Anamorphic images JL. Hunt, 8. G. Nickel, and Christian Gigauit Department of Physics, University of Guelph, Guelph, Ontario NIG 2W1, Canada (Received 2 April 1999; accepted 10 August 1999) ‘Anamorphic images are images of objects which have been distorted in some way so that only by ‘viewing them from some particutar direction or in some particular optical surface do they become recognizable. Anists have been fascinated with these transformations since the 16th century, but there seems to be no modem explication of the mathematics of these transforms. In this paper we describe the most common of the anamorphic images found in art and derive the transform ‘equations for plane, conical, and c; indrical cases. With these equations it is possible to analyze carly anamorphs, and with computation to create modern ones with ease. © 2000 American Assocation of Physics Teachers I. INTRODUCTION ‘The revolution in art in the 16th century was accompanied by a rediscovery of perspective. The artists, mathematicians, and philosophers of the Renaissance studied and exploited perspective, sometimes to the extent of obsession, as they saw it bearing on the nature of illusion, truth, and reality, Prodigious essays in perspective were executed resulting in startling trompe-t'oeil decorations such as the Colonna and. Mitelli false architecture of 1616 in the Ducal Palace, Sas- suole. A critical essay, with copious illustration of the devel- ‘opment, of perspective, can be found in the article by Cleric." Iis only a small further step to experiment with the rules of perspective and to carry them to extremes, producing distorted images which appear normal only when observed from a particular point of view, or via some optical device. Such images are generally known as “‘anamorphic images."" Some artist-mathematicians produced treatises on the methods for producing these images, the most notable being Jean-Francois Niceron’ who provided geometrical construc- tions for making many types of anamorphic transforms. Some of these methods are exact, while others are clearly approximations. The most extensive modem treatment of the history of anamorphic art is the monograph by Baltrusaitis.” An exh bition in Amsterdam in 1975 prompted a book containing several high-quality color reproductions by Leeman.‘ Earlier brief articles by Gardner’ and Walker® cover much the same ‘ground but from a more mathematical point of view. Walker deals explicitly with the geometrical optics of the anamor- phic transformations and provides techniques devised by David G. Stork for producing anamorphic images photo- graphically. Familiar examples of anamorphic transformations are map projections and the images in ‘“fun-house”” mirrors. In this paper we will present the transformations for the “plane,"" “‘conical,”” and “cylindrical’” anamorphs which are the most common in anamorphic art. Il. PLANE ANAMORPHS: In these images a planar figure is distorted, mostly in one ‘dimension, So that it appears normal (or a hidden image ap- pears) only when viewed from a particular direction— usually very close to the plane of the drawing. By far the ‘most famous of these is the painting ““The Ambassadors” (1533) by Hans Holbein which hangs in the National Gallery in London.’ This painting of two prosperous men standing 232 Am J. Phys. 68 (3), March 2000 ‘on either side of a curio cabinet is a virwoso exercise in perspective. The cabinet is filed with objects wich were the standard subjects for artists 10 practice in their sketch- bbooks—spheres, polyhedra, a lute, etc., and the tessellated floor contains a variety of geometric shapes. However, spread across the lower portion of the painting is a gray smear which startlingly resolves itself into the image of @ skull when the viewer stands on the right-hand side and glances down at an elevation of about 20° above the surface. This memento mori may also have been intended, by the anist, asa rubric on his name, a5 “‘hoble Bein"’ means “*hol- low bone.”” A well-known Vesierbild (puzzle picture) by Eshard Schén (c. 1535) is shown in the left of Fig. 1.® The woodcut looks like a scene in which familiar objects are presented as pethaps in a nightmare landscape. When viewed from the left edge at an elevation of about 10° above the surface, the portraits of Charles V, Ferdinand 1, Pope Paul III, and Fran- cis are seen, as showin in the reconstruction atthe right in the figure. How was such 2 picture produced? Was it drawn by viewing the surface from that angle or was a formule of some kind employed? We know that both methods were used. What then are the transforms that produce these draw- ings? “The simplest transform is shown in the example of Fi. 2 by Samuel Marolois (1614), In this case only one dimension is distorted and so the correct position from which to view the reconstruction is from infinity, The transformation equa- tions are trivial in this ease: x'sx, and y'=y/sina, ) where the primes represent the transformed coordinates and ais the angle of elevation above the plane. Measurements on the drawing indicate that the angle is 13°, "Anamorphs of this type are seen by almost everyone every day in the symbols and drawings painted on roadways. The distorted figures appear undistorted to the car driver or bi- cycle rider as shown in Fig. 3." ‘A more realistic situation for smaller scale images is one ‘where the anamorph is viewed at an elevation angle a at a finite distance which we choose to be d, at height h from the center of the leading edge of the anamorph as shown in the top of Fig. 4. The transformation is found by mapping points xy in a “target” grid to points x',y’ on the anamorphic plane, We take the orientation of the target plane to be nor- ‘mal to the origin-line-of-sight, VO. From the figure, (© 2000 American Association of Physics Teachers 232Fig. 1. Plane anamorph by Erhard chin (c 1535). At the ight the transformed image cbiaine’ by looking fom te left at an elevation of 1" above the anamorptic plane iasy and tan B= y/(a?-+h?) ‘Using the relation for the tangent of a difference, and writing the functions in terms of d, h, and a, we readily obtain for the y transformation and its inverse tan y=tan(a—B) yisin Y= Griese 8d Y= y' sine > Poder ° For each value of y. the x transformation follows from the lower part of Fig. 4 which is the (almost) horizonta? plane including the line VAB, Writing «/VA=x'/VB in terms of d hy, and 5" gives x) ie (dey aa) Ey @) Equations (2) and (3) have Eqs. (1) as a limit as d and h go to infinity with a constant ratio hid=tan a. ‘This transform is depicted in Fig. 5, originally drawn by Niceron.!"" The square grid (enlarged) is mapped onto the trapezoidal grid, The rest of the construction is Niceron’s eomeuical equivalent of Eqs. (2) and (3) and is given in derail by Gardner.’ Analysis of the figure shows that itis drawn for a= 19° and a™=26 times the untransformed grid element Fig, 2 Plane sramomph by Marlo 1624) constcted for viewing from Init. The fgue sto Ge viewed at an elevation angle of 13" 233m. J. Phys, Vol 68, No.3, March 2000 lane symbol designed fr viewing at about 20". Right: View ofthe symbol at abou the proper enle INI. CONICAL ANAMORPHS Anamorphic images which require an optical reflecting surface for reconstruction have, over the centuries, proven of ‘more interest to both artist and viewer. The simplest is the conical reflector where the image is viewed from directly above the cone apex as shown ia Fig. 6. The cone is charac- terized by an apex angle 6, and a base radius taken as unity so that the cone height is = cot | é. Clearly, the transforma: tion has circular symmetry, so the target is taken to be a series of concentric rings of radius r which transform to rings of radius r' when reflected off the conical surface For the view from infinity (@ not unreasonable assumption in practice) the transform is the strictly linear (a) More generally, the situation is as depicted in Fig. 6, where the anamorph is viewed from a point V at height d above the apex. From the figure, 1 "= AR+(h—a)tan (-a)tan( @~ a) and r_ ah tne" TER dea ‘The last line can be used to eliminate both a and a and we find Fig 4. The gomeury of the plane aramorph viewed from afte distance ont, Neko, and Gignult 233,ie ooo. Fig, 5. Geometrical soltion to the plane ansmorph viewad fom a Fite distance by JF. Niceron (1638) ar (a+ hXL aa aT ‘The limit of this as d goes to infinity is Ea. (4). ‘An example of this transformation is shown in Fig, which is again a construction by Niceron.? An analysis of the figure shows that the cone angle is 60°, a natural choice for a real situation, By fitting the measured values of r' itis fur- ther possible to show that the situation is not drawn for d =e, but rather that d-+h~5. The measured values of 7’ show some scatter and itis not unreasonable to speculate that the tansformed rings were positioned experimentally by viewing a target of equally spared circles in the cone about 5 radius units above its base. ‘Niceron also provided templates for making anamorphs by looking into a mirror from the side, but the templates, being circles in a sector, are certainly approximations." Such an- amovphs are rare, IV. CYLINDRICAL ANAMORPHS ‘The anamorphs which have retained their interest for the longest time have been those which are reconstructed by ob- Fi. 6 The geomety ofthe conical anamorph viewed from rite distance above me base. 234 Am J. Phys. Vol 68, No. 3, March 2000 Fig, 7. Geometrical mamix tor the consti of conical anamoephs by oF, Nieron (1638) serving them ina cylindrical mirror standing on the anamor- phic plane. From objects of mystery in the 16th century to a rich man’s possession in the 17th and 18th to children’s toys. in the 19th century," the cylindrical anamorph has long history, ‘The situation is depicted in Fig. 8. The cylinder with unit radius stands on the anamorphic plane and the transforma- tion is determined by imagining rays from the point P” in the ‘anamorphic plane striking the mirror at R and being reflected to the point of observation V. The virtual extension of the ray VR passes through the anamorphic plane at P’. The left part of Fig, 8 shows the projection of the geometry onto the horizontal anamorphic plane, and on the right is a projection onto a vertical plane through the cylinder axis and V. ‘We take the origin of coordinates as the center of the base of the cylinder and V at an elevation h and distance d in Fig. & The gcomeuy of the cylindrical anamorph viewed from a finite mance at an elevation angle @ nt, Nickel and Gigault 234Fig. 9. Right The cylindrically anamorphized contours of q 10X20 grid of squares viewed at d= 15,410. Left The same viewed fm infinity ata clevation angle which case the geometry of the ray VP" is exactly that shown in Fig, 4, As a consequence we now can take the total ‘anamorphic transform to P" as the combination of the trans- formation Eqs, (2) and (3) plus that generated by the rotation of the ray segment RP’ to RP" by angle 7-26. ‘To obtain the latter transformation equations we start with x sing By esd as shown in Fig. 8, The second equality in this equation can be used to solve for yin terms of x',y', and d. We get, from the resultant quadratic in cos i, dx'2+ (d+ y' NEU =x) Bdy Fey (dty Px The point P” coordinates are in + RP” sin( b+ 8), y"=cos + RP" cos( +8), where the horizontal projection RP (cos #+y’Ve0s On combining the above equations to- gether with 6= v+ 4 we obtain in y+ (cos U+ y’ (sin 2y+cos 2uhtan d) cos d= © 2 H+ pyr loos w+ y" Videos WL) a cos U+ (cos + y")(cos 2W—sin 2 yan 6) fs (deo y'42.c05 Hoos W+y")) Tan) @ Since sin y=x'(d~c0s yyld'+y’), the second expressions in each of (7) and (8) make apparent the geometrical relation- ships (x”—x')/()"+y")=tan die, P’PY is parallel to OR ‘The combined Eqs. (2), (3) and (6)~(8) solve the cylindrical anamorph problem of mapping (x3) 0 (x",9"), The nature of these equations is illustrated in the right hand section of Fig. 9 which displays the transformation of 10%20 grid of squares on the interval O=x= 1, O=Sy=2 235 Am.J. Phys. Vol. 68, No.3. March 2000 with h= 10 and d= 15, ic., an elevation angle of 33.7°. Note that the contours are not circles and the reflecting cylinder is displaced from what would appear to be the “center” of the array. The curved contours acquire an increasingly wider spacing further ftom the mirror as a consequence of the finite viewing distance. In many, if not most, situations the view- ing distance can be taken to be infinity and a considerable simplification in the equations is realized. If h=dtan a, ‘where a is the elevation angle of the line-of-sight above the plane, then Eqs. (2), (3), and (6)~(8) become (9) r= 2(1-22)82+ (1-222) yra2(1aat)+ 2 (1-292) (19) [As an example ofthese equations, another 10% 20 grid with = 15x50 viewed from infinity sf a= 33.7" is shown in the lefehalf of Fig. &. The contours are not circles but they are equally spaced, The miror is sill not inthe center of the array. The contrast between the let and right sides of Fig. 9 iustrates thatthe edges of the cylinder are not lines of con- stant x except for viewing from infinity. At fnte distance the image of the cylinder in the eye isa trapezium, wider atthe top. which the brain insists on interpreting as a rectangle ‘This in tum means that. when viewing the anamorph a @ “picture framed by the eylinder, one is left with the im- pression that something is not quite right asthe picture e- Eedes from the cylinder edges with increasing y. To obtain an “anisically pleasing” transformation we must somehow compensate for the illusion thatthe edges of the eyinder appear as parallel lies. One possibility is t0 leave Eq. (2) and the corresponding target plane in Fig. 4 unchanged for Specifying the y coordinate, but replace Eq, (3) with xMdty)=xld, an which corresponds to specifying the x coordinate on a verti- cal target plane through the cylinder axis. By defining sepa- rate target planes we are clearly leaving the domain of per- spective drawing in the strict sense, but itis precisely such apparent inconsistencies that prevents a unique solution. ‘Niceron* has provided both an exact geometrical method for the construction of such anamorphs and an approxima- tion which is shown in Fig. 10.!° In the approximation the contours are circles"® with the spacing increasing with radial distance, indicating that it has been drawn for a finite view- ing distance. In addition, the position of the cylinder is dis- placed from the center of the array. Finally, the radial lines in the correct transformation (see Fig. 9) do not converge to a ‘common vertex whereas in the approximation they are taken to point back to the center of the displaced cylinder. The radial ines are drawn at equal angular separation, making the sectors in the forward direction larger than at the sides. From Fig. 9 it can be seen that this is exactly the opposite of the ‘cortect transformation. When the Niceron matrix is tested by transforming straight lines (a triangle, for example), the re- constructed image retains small curvatures. This failing ‘would not be so noticeable or important for a drawing or painting which was probably to be finished by direct obser- vation in the misror. ‘The forward transformation Eqs. (2), (3). and (6)~(8) are ‘appropriate if the target consists of simple elements, An ex- Hunt, Nickel and Giga 235SMMOMY Fig. 10. Geometric matrix for the constton of cylindrical anamorphs by IE. Niceron (1638) ample of such a case and the resulting reconstruction is shown in Fig. 11. In more complicated cases such a5 a igi- tized photograph itis preferable to have the inverse equa- tions. These are derived in the Appendix, and Fig. 12 is an example of an anamorphized photograph based on the in- verse Eqs. (A1)-(AS) supplemented with (2) and (11). \V. VIEWING THE IMAGE ‘There is an ambiguity involved in describing where the Virtual image in th eylindrcal mieror appears to be located ‘The sinuation is described in Fig. 13. The anamorphic ele~ ‘ment at distance s is imaged at 5 by the cylindrical mirror, 11.2 FUS"R Since 2=(s—s' tan a, then, 2s"? ) R-2| This image plane is shown in the figure and is esponsible for some of the fascination ofthese objects in the pas; the image jana, ay Fig. I. A simple namorph constructed with Eas. (2), (3), and (6)(8), 236 Am.J. Phys, Vol. 68, No, 3, Mach 2000 Fig. 12. An anamorphized photograph using Eqs. (AI)-(AS), (2), and (1D, seems to stand up in the mirror and have a certain three- dimensional air about it” There is, however, another possibility: Considered strictly as a plane mirror in the axial direction, then s=s' and z 0. In other words, the image will be seen to lie in the anamorphic plane, Is such an image ever seen? Almost all viewers see the cylindrical situation (Bq, (12)] when viewing the image normally, ic., with the interoeular axis perpendicular to the cylinder axis. It is, however, re- markable that when the head is tured to the side (interocular axis in the direction of the cylinder axis), then almost all ‘viewers we have sampled see the image lying in the anamor- phic plane. This illusion is very strong and viewing with one ‘eye closed does not seem to affect it! We are not aware that this illusion has been previously reported. APPENDIX: THE INVERSE TRANSFORMS FOR ‘THE CYLINDRICAL ANAMORPH The inverse mapping for the cylindsical anamorph has been referred to by Domrie!® as Alhazen’s Billiard Problem. We start by combining Eqs. (7) and (8) into the alternative pair, (x"~sin g)(cos 2 sin 2dtan §) =(y"~cos #)(sin2¢-+ cos 2ytan 6), (x"—sin ¥)?+(y"—cos $)?=(cos y+ y'P(1 + tan? 4). bE: —sbkeo Fig. 13. The geomety of the image plane fr the cylindvcalanamexph Hont, Nickel and Gignolt 236‘The first of these equations, tan d=sin Yé(d—cos ), reduces to (x"~sin W)(d cos24—cos ¥) (9" ees i(d sin2y—sin ), which is an equation for y in terms of x".y", and d, ‘The remaining equation gives y’ once wis known. Since cos y+’ is necessarily positive, cosy when used with d-cos VE = 2d cos +1 V(a"= Sin YG =c05 WF, (a) and, for completeness, from the discussion leading to Ea, (6), wa thiatyy (a2) cos ‘The equation for w given above simplifies considerably if we introduce polar coordinates r, 2¢ to replace x".y" and write qin terms of the deviation from the bisecting angle €, ie, d= €+é. With these changes the dequation becomes equivalent to sin 3cos 5=A sin 3+ B cos 3, where p[t+thcose, ae z(4~+)sine 2 \ ried) eee een? cd) ped x"=rsinde, y"=reos2e, (a3) ‘Since the equation for the coordinates x".y" define r and € and hence A and B, we are left with the equation for as the ‘only nontrivial relationship. This, in tur, is the quartic sin 3~2B sin? 5—(1—A?~ 82) sin? 5428 sin 3—B" which yields, asthe physical rolution, > (at ya?=28)" with (1/22, The solution of the inverse problem is now trivially completed with Eqs. (Al) and (A2) and sind inet 2). ‘The solution given in Eq, (AA) is numerically stable with |8|1 in the physically allowed region [2e! cos '(I/d)+cos™'(I/r). In the extended or *shadow"’ re 237 Am J. Phys, Vol. 68, No.3, March 2000, gion cos"(I/d)+cos™4I/r)<|2¢|<2, the solution pre- dicted by Eq, (A4) comesponds to rays penetrating the cyl inder and refectng off the inner face. Thete is no dramatic simplification that occurs in the inverse Eqs. (A1)~(A3) when d= in contrast to the forward transformation case fef. Eqs. () and (10)] It ig pethaps worth noting that none of the intermediate functions a, bv, oF w in Eqs. (Ad) and (AS) have any par- ticular symmetry under reflection ¢——€, and thus the sym- retry requirement that sin 5be an odd function of ¢ is not explicit This does not mean Eq, (Ad) is defective. On the contrary, Eq, (Ad) is our “best” choice as we have found that other representations of the solution, which do satisfy the reflection symmetry explicitly, are numerically less stable and/or more complicated. "Fabio Claris, “The Grand Hlsion: Some Considerations of Pespec= sve, Hlsionsm and Trompe eel” An News Aaa! XXIM, 96-180 (958) Sean-Franois Niceon, La perspective curieuse (Pas, 1638) orgs Baliains (wanted. by W. J. Suachan), Anamorphic Are (Abrams, New York, 1977 SF Leeman, Hidden Images: Games of Perception, Anamorphic Art Hl sion (Abrams, New York, 1976) SManin Gardner, "Mathematical Gams,” Sci, Am. 282 (I), 110-116 0975), east Walker, 98. “Figure p-125, 1S1(bw) of Ref 1: Chap. 7 (bw) of Ret 3 Plate 2, p18 (colo) of Ref. d, Fig. p12 (hw) of Re S. igutep-150 of Ref Is Fig 7 of Ret. 3: Fig 2, p12 of Ref. 4 Seite 24 of Ra. 3 The image onthe left in Fig, 3 could ot be photographed from drecy hove but fom te side a2 steep angle: thas ben comes gall for this anamorphic enor. Oneimesionalanzmorphs ae also the wa it which wide-sreen motion pictures are made. The camera imposes a com pression in the honzonal dlretion to St the wide image ota the fim and the projector's sramorphic lens reverses the compression. "opigie 28 of Ret. 3: Fig. 27, p-106 of Ret: Fig redrawn) pl of Ret 5, "apigure 118 of Ret. 3; Fig. p15 of Ref. 5 rigor 34 of Ret. 3 ‘yeh’ toy consisting of 24 colored drawings anda elindrcal mitor was published aboot 1900 by the MeL-oughlin Bros. of New York. It was republished as The Magic Mirror (Dover, New York, 1979). A more re- ent chil’s book is Anna's Mogical ABC. by Mitstmasa and Masahiro ‘Anvo (Bodley Head, Landoa, 1981) "opigue 115 of Ret. 3: Fig 35, 9-132 of Ref. 4 Fig. pl tS of Ref. 5. The base af the mir is the smallest cele centred at the convergence of te rail lines. When the curves ar iosarured they are not sel cles: the ri vary about IS. Me geometry ofthe drawing strongly sugges that they were onpinally cies. Various copying processes andthe diferent shrinking of almost four-cenury-old paper can well account fr the deviations Aanists may have acully taken advantage of the shape of this image surface. A pretty example on p. III of Ref. Sis an anamorph by an ‘unknown are of Nicolas Lane's Par ane Tendre Chansonnete(Wih 2 tender lial song). The reconstructed image of aman setenading 2 woman dnd her anendans ina pastoral seene has an almost horizontal wa athe lower portion, and the figures erect further in the tear The thre dlensional effet is very siking. It can be viewed wsing a clinical rior 43 mm in diameter, "enrich Dowie. A Hundred Great Problems of Elementary Mathematics: ‘heir Husory and Solution (ansaed by David Artin) (Dover, New York, 1965). “The Amateur Scientist" Sci, Am, 245 (I), 176-187 ont, Nickel, and Gigault 237