and wavelength in nm (nanometers) for the colour
pixels in row 150
|C2 − C1 |
cSim = 1 − .
|C1 |
A.10 K
A.11 L
[Light:]
Huygens’s View: Light consists in the motion of some sort of matter: Huy-
gens [41].
Oudet’s View: Light is a flow of photons: Oudet [42].
Young’s View: Light physically is a wave: Young [43]. See, also, Dennis [44].
Newton’s View: Light consists of parts both successive and contemporary:
Newton [45, Sect. 1, p. 1]. The least light or part of light, which may be
stopped alone without the rest of the light, or propagated along, or do or suffer
anything alone, which the rest of light does not or suffers not, I call a ray of
light Newton [45].
Glossary 397
A.12 M
E = m × c2 .
Shape shifting view of the energy of an object
E
m≈ .
c2
Lemma A.13 (Based on Observation by L. Susskind) The mass of an object is
proportional to shape shifting.
Recall that the shape shA of an object A is defined by the combination of its
boundary and its interior, which is nonempty. Let sh A be the shape of an object
A. From Lemma A.13, we propose a shape function sh A (bdy A , int A , s A ) that
depends on the momentary values of bdy A (boundary of shape sh A ), int A (interior
of shape sh A ) and s A (speed of shape sh A ), which returns a single real value that
represents the combined results of the movement and shape-changing of A, i.e.,
the momentary energy of A.
398 Glossary
Theorem A.14 The mass of an object is derivable from its changing shape and
speed.
Proof By definition, the shape of an object A (denoted by sh A ) travelling in
spacetime, has a boundary bdy A and nonempty interior int A with a speed s A at
any given instant. Let sh A : R × R × R −→ R be a scalar function such that
a scalar sh A (bdy A , int A , s A ) = sh(r) is assigned to each point P with position
vector r. In addition, div shA denotes the divergence of the vector field shA and
grad sh A denotes the gradient (change in direction) of the scalar field sh A .
The differential operator sh A (bdy A , int A , s A ) is the Laplace delta of func-
tion sh A at a point (bdy A , int A , s A ). Recall that the Hamilton nabla operator ∇
provides a shorthand for standard derivatives.
The Laplace delta of sh A is defined by
vertices, its length), shape interior (its area) and displacement relative to its initial
and next appearance in a pair of video frames. Evidence of this shape-shifting
view of mass can be found in the ever-changing boundaries of surface holes,
which directly influence the surface shapes containing those holes. Video frames
provide a record of shape-shifting and minute changes in path energy levels from
instant to instant. What we have in a sequence of video frames is a little history of
shape-shifting as the surfaces in a recorded visual scene undergo change (erosion,
inflation, contraction) and as the recorded surfaces move in spacetime.
(b) The mass of a physical body is a property of the body and a measure of the
resistance to acceleration by the body. The kilogram (kg) is the standard unit of
mass.
Mass Point: Planck mass m is the maximum mass of point particles ≈2.18×10−8
kg. For more on mass points, see the optics–mechanics analogy in Fermi [49],
handwritten notes on quantum mechanics from Fermi’s University of Chicago
lectures.
A.13 N
[Nerve Complex:]
Alexandroff nerve: Collection triangles with a common vertex (see Sect. 1.23).
MNC Nerve Complex: A nerve complex containing the maximal number of
components. For example, an Alexandroff MNC contains the maximum num-
ber to triangles with a common vertex and an MNC vortex nerve contains the
maximum number of vortexes with either a common vertex or a common edge.
Skeletal Nerve Complex: A collection of skeletons with nonempty intersec-
tion (see Sect. 2.6).
Gemini Nerve Complex: A collection of skeletons with a common vertex or
a common edge (see Sect. 2.13).
From Lemma 8.13, spokeE = p, a vertex common to a pair of intersecting
spokes. “
A.14 O
A.15 P
Fig. A.5 Light caustic cusp vector qubits derived from an optical vortex nerve
See 1-cycle.
Pixel: Picture element in a digital image. For the details about raster image pixels,
see Peters [53, p. 14, p. 88].
[Pixel Qubit:] An isolated quantum spin is an example of a qubit. Susskind and
Friedman [54, pp. 2–3] observe that attached to an electron is an extra degree of
freedom called its spin (aka quantum spin, which is the momentum and angular
momentum isolated from the electron). A qubit is a basic unit of quantum infor-
mation such as an electron spin, a two state (or level) of a quantum mechanical
system, which has two levels (spin up and spin down). In our case, the focus is on
the photon qubits of recorded polarized light. A photon qubit is an isolation of
photon energy in a two state quantized system in which a single photon is either
in a vertical polarization state or a horizontal polarization state. A photon qubit
is a quantum mechanical description of the state of a polarized sinusoidal plane
electromagnetic wave.
Polarization of light refers to the geometrical orientation of an electromagnetic
wave, which rotates either in a right direction or in a left direction. A photon
qubit is a description of one of two possible spin states in which a photo spin
is either in the right hand or in the left hand in direction of travel. In our case, a
light cusp filament vector qubit records the left or right direction of polarized
light (see, e.g., Fig. A.5a as the basis for a partial light caustic cusp vector qubit
information system shown in Fig. A.5). A left direction of a cusp vertex vector
yields an angle past 90◦ (see, e.g., Fig. A.5b), a right direction of a cusp vertex
vector is an angle less than 90◦ (see, e.g., Fig. A.5c). Zizzi [55] introduces one
qubit (quantum bit), instead of a bit. See seed points: Appendix A, Sect. A.18.
[Polarization:] Polarization of light refers to the geometrical orientation of an
electromagnetic wave, which rotates either in a right direction or in a left direction.
402 Glossary
A.16 Q
∂i = i th p − simplex in K .
λi = ±1 or 0.
c= λi ∂i .
(b) Let G, H be a pair of groups. Giblin [3] observes: The quotient group G/H is
a device for ‘ignoring’ or ’making zero’ the elements of H . Notice, for example,
with a homology group, we consider only the right cosets of H in G and ignore
H . See Homology Group: Appendix A.8.
In addition, let K p be a set of simplexes in K of dimension at most p. Let
Z p be a set of simplicial n-cycles, which are boundaries of the union of some
(n+1)-simplexes. For example, in Fig. A.6, Z 1 is a group of 1-cycles that are
boundaries of 2-simplexes. In this example, the sequence of connected seg-
ments pq, qr , r s, sp is a 1-cycle boundary on the pair oriented 2-simplexes
pqs, qr s. (Z n , +) , n ≥ 0 (briefly, Z n ) is the kernel subgroup ker ∂n ⊆ Cn (set
of all simplicial n-cycles. In addition, let Bn be a set of simplicial n-boundaries
of holes. Tne nth simplicial homology group of a finite simplicial complex K is
the quotient group (denoted by Hn (K )) defined by
A.17 R
[Reflection:] Light bounced off a surface. A smooth surface such as water or glass
or polished metal will reflect light so that the reflection angle of an incident ray of
light equals the angle of reflection of the reflected ray of light. A bumpy surface
will result in varying angles of reflection relative to the angle of incidence of the
rays of light. Diffuse reflection results from rays of light striking a bumpy surface.
[Refraction:]
μ1 sinα = μ2 sinβ.
A.18 S
Saturation: Let img be a colour image in which pixel has an R (red), G (green)
and B (blue) intensity. The saturation of img is the amount of white in each image
pixel colour. And let p ∈ img be a pixel with saturation S p , which is estimated
using
404 Glossary
Seed Points: A seed point is a vertex used in either the tessellation or triangulation
of a finite, bounded region. Here is a list of possible seed points useful in Voronoï
and Delaunay mesh generation. Seed points are also called mesh generating
points.
1o centroid. The earliest known mesh generating point. Useful in the study of
the shapes of finite, bounded surfaces, especially in the context of vertexes
on Alexandroff nerves on triangulated centroids on surface holes, leading
to barycentric nesting, non-concentric, overlapping vortex cyclic skeletons
and a collection of light caustic folds and cusp filaments that provide a
framework for optical vortex nerves, denoted by skcyclic NrvE. The cusp
filament vertexes are triangle barycenters (between the centroids of holes,
which are vertices in an Alexandroff nerve). For this reason, a cusp filament
represents a path followed by reflected light from the surface of a shape in
a visual scene. The orientation of a cusp filament is a source of a photon
spin qubit (see qubit: Appendix A, Sect. A.15).
2o corner. Useful, provided the selection is restricted. See, e.g., Mscript 19
in Appendix A in [50, pp. 356–357].
3o rectangular grid line intersections.
4o pixel colour wavelength. See Wavelength of Light.
5o pixel intensity. Useful, provided the selection is restricted.
6o edge pixel. Typically, edge pixels provide too many seed points.
7o key point. Distinguished edge pixel in an image (cf. ImageKeypoints in
Mathematica). See, e.g., Mscript 32and Mscript 33 in Appendix A in [50,
pp. 370–371].
8o critical point. Also called an interesting point.
9o randomly selected points. Useful for some experiments.
10o salient point. cf. ImageSaliencyFilter in Mathematica.
11o hybrid centroid-corner. A centroid that is also corner.
12o hybrid centroid-edge. A centroid that is also an edge.
13o hybrid pixel-intensity-centroid. A centroid that is a distinguished point in
a level set of intensities.
14o hybrid pixel-intensity-corner. A corner that is a distinguished point in a
level set of intensities.
15o hybrid pixel-intensity-edge. An edge pixel that is a distinguished point in
a level set of intensities.
16o hybrid pixel-intensity-keypoint. An keypoint that is a distinguished point
in a level set of intensities.
17o hybrid pixel-intensity-critical point. An critical point that is a distinguished
point in a level set of intensities.
Glossary 405
Hn (K ) = Z n (K )/Bn (K ).
406 Glossary
After factoring out all copies of boundary cycles in Bn that are also cycles in
Z n , what survives in this quotient group are the n-dimensional holes, i.e. those
n-cycles that are not n-boundaries: Rotman [12, Sect. 1.1].
Singularity: A singularity is point at which a function [also, surface shape] blows
up (tends to infinity) or becomes degenerate (tends to zero). A nerve complex
becomes degenerate when it shrinks to a single vertex. For example, a singularity
of a cusp optical nerve is a single vertex common to pair of very thin optical vortex
nerve spokes, i.e., when the spokes shrink skeletons.
Spoke: A spoke is a pair of polygons that include the nucleus of a mesh nerve
and which have an edge in common. See Observation A.13. In the case where a
mesh nerve Nrv Alexandroff E is Alexandroff, then the pair of triangles in a nerve spoke
can have either a vertex or an edge in common. Again, for example, there is a
singularity in the width of a cusp of a coffee cup caustic that tends to zero as the
cusp tends to a point (the tip of the cusp).
A.19 T
A.20 U
Upper bound: Threshold th > that defines a class of shapes in which each mem-
ber of the class has approximate proximity to a particular shape relative the thresh-
old th. For more about this, see Sects. 7.3 and 7.7.
A.21 V
Value: Let img be a colour image in which pixel has an R (red), G (green) and B
(blue) intensity. The value of img is the brightness of each pixel in img. And let
p ∈ img be a pixel with value V p , which is estimated using
Vortex [Vorticity]: (a) A solid body-like rotation that can be imparted to the ele-
ments because of a stress distribution in a fluid: Cottet and Koumoutsakos [57,
Sect. 1, pp. 1–2]. (b) Optical rotation (Kamandi, Albooyey, Veysi, Rajaei, Zeng,
Wickramasinghe and Capolino [58]). See Edge: Sect. A.5.
A.22 W
For a complete introduction to the wavelength of light waves, see Susskind and
Friedman [54, Sect. 8.2, p. 260].
[Wavelengths of Colour Image Pixels:] The wavelengths of colour image pixels
are sources of seed points, useful in generating image meshes and also useful as
distinguished features of selected pixels such as image region centroids.
Weak Topology: (1) Let X be a Hausdorff space and let E be a cell decomposition
of X . Recall that a cell decomposition of a topological space is a partition of X
into subspaces called cells so that every member of X lies in exactly one cell. A
topological space is an n-cell, provided the space is homeomorphic to Rn . The
collection τE is a Jänich weak topology, provided A ⊂ X is closed if and only if
A∩cle, e ∈ E is also closed [59, Sect. VII.3]. (2) Let be a collection of mappings
f : X −→ Y on a set X into a topological space Y . Define the topology τ to be
a collection of all unions and finite intersections of the f −1 (V ) for each open set
V ⊂ Y . The topology τ is a weak topology on X .
See CW-Complex, Homeomorphism, Topology.
A.23 X
The constant K is called the modulus of the X 9 germ φ(x, y). The complete
unfolding of the X 9 caustic requires 8 control variables. For just 3 controls a, b, c,
the unfolding of the X 9 caustic is defined by
See, also, Nye [61, Sect. 1, p. 2]. The representation of the unfolding of X 9 with
K = 2 in Fig. A.7 is based on the unfolding of X 9 introduced by Nye [60, Sect.
4, p. 407].
In this work, X 9 optical catastrophes are of interest because of their similarity
to the elementary structures in optical cusp nerves (see Sects. 8.6 and 8.7) and
Appendix A.13 as well as in optical vortex nerves (see Sect. 4.11 on optical vortex
nerves) and Appendix A.13.
X. Oudet view of light: Flow of photons [42].
A.24 Z
[Set of integers:] Z.
[Zero shot recognition:] Classification of images with no training data. See
Sect. 5.13 and Application No. 5.13.
References
1. Ahmad, M., Peters, J.: Descriptive unions. A fibre bundle characterization of the union of
descriptively near sets. arXiv 1811(11129v1), 1–19 (2018)
2. Hilton, P.: A brief, subjective history of homology and homotopy theory in this century. Math.
Mag. 61(5), 282–291 (1988). MR0979026, homology until 1940
Glossary 409
3. Giblin, P.: Graphs, Surfaces and Homology, 3rd edn. Cambridge University Press, Cambridge
(2016). Xx+251 pp. ISBN: 978-0-521-15405-5, MR2722281, first edition in 1981, MR0643363
4. Eilenberg, S.: Homology of spaces with operators. i. Trans. Am. Math. Soc. 61, 757–831
(1947). MR0021313
5. Hatcher, A.: Algebraic Topology. Cambridge University Press, Cambridge, UK (2002). xii+544
pp. ISBN: 0-521-79160-X, MR1867354
6. Adhikari, M.: Basic Algebraic Topology and its Applications. Springer, Berlin (2016).
Xxix+615 pp. ISBN: 978-81-322-2841-7, MR3561159
7. Tourlakis, G.: Group extensions and homology. SIAM J. Appl. Math. 33(1), 51–54 (1977).
MR0515290
8. Corcoran, P., Winstanley, A., Mooney, P., Tilton, J.: Self-intersecting polygons resulting from
contour evolution for shape similarity. Int. J. Shape Model 15(1), 93–109 (2009). MR2804504
9. Smirnov, Y.M.: The theory of shapes. i. (Russian). Algebra Topol. Geom. 19, 181–207, 276
(1981). MR0639760; Translated from Itogi Nauki i Tekhniki, Seriya Algebra, ropologiya,
Geometriya, vol. 19, pp. 181–207 (1981)
10. Ibrahim, S.: Data-inspired advances in geometric measure theory: Generalized surface and
shape metrics. Ph.D. thesis, Washington State University, Department of Mathematics (2014).
Chair: K.R. Vixie, MR3295312; arXiv:1408.5954v1, 26 Aug. 2014
11. Vixie, K., Clawson, K., Asaki, T., Sandine, G., Morgan, S.: Multiscale flat norm signatures for
shapes and images. Appl. Math. Sci. 4(667-680), 93–109 (2009). MR2595506
12. Rotman, J.: An Introduction to Homological Algebra, 2nd edn. Universitext. Springer, New
York (2009). Xiv+709 pp. ISBN: 978-0-387-24527-0, MR2455920
13. Carter, N.: Visual Group Theory. Mathematical Association of America, Classroom Resource
Materials Series, Washington, DC (2009). xiv+297 pp. ISBN: 978-0-88385-757-1, MR2504193
14. Carr, H., Duke, D.: Joint contour nets. Found. Comput. Math. 18(6), 1333–1396 (2018).
MR3875842
15. Weisstein, E.: Cover. Wolfram MathWorld (2018). http://mathworld.wolfram.com/Cover.html
16. Willard, S.: General Topology. Dover Publications, Inc., Mineola (1970). Xii + 369pp, ISBN:
0-486-43479-6 54-02, MR0264581
17. Alayragues, S., Damiand, G., Lienhardt, P., Peltier, S.: A boundary operator for computing the
homology of cellular structures. HAL archives-ouvertes.fr (2011). http://hal.archives-ouvertes.
fr/hal-00683031v1
18. Ww2010.: Department of Atmospheric Sciences. University of Illinois (2010). http://ww2010.
atmos.uiuc.edu/(Gh)/guides/mtr/opt/mch/refr/less.rxml
19. Nye, J.: Natural Focusing and Fine Structure of Light. Caustics and Dislocations. Institute of
Physics Publishing, Bristol (1999). xii+328 pp. MR1684422
20. Wright, F.: Wavefield singularities: a caustic tale of dislocation and catastrophe. Ph.D. thesis,
University of Bristol, H.H. Wills Physics Laboratory, Bristol, England (1977). https://research-
information.bristol.ac.uk/files/34507461/569229.pdf
21. Susskind, L.: The Black Hole War, p. 470. Back Bay Books, New York (2008)
22. Eilenberg, S., MacLane, S.: Group extensions and homology. Ann. Math. 43(2), 757–831
(1942). MR0007108
23. [pseudonym], H.: Formal sum. Stackexchange (2017). https://math.stackexchange.com/
questions/2308741/what-is-the-meaning-of-formal-in-math-speak
24. MacLane, S., Birkhoff, G.: Algebra. The Macmillan Co., New York (1967). xix+598 pp
25. MacLane, S., Birkhoff, G.: A Survey of Modern Algebra. The Macmillan Co., New York
(1941). xi+450 pp. MR0005093
26. Rotman, J.: The Theory of Groups. An introduction. 4th edn. Springer, New York (1965, 1995).
xvi+513 pp. ISBN: 0-387-94285-8, MR1307623
27. Alexandrov, P.: Combinatorial Topology. Graylock Press, Baltimore (1956). xvi+244 pp. ISBN:
0-486-40179-0
28. Clifford, A., Preston, G.: The Algebtaic Theory of Semigroups, Vol. 1. American Mathematical
Society, Providence (1961). Mathematical Surveys No. 7
410 Glossary
29. Peters, J., Öztürk, M., Uçkun, M.: Klee-Phelps convex groupoids. arXiv 1411(0934), 1–5
(2014). Published in Mathematica Slovaca 67 (2017), no. 2.397-400
30. Hausdorff, F.: Set Theory, trans. by J.R. Aumann. AMS Chelsea Publishing, Providence (1957).
352 pp
31. Hausdorff, F.: Grundzüge der Mengenlehre. Veit and Company, Leipzig (1914). Viii + 476 pp
32. Krantz, S.: A Guide to Topology. The Mathematical Association of America, Washington,
D.C. (2009). ix + 107pp, The Dolciani Mathematical Expositions, 40. MAA Guides, 4, ISBN:
978-0-88385-346-7, MR2526439
33. Tozzi, A., Peters, J.: Topology of black holes horizons. Emerg. Sci. J. 3(2), 58–63 (2019).
http://dx.doi.org/10.28991/esj-2019-01169
34. Edelsbrunner, H., Harer, J.: Computational Topology. An Introduction. American Mathematical
Society, Providence (2010). xii+241 pp. ISBN: 978-0-8218-4925-5, MR2572029
35. Munkres, J.: Elements of Algebraic Topology, 2nd edn. Perseus Publishing, Cambridge (1984).
ix+484 pp. ISBN: 0-201-04586-9, MR0755006
36. Herstein, I.: Topics in Algebra, 2nd edn. Xerox College Publishing, Lexington (1975). Xi+388
pp. MR0356988; 1st edn. in 1964, MR0171801 (detailed review)
37. Krantz, S.: Essentials of Topology with Applications. CRC Press, Boca Raton (2010). Xvi+404
pp. ISBN: 978-1-4200-8974-5, MR2554895
38. Weisstein, E.: Homomorphism. Wolfram MathWorld (2017). http://mathworld.wolfram.com/
Homomorphism.html
39. Klette, R., Rosenfeld, A.: Digital Geometry. Geometric Methods for Digital Picture Analysis.
Morgan-Kaufmann Publishers, Amsterdam (2004). MR2095127
40. Ilieva, J., Zlatev, Z., Yordanova, R.: Application of Fibonacci series in computer-generated
patterns for contemporary textiles. Int. J. Textile Sci. Eng. TSE-109, 1–7 (2018). https://doi.
org/10.29011/IJTSE-109.100009
41. Huygens, C.: Treatise on light. In: Which Are Explained the Causes of that Which Occurs
in Reflexion and Refraction and Particularly in the Strange Refraction of Iceland Crystal.
University of Chicago Press, Chicago (1690 (1912)). xi+129 pp. translated by S.P. Thompson
42. Oudet, X.: Light as flow of photons. Technical report, Université Paris-Sud (2018). https://
www.researchgate.net/profile/Xavier_Oudet
43. Young, T.: The bakerian lecture: On the theory of light and colours. Philos. Trans. R. Soc.
Lond. 92, 12–48 (1802). https://www.jstor.org/stable/107113
44. Dennis, M.: Topological singularities in wave fields. Ph.D. thesis, University of Bristol, Depart-
ment of Physics and Astronomy (2001). Supervisor: M. Berry, 226pp
45. Newton, I.: Opticks: or, A Treatise of the Reflexions, Refractions, Inflexions and Colours of
Light. Also Two Treatises of the Species and Magniture of Curvilinear Figures. S. Smith, and
B. Walford, Printers to the Royal Society, London, UK (1704). 211pp
46. Saji, K.: Criteria for singularities of smooth maps from the plane into the plane and their
applications. Hiroshima Math. J. 40(2), 229–239 (2010). MR2680658
47. Nishimura, T.: Topological equivalence of k-equivalent map germs. J. Lond. Math. Soc. 60(1),
308–320 (1999). MR1722153
48. Seade, J.: Remarks on the topology of real and complex analytic map-germs. In: Singularities
and Computer Algebra, pp. 257–273 (2017). MR3675730
49. Fermi, E.: Notes on Quantum Mechanics. A Course Given by Enrico Fermi at the University of
Chicago. The University of Chicago Press, Chicago (1961). vii+188 pp. ISBN 0-226-24181-8
50. Peters, J.: Computational Proximity. Excursions in the Topology of Digital Images. Intelligent
Systems Reference Library, vol. 102 (2016). Xxviii + 433pp. https://doi.org/10.1007/978-3-
319-30262-1, MR3727129 and Zbl 1382.68008
51. Peters, J., İnan, E.: Strongly proximal edelsbrunner-harer nerves. Proc. Jangjeon Math. Soc.
19(3), 563–582 (2016). MR3618825
52. Bredon, G.: Topology and Geometry. Springer, New York (1997). Xiv+557 pp. ISBN: 0-387-
97926-3, MR1700700
Glossary 411
53. Peters, J.: Foundations of Computer Vision. Computational Geometry, Visual Image Structures
and Object Shape Detection, Intelligent Systems Reference Library, vol. 124. Springer Inter-
national Publishing, Switzerland (2017). i-xvii, 432 pp. https://doi.org/10.1007/978-3-319-
52483-2, Zbl 06882588 and MR3768717
54. Susskind, L., Friedman, A.: Quantum Mechanics. The Theoretical Minimum. Penguin Books,
UK (2014). xx+364 pp. ISBN: 978-0-141-977812
55. Zizzi, P.: Entangled spacetime. Mod. Phys. Lett. A 33(29), 1–21 (2018). https://doi.org/10.
1142/S217732318501687
56. May, J.: Finite Spaces and Larger Contexts. University of Chicago, Chicago (2003). http://
math.uchicago.edu/~may/REU2017/FiniteAugBOOK.pdf
57. Cottet, G.H., Koumoutsakos, P.: Vortex Methods. Theory and Practice. Cambridge University
Press, Cambridge (2000). xiv+313 pp. ISBN: 0-521-62186-0, MR1755095
58. Kamandi, M., Albooyey, M., Veysi, M., Rajaei, M., Zeng, J., Wickramasinghe, K., Capolino,
F.: Unscrambling structured chirality with structured light at nanoscale using photo-induced
force, 1–19 (2018)
59. Jänich, K.: Topology. With a chapter by T. Bröcker. Translated from the German by Silvio
Levy. Springer, New York (1984). ix+192 pp. ISBN: 0-387-90892-7 54-01, MR0734483
60. Nye, J.: Rainbows from ellipsoidal water drops. Proc. R. Soc.: Math. Phys. Sci. 438(1903),
397–417 (1992). https://www.jstor.org/stable/52118
61. Nye, J.: Wave dislocations in the diffraction pattern of a higher-order optical catastrophe. J.
Opt. 12(015702), 1–10 (2010). https://doi.org/10.1088/2040-8978/12/1/015702
Author Index
G
Gairola, U.C., 36 I
Galton, F., 220 Ibrahim, S.N., 387
Gellert, W., 146, 277 Illinois, University, 390
Gerla, G., xiv, 227 İnan, E., xiv, 81
Gersho, A., 77 Ion, A., 149
Ghrist, E., 82
Ghrist, R., 123
Ghrist, R.W., 31, 40, 114, 296 J
Giblin, P., x, 123, 160, 180, 196, 217, 384, Jack, B., 266
391, 392, 394, 402 Janeczko, S., 196
Author Index 415
Jänich, K., 19, 28, 72, 78, 113, 123, 407 Molina, M., 262
Jaquette, J., 28 Mooney, P., 387
Jeffs, R.A., 218 Morgan, F., 341
Morgan, S.P., 387
Morris, W.D., 14
K Moschovakis, Y., 154
Kaczynski, T., xi, 13, 17, 52, 82 Mrozek, M., xi, 13, 17, 52, 82
Kamandi, M., 407 Mughal, A., 23
Karimi, M., xiv Mumford, J., 219
Kästner, H., 277 Munch, E., ix
Kettunen, L., 283 Munch, R.W., 82
Khaldi, B., 379 Munkres, J.R., 20, 217, 394
Kherfi, M.L., 379
King, R., 266
Kirkby, W., 219 N
Klette, R., 395, 400 Naimpally, S., 266
Kopf, N., 57 Naimpally, S.A., xiv, 7, 81, 91, 227, 266
Koumoutsakos, P.D., 92, 102, 104, 407 Nakahara, M., 3
Kpalma, K., 31 Newton, I., 396
Kramár, M., 28 Niedermann, B., 81
Krantz, S.G., 91, 217, 393, 394 Nishimura, T., 397
Kropatsch, W.G., 149 Novik, I., 218
Küstner, H., 277 Nye, J.F., ix, 171, 172, 196, 208, 267, 299,
390, 403, 407
L
Larman, D.G., 14 O
Leader, S., 151, 266, 272 O’Dell, D.H.J., 219
Lebesgue, H., 341 Orłwoska, E., xiv
Lewis, G.N., 368 Ostavari, M., 36
Libertiny, T., 23 Oudet, X., 88, 396
Lienhardt, P., 389 Oudot, S., 3
Li, J., 262 Overmars, M., 81
Litchinitser, N.M., 32 Özturk, M.A., xiv, 392
Li, X., 109
Livingston, W., 171
Lockery, D., xiv P
Longhi, S., 209 Padgett, M., 266
Lowe, D., 25 Pal, S.K., xiv
Lu, J., 262 Pareek, C., 227
Lynch, D.K., 171 Pareek, C.M., 266
Pawlak, Z., xiv
Pedrycz, W., xiv
M Pellikka, M., 281, 283
Mackenroth, F., 266 Peltier, S., 149, 389
MacLane, S., 391 Peters, J.F., ix, 3, 7, 12, 13, 30–32, 72, 78,
Marino, F., xiv, 81 81, 82, 91, 113, 114, 123, 177, 180,
May, J.P., 405 181, 217, 220, 227, 245, 266, 379,
McAllister, B.L., 149, 153 383, 392, 393, 401
McBride, R., xiv Pham, D., xiv
Meghdadi, A.H., xiv Pham, H.D., 189
Meijer, D.K.F., 177 Pike, O., 266
Milnor, J., 377 Planck, M., 320
Mischaikov, K., xi, 13, 17, 52, 82 Poincar’e, H., 7
416 Author Index
example, 200 C
See Section 4.8, 199 Camera image, 180
Barycentric star nerve, 187, 189, 199 symmetry group, 180
application, 189 Caustic, 196, 209
defined, 187, 199 coffee-cup, 196
example, 188 defined, 196, 209
See Section 4.2, 199 optical, 196
Barycentric subdivision, 161 Caustic fold, 209
definition, 161 analogue, 209
Barycentric triangle, 187, 199 Cell, 51, 148, 385
defined, 187, 199 1.5-cell, 51
2-cell, 51
Betti number, 122, 170, 177, 179, 215, 283,
3-cell, 51
287, 296, 326, 374, 384
defined, 385
application, 283, 374
examples, 385
application No. 1, 326
vertices, edges, faces, 148
barcode, 122, 296
Cell complex, xi, 3, 5, 7, 28, 29, 67, 72, 78,
cusp nerve system, 374 87, 113, 126, 130
defined, 170, 384 Alexandroff–Hopf, 72
example, 172, 177, 215, 287 basic idea, 3
history, 179 connected cells, 5
Obs. No. A.2, 384 cycle, 130
optical vortex nerve, 215, 296 defined, 5, 87, 126, 130
persistence, 283 definition, 7, 28, 78
rank of a free Abelian group, 179 examples, xi
rank of the pth homology group, 179 introduction, 78
Section. 7.3, 384 known geometries, 5
Theorem 4.26, 215 natural approach, 3
Betti-Nye optical vortex nerve, 172 oriented, 113
defined, 172 properties, 78
Betti-Nye vortex nerve, 172 shape persistence fingerprint:Sec. 1.12,
example, 172 29
finite free Abelian group, 172 studies, 113
Binary operation, 153 topology, 73
definition, 153 triangulation, 67
Binary relation, 384, 385 two conditions, 73
defined, 384 Cell complex topology, 73
Alexandroff -Hopf , 73
on a video, 385
defined, 73
Boltyanskiĭ–Efremovich trail, 142
Topologie der Komplexe, 73
defined, 142
two conditions, 73
Boundary, 148, 162
Cell decomposition, 407
definition, 148, 162 definition, 407
Boundary chain, 162 Cellular geodesic line, 143
Cn = Cn (K ), 162 definition, 143
defined, 162 Centroid, 19, 25, 27, 404
definition, 162 Alexandroff nerve vertex, 404
Boundary group, 165 Centroid-based tessellation of a video
definition, 165 frame: Alg. No. 2, 27
Boundary of a hole, 162 definition, 25
definition, 162 deformation retract, 19
Boundary of a planar polygon, 12 2D region, 25
definition, 12 3D region, 25
Subject Index 425
⩕
conn
examples, 25, 27 Class of δ , δ||Φ|| -based cusp nerve sys-
sec: A.18, 404
Chain, 125, 162, 165, 386 tem shapes, 361
⩕
⩕
Bn boundary chain, 165 conn δ||Φ||
δ , δ||Φ|| -based clscuspNr vShapeSys
boundary, 165
briefly explained, 125 class construction, 361
1-chain, 125, 386 defined, 361
⩕
complex, 386 Class of δΦ -based shapes, 284
cycle, 165 ⩕
defined, 162, 386 δΦ class construction, 284
of edges, 386 defined, 284
Z n cycle chain, 165 Class of shapes
⩕
Chain complex, 149, 393 δ
||Φ||
clscuspNr vShapeSys class construction,
definition, 149
361
Chain group, 163 ⩕
δ
||Φ||
(Cn , +), 163 clscuspNr vShape class construction, 357
definition, 163 ⩕
Chain map, 387, 393 δΦ class construction, 285
Fundamental Theorem, 387 Clock addition, 388
Class, 271, 272, 283–285, 288, 324, 357, analogy, 388
361, 374 Closed 1-cell, 126
δΦ , 283 defined, 126
⩕ Closed cell complex, 70
conn ⩕ δ||Φ||
δ , δ||Φ|| -based clscuspNr vShapeSys , defined, 70
Closed half plane, 16, 51
361 definition, 16, 51
⩕
δ
||Φ|| Closure, 64
clscuspNr vShapeSys , 374
⩕ defined, 64
δ
||Φ|| Closure of a nonempty set, 217
clscuspNr vShapeSys class construction, Closure of a set, 341
361
⩕ defined, 217
δ
||Φ||
clscuspNr example, 341
vShape class construction, 357
clsshapeδΦ , 271 proximity-based, 217
⩕ Sec. 4.14, 217
δΦ -based, 288 Cluster, 273
C ⩕ (skcyclic NrvE), 288 defined, 273
δΦ
⩕
Leader, 273
Φδ Coffee cup caustic, 171, 172, 208
clsshape , 284
⩕
chain-of-quanta view, 172
Φδ cusp, 208
clsshape class construction, 285
⩕
defined, 171, 208
δ
clsshape , 271 example, 207
⩕ fold caustic, 208
δ
||Φ||
cls Shape shape class, 324 geometric rendition, 209
descriptive, 272 J.F. Nye, 208
descriptively near shapes class See Section 4.11, 208
clsshapeδΦ , 271 Colour pixel, 395
optical vortex nerve, 288 corresponding wavelength example, 395
spatial, 272 hue angle, 395
⩕ Complex, xi, 180
δ
strongly near shapes class clsshape , 271 covering, xi
Classical general topology, 8 CW complex, 180
definition, 8 finite, 180
426 Subject Index
Cusp filament, 209, 211, 220, 279, 397, 400, simple closed path, 130
401, 404 Cycle chain complex Z n , 165
•——•, 279 ∂Cn = 0, 165
angle between cusp filament vertexes, Zn , 165
279 definition, 165
angle between vectors, 277 Cyclic, 149
angle example, Sec. 6.3, 279 chain-wise connected, 149
angle, Sec. 6.3, 277 Cyclic chain-wise connected, 149
defined, 209, 220 definition, 149
Observation No. 4.11 in Sec. 4.11, 279 Cyclic filament skeleton, 209
pathway for reflected light, Sec. 4.12, 277 example, 209
photon spin qubit, 404 innermost, 209
qubit: Appendix A, Sec. A.15, 401 outermost, 209
reflected light path, 404 Sec. 4.11, 209
reflected light pathway, 211 Cyclic group, x, 123, 156, 180, 193, 389
Sec. 6.3, 279 +mod2,
124
Section 4.12, 400 arcs, +, 157
simplest structure, 279 Alexandroff definition, 180
vector, 401 defined, x, 389
vertex, 277 definition, 123, 156
waveform of light: Obs. No. 8, 397 example, 389
Cusp nerve, 400 generator, 156, 193
origin of optical cusp nerves: Obs. No. geometric realization, 124
A.13, 400 integers, +, 156
CW complex, 69, 73, 91, 113, 180
containment condition, Sec. 1.27, 69
defined, 92 D
intersection condition, Sec. 1.27, 70 Delaunay nerve
Alexandroff–Hopf conditions, 73 barcode, 56
closure finite property, 91 bc M N C , 56
Containment condition, 73 Delaunay Triangle, 36, 37
Intersection condition, 73 boundary seed points, 37
introduction, 180 construction, 36
origin, 69 definition, 36
path-connected, 180 example, 37
studies, 113 polytope, 37
weak topology property, 91 Delaunay triangle, 40, 51, 79, 80
CW topology, 8, 131 barcode, 40
briefly explained, 131 barycenter, 39, 40, 79
closure finiteness, 72 cell
CW, 131 1.5-cell, 51
defined, 72 2-cell, 51
definition, 8 centroidal, 40
J.H.C. Whitehead, 72 construction, 80
weak topology, 72 Cover property, 52
CWΦ complex, 276 definition, 51
Cycle, 112, 130, 132, 149 distinguished point, 79
cyc A, 132 Edge property, 51
defined, 132 intersecting closed half planes, 51
definition, 149 seed point vertices, 79
homologous, 149 Delaunay triangulation, 51, 67, 69, 72, 79,
oriented, 112 80
path-connected, 112 1.5-cells, 51
428 Subject Index
2-cells, 51 definition, 18
2.5-cells, 51 Edge, 390
3-cells, 51 Edge path, 17, 53, 92
Closure Finiteness property, 67 barcode, 53
Cover property, 52 defined, 92
CW topology, 79 definition, 17
definition, 51 length, 53
edge connected, 52 Edge path length, 53
Edge-connected property, 52 definition, 53
edge path, 52 Edges, 149
Edge property, 51 directed, 149
Nerve property, 55 oriented, 149
path-connected, 79 Electric field, 176, 177
Weak Topology property, 69, 72 equipotential lines, 176
Descriptive CW complex, 276 free Abelian group, 177
example, 276 non-concentric, nesting circles, 177
Descriptive nearness Electron, 266, 379
A δΦ B, 244 charge, 266
Descriptive proximity, 226 mass, 266
example, 226 spherical model, 379
Descriptive shape class clsδshape
Φ
, 272 Elementary shape, 146
defined, 272 definition, 146
example, 272 Energy, x, 334, 369, 377, 390
Diffraction, 390 cusp nerve system, 369
defined, 390 cusp nerve system: Sec. 8.12, 368
example, 390 cusp nerve system shape: Sec. 8.12, 369
patterns, 390 filament kinetic, 390
Digital image, 124 filament kinetic. See, also:
defined, 124 Appendix A.5, 334
picture, 124 kinetic, 377, 390
Digital image hole, 148 Milnor path energy, 390
definition, 148 Milnor path, Sec. 8.15, 377
interior, 149 nerve system, 390
shape, 149 photon, 390
uniform intensity, 149 shape kinetic, 390
Digital image shape, 148 shape shifter, 390
definition, 148 shape shifting, x, 390
Distinguished point, 19 surface shape, 369
defined, 19
Distinguished point:Voronoï region interior
seed point, 16 F
Dot product, 278 Fat line, 161
p · q, 278 definition, 161
example, Sec. 6.3, 278 Fat line segment, 161
inner product, 278 definition, 161
Drone video frame, 9 Filament, 92, 193, 209
MNC, 9 1-cell, 220
polytope, 9 cusp, 209, 220
tessellated, 9 defined, 92
example, 209
length, 193
E Sec. 4.11, 209
Edelsbrunner–Harer nerve, 18 vortex, 92
Subject Index 429
Shape of interest, 199 Skeletal vortex, 104, 106, 112, 167, 168, 170,
defined, 199 176, 178
MNC, 199 defined, 104
See Sec. 4.7, 199 example, 105, 167
Shape persistence, 28, 40 free Abelian group, 168
defined, 40 intersection filament skeletons, 178
definition, 28 largescale, 176
example, 40 modeling behavior of light waves
fingerprint, 29 reflected off surfaces, 106
Shape signature, 31 oriented, 112
definition, 31 revisited, 168
Ghrist barcode, 31 see Sec. 3.21, 168
Shape space, 117 Skeleton, 3, 5, 28, 29, 90, 113, 193, 342
definition, 117 charged with physical meaning, 5
digital image, 117 cyclic, 342
Shape wearout, 122 cyclic group representation, 193
defined, 122 defined, 90
Simple closed curve, 146 definition, 28
definition, 146 filaments, 5
extA, 146 fractional dimension , 29
exterior region, 146 n-skeleton, 91
intA, 146 oriented, 113
interior region, 146 replicas of 3D tubes, 5
point, 147 shape closed boundary, 3
Simple closed surface, 147, 165 simple, 342
Simplex, xi Space, 28, 89, 117
planar, xi cellular, 28
Simplicial ball, 136 contractible, 28
Br (c), 136 defined, 89
cl(Br (c)), 137 definition, 28, 117
closed, 137 Hausdorff, 89
defined, 136 topological, 28
open, 136 Spatial shape class, 273
Simplicial complex, 124 defined, 273
⩕
covering, 124 δ
Spatial shape class clsshape , 272
defined, 124 defined, 272
Singularity, 406 Spin behaviour, 128
Skeletal nerve, 94, 101, 102, 106, 168, 179, defined, 128
190, 192, 200 Split feasibility, 58
barycentric, 200 Close Alexandroff Nerve Shapes on Tri-
Betti number, 179 angulated Video Frames, 58
collision of skeletal vortexes, 106 Split feasibility problem, 57
defined, 94 Alexandroff nerves on video frames, 58
energy, 101 defined, 57
example, 94, 106, 179 Spoke, 6, 46, 48, 210, 400, 406
free Abelian group, 168, 179 skA, 48
hybrid, 102 ◦–◦, 210
In spacetime, 101 construction, 48
proximity, 102 defined, 6
revisited, 168 defined: Appendix A.18, 406
see Sec. 3.21, 168 Delaunay nerve, 50
See Sec. 4.8, 200 example, 46
simplest form, 192 leaf, 6
438 Subject Index
MNC, 50 Voronoï, 7, 78
nerve, 48 Voronoï diagram, 9
nucleus polygon, 6 Theorem, 2, 21, 35, 90, 110, 111, 147, 188,
persistence, 50 212, 213
winding, 50 Ahmad Betti Number Measure, 212, 213
Strong Gemini complex, 108, 109 barycentric star nerve, 188
defined, 108 Coffee Cup Caustic Cusp Filament, 213
example, 109 connected space, 229
Gemini Nerve, 111 contour cyclic group, 158
skeletal nerve, 111 Di Concilio–Guadagni–Peters, 147
Surface, 88, 196, 338 finite planar vortex complex, 2
bounded, 196 Fundamental optical vortex nerve theo-
defined, 338 rem, 213
hole, 88 Jordan Curve Theorem, 2
Martian, 88 nerve, 21
plane, 338 Optical Vortex Nerve Free Abelian
projections, 196 Group Representation, 213
reflections of light, 196 sec. 4.13, 213
shape, 88, 338 Shape shifting mass, 399
Surface hole, 12, 13 skeletal nerve, 111
defined, 12, 88 skeleton cyclic group, 159
physical surface, 13 strong Gemini complex, 110
Surface shape, 13, 88 2D physical shape boundary, 149
defined, 13, 88 video frame, 90
Vortex cycle contractible, 35
Vortex cycle retract, 35
T Tile, 6, 339
Table, 3, 87, 117, 186, 223, 273, 337, 357 huge‘, 6
Additional Computational Geometry and defined, 339
Topology Symbols, 87 examples, 339
Nerve Complexes and Their Symbols, Tiling, 6, 8, 78, 81, 339, 341
186 carta marmorizzata, 81
Optical Cusp Nerve Complexes and cells, 78
Their Symbols, 337 defined, 339
proximites and their symbols, 223 definition, 8
Proximity-Based Cusp Nerve Shape disegno cachemire, 81
Classes and Their Symbols, 357 example, 6, 81
Proximity-Based Shape Classes and known shapes, 81
Their Symbols, 273 proximity function, 81
Shape Complex, Skeleton and Other space-filling shapes, 81
Useful Symbols, 117 sufficiently small, 341
Some Computational Geometry and Theorem 8.5, 341
Topology Symbols, 3 Topology, xi, 5, 151, 406
Tessellation, 5–7, 9, 14, 24, 78 computational, 5
defined, 6 construction, 151
definition, 9, 14 Leader uniform, 151
Descartes, 7 Topology of complexes, 78
example, 5 introduction, 78
maximal nucleus cluster, 9 Trail, 141, 142
nucleus, 6 Boltyanskiĭ–Efremovich, 142
nucleus polygon, 9 definition, 141
See good tessellation, 24 Triangle, 187
video frame, 9 barycenter, 187
Subject Index 439
W Z
Water, 40 Zero-shot recognition, 262
detected by MARSIS, 40 application, 262
Martian, 40 application No. 5.13, 262
Wave function, 299 defined, 262