Professional Documents
Culture Documents
Image Analysis and Recognition 17th International Conference ICIAR 2020 Póvoa de Varzim Portugal June 24 26 2020 Proceedings Part II Aurélio Campilho
Image Analysis and Recognition 17th International Conference ICIAR 2020 Póvoa de Varzim Portugal June 24 26 2020 Proceedings Part II Aurélio Campilho
Image Analysis and Recognition 17th International Conference ICIAR 2020 Póvoa de Varzim Portugal June 24 26 2020 Proceedings Part II Aurélio Campilho
https://textbookfull.com/product/image-analysis-and-
recognition-17th-international-conference-iciar-2020-povoa-de-
varzim-portugal-june-24-26-2020-proceedings-part-i-aurelio-
campilho/
https://textbookfull.com/product/image-analysis-and-
recognition-11th-international-conference-iciar-2014-vilamoura-
portugal-october-22-24-2014-proceedings-part-ii-1st-edition-
aurelio-campilho/
https://textbookfull.com/product/image-analysis-and-
recognition-11th-international-conference-iciar-2014-vilamoura-
portugal-october-22-24-2014-proceedings-part-i-1st-edition-
aurelio-campilho/
https://textbookfull.com/product/hci-
international-2020-posters-22nd-international-conference-
hcii-2020-copenhagen-denmark-july-19-24-2020-proceedings-part-ii-
constantine-stephanidis/
https://textbookfull.com/product/graph-transformation-13th-
international-conference-icgt-2020-held-as-part-of-
staf-2020-bergen-norway-june-25-26-2020-proceedings-fabio-
Aurélio Campilho
Fakhri Karray
Zhou Wang (Eds.)
LNCS 12132
Image Analysis
and Recognition
17th International Conference, ICIAR 2020
Póvoa de Varzim, Portugal, June 24–26, 2020
Proceedings, Part II
Lecture Notes in Computer Science 12132
Founding Editors
Gerhard Goos
Karlsruhe Institute of Technology, Karlsruhe, Germany
Juris Hartmanis
Cornell University, Ithaca, NY, USA
Image Analysis
and Recognition
17th International Conference, ICIAR 2020
Póvoa de Varzim, Portugal, June 24–26, 2020
Proceedings, Part II
123
Editors
Aurélio Campilho Fakhri Karray
University of Porto University of Waterloo
Porto, Portugal Waterloo, ON, Canada
Zhou Wang
University of Waterloo
Waterloo, ON, Canada
LNCS Sublibrary: SL6 – Image Processing, Computer Vision, Pattern Recognition, and Graphics
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
ICIAR 2020 was the 17th edition of the series of annual conferences on Image Analysis
and Recognition, organized, this year, as a virtual conference due to the pandemic
outbreak of Covid-19 affecting all the world, with an intensity never felt by the
humanity in the last hundred years. These are difficult and challenging times, never-
theless the situation provides new opportunities for disseminating science and tech-
nology to an even wider audience, through powerful online mediums. Although
organized as a virtual conference, ICIAR 2020 kept a forum for the participants to
interact and present their latest research contributions in theory, methodology, and
applications of image analysis and recognition. ICIAR 2020, the International
Conference on Image Analysis and Recognition took place during June 24–26, 2020.
ICIAR is organized by the Association for Image and Machine Intelligence (AIMI), a
not-for-profit organization registered in Ontario, Canada.
We received a total of 123 papers from 31 countries. The review process was carried
out by members of the Program Committee and other external reviewers. Each paper
was reviewed by at least two reviewers, and checked by the conference co-chairs.
A total of 73 papers were accepted and appear in these proceedings. We would like to
sincerely thank the authors for their excellent research work and for responding to our
call, and to thank the reviewers for dedicating time to the review process and for the
careful evaluation and the feedback provided to the authors. It is this collective effort
that resulted in a strong conference program and a high-quality proceedings.
We were very pleased to include three outstanding keynote talks: “Deep Learning
and The Future of Radiology” by Daniel Rueckert (Imperial College London, UK);
“Towards Human-Friendly Explainable Artificial Intelligence” by Hani Hagras
(University of Essex, UK); and “Embedded Computer Vision and Machine Learning
for Drone Imaging” by Ioannis Pitas (Aristotle University of Thessaloniki, Greece). We
would like to express our gratitude to the keynote speakers for accepting our invitation
to share their vision and recent advances made in their areas of expertise.
This virtual conference was organized in two parallel tracks, corresponding to nine
sessions, each one corresponding to the following chapters in this proceedings with two
volumes:
1. Image Processing and Analysis
2. Video Analysis
3. Computer Vision
4. 3D Computer Vision
5. Machine Learning
6. Medical Image Analysis
7. Analysis of Histopathology Images
8. Diagnosis and Screening of Ophthalmic Diseases
9. Grand Challenge on Automatic Lung Cancer Patient Management
vi Preface
General Chairs
Aurélio Campilho University of Porto, Portugal
Fakhri Karray University of Waterloo, Canada
Zhou Wang University of Waterloo, Canada
Program Chairs
Image Processing and Analysis
Armando Pinho University of Aveiro, Portugal
Ed Vrscay University of Waterloo, Canada
Computer Vision
J. Santos Victor Instituto Superior Técnico, Portugal
Petia Radeva University of Barcelona, Spain
Machine Learning
Jaime Cardoso University of Porto, Portugal
J. Salvador Garreta University of Jaume I, Spain
Faculty of Engineering
University of Waterloo
Canada
Organization ix
Waterloo AI Institute
Canada
Program Committee
Alaa El Khatib University of Waterloo, Canada
Alberto Taboada-Crispi Universidad Central Marta Abreu de Las Villas, Cuba
Alexander Wong University of Waterloo, Canada
Ambra Demontis Università di Cagliari, Italy
Ana Filipa Sequeira INESC TEC, Portugal
Ana Maria Mendonça University of Porto, Portugal
Andreas Uhl University of Salzburg, Austria
Angel Sappa ESPOL Polytechnic University, Ecuador,
and Computer Vision Center, Spain
António Cunha University of Trás-os-Montes and Alto Douro, Portugal
Arjan Kujiper TU Darmstadt, Fraunhofer IGD, Germany
Armando Pinho University of Aveiro, Portugal
Aurélio Campilho University of Porto, Portugal
Beatriz Remeseiro Universidad de Oviedo, Spain
Bob Zhang University of Macau, Macau
Carlos Thomaz FEI, Brazil
Catarina Carvalho INESC TEC, Portugal
Chaojie Ou University of Waterloo, Canada
Dariusz Frejlichowski West Pomeranian University of Technology, Poland
Dipti Sarmah Sarmah’s Algorithmic Intelligence Research Lab,
The Netherlands
Dominique Brunet Environment and Climate Change Canada
(Toronto Area), Canada
Edward Vrscay University of Waterloo, Canada
Fabian Falck Imperial College London, UK
Fakhri Karray University of Waterloo, Canada
Farzad Khalvati University of Toronto, Canada
Francesco Camastra University of Naples Parthenope, Italy
Francesco Renna University of Porto, Portugal
Francesco Tortorella Universita’ degli Studi di Salerno, Italy
Gerald Schaefer Loughborough University, UK
Giang Tran University of Waterloo, Canada
Gilson Giraldi LNCC, Brazil
Giuliano Grossi University of Milan, Italy
Guillaume Noyel International Prevention Research Institute, France
Hasan Ogul Baskent University, Turkey
Hassan Rivaz Concordia University, Canada
Hélder Oliveira INESC TEC, Portugal
Hicham Sekkati National Research Council of Canada, Canada
Howard Li University of New Brunswick, Canada
Huiyu Zhou Queen’s University Belfast, UK
Jaime Cardoso University of Porto, Portugal
Jinghao Xue University College London, UK
João Pedrosa INESC TEC, Portugal
Organization xi
Additional Reviewers
Américo Pereira INESC TEC, Portugal
Audrey Chung University of Waterloo, Canada
Devinder Kumar Stanford University, USA
Dongdong Ma Tsinghua University, China
Guilherme Aresta INESC TEC, Portugal
Honglei Su Qingdao University, China
Isabel Rio-Torto University of Porto, Portugal
Juncheng Zhang Tsinghua University, China
Khashayar Namdar University of Toronto, Canada
Lu Zhang INSA Rennes, France
Mafalda Falcão INESC TEC, Portugal
Pedro Costa INESC TEC, Portugal
Saman Motamed University of Toronto, Canada
Tânia Pereira INESC TEC, Portugal
Tom Vicar Brno University of Technology, Czech Republic
Youcheng Zhang Tsinghua University, China
Contents – Part II
Machine Learning
Optic Disc and Fovea Detection in Color Eye Fundus Images . . . . . . . . . . . 332
Ana Maria Mendonça, Tânia Melo, Teresa Araújo,
and Aurélio Campilho
Deep Retinal Diseases Detection and Explainability Using OCT Images . . . . 358
Mohamed Chetoui and Moulay A. Akhloufi
xvi Contents – Part II
Video Analysis
Color Inference from Semantic Labeling for Person Search in Videos . . . . . . 139
Jules Simon, Guillaume-Alexandre Bilodeau, David Steele,
and Harshad Mahadik
Computer Vision
Learning to Search for Objects in Images from Human Gaze Sequences . . . . 280
Afonso Nunes, Rui Figueiredo, and Plinio Moreno
3D Computer Vision
Dense Disparity Maps from RGB and Sparse Depth Information Using
Deep Regression Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
Pedro Nuno Leite, Renato Jorge Silva, Daniel Filipe Campos,
and Andry Maykol Pinto
1 Introduction
Fisher Discriminant Analysis (FDA) [1], first proposed in [2], is a powerful sub-
space learning method which tries to minimize the intra-class scatter and max-
imize the inter-class scatter of data for better separation of classes. FDA treats
all pairs of the classes the same way; however, some classes might be much fur-
ther from one another compared to other classes. In other words, the distances
of classes are different. Treating closer classes need more attention because clas-
sifiers may more easily confuse them whereas classes far from each other are
generally easier to separate. The same problem exists in Kernel FDA (KFDA)
[3] and in most of subspace learning methods that are based on generalized
eigenvalue problem such as FDA and KFDA [4]; hence, a weighting procedure
might be more appropriate.
In this paper, we propose several weighting procedures for FDA and KFDA.
The contributions of this paper are three-fold: (1) proposing Cosine-Weighted
c Springer Nature Switzerland AG 2020
A. Campilho et al. (Eds.): ICIAR 2020, LNCS 12132, pp. 3–15, 2020.
https://doi.org/10.1007/978-3-030-50516-5_1
4 B. Ghojogh et al.
tr(U S B U )
maximize , (1)
U tr(U S W U )
where tr(·) is the trace of matrix. The Fisher criterion is a generalized Rayleigh-
Ritz Quotient [8]. We may recast the problem to [9]:
maximize tr(U S B U ),
U
(2)
subject to U S W U = I,
where the S W ∈ Rd×d and S B ∈ Rd×d are the intra- (within) and inter-class
(between) scatters, respectively [9]:
c
nr
c
− μ(r) ) =
(r) (r)
S W := nr (xi − μ(r) )(xi nr X̆ r X̆ r , (3)
r=1 i=1 r=1
c c
c
S B := nr n (μ(r) − μ() )(μ(r) − μ() ) = nk M r N M
r , (4)
r=1 =1 r=1
(r) (r)
where Rd×nr X̆ r := [x1 − μ(r) , . . . , xnr − μ(r) ], Rd×c M r := [μ(r) −
μ(1) , . . . , μ(r) − μ(c) ], and Rc×c N := diag([n1 , . . . , nc ] ). The mean of the
nr (r)
r-th class is Rd μ(r) := (1/nr ) i=1 xi . The Lagrange relaxation [10] of the
optimization problem is: L = tr(U S B U )−tr Λ (U S W U −I) , where Λ is
Weighted Fisher Discriminant Analysis in the Input and Feature Spaces 5
a diagonal matrix which includes the Lagrange multipliers. Setting the derivative
of Lagrangian to zero gives:
∂L set
= 2S B U − 2S W U Λ = 0 =⇒ S B U = S W U Λ, (5)
∂U
which is the generalized eigenvalue problem (S B , S W ) where the columns of U
and the diagonal of Λ are the eigenvectors and eigenvalues, respectively [11]. The
p leading columns of U (so to have U ∈ Rd×p ) are the FDA projection directions
where p is the dimensionality of the subspace. Note that p ≤ min(d, n − 1, c − 1)
because of the ranks of the inter- and intra-class scatter matrices [9].
maximize tr(Y ΔB Y ),
Y
(6)
subject to Y ΔW Y = I,
where ΔW ∈ Rn×n and ΔB ∈ Rn×n are the intra- and inter-class scatters in
the feature space, respectively [3,9]:
c
ΔW := nr K r H r K
r , (7)
r=1
c c
c
ΔB := nr n (ξ (r) − ξ () )(ξ (r) − ξ () ) = nr Ξ r N Ξ
r , (8)
r=1 =1 r=1
where Rnr ×nr H r := I − (1/nr )11 is the centering matrix, the (i, j)-th
(r)
entry of K r ∈ Rn×nr is K r (i, j) := k(xi , xj ), the i-th entry of ξ (r) ∈ Rn is
nr (r)
ξ (r) (i) := (1/nr ) j=1 k(xi , xj ), and Rn×c Ξ r := [ξ (r) −ξ (1) , . . . , ξ (r) −ξ (c) ].
The p leading columns of Y (so to have Y ∈ Rn×p ) are the KFDA projection
directions which span the subspace. Note that p ≤ min(n, c − 1) because of the
ranks of the inter- and intra-class scatter matrices in the feature space [9].
maximize B U ),
tr(U S
U (9)
subject to U S W U = I,
6 B. Ghojogh et al.
where R αr ≥ 0 is the weight for the pair of the r-th and -th classes,
Rc×c Ar := diag([αr1 , . . . , αrc ]). In FDA, we have αr = 1, ∀r, ∈ {1, . . . , c}.
However, it is better for the weights to be decreasing with the distances of
classes to concentrate more on the nearby classes. We denote the distances of
the r-th and -th classes by dr := ||μ(r) − μ() ||2 . The solution to Eq. (9) is the
generalized eigenvalue problem (S B , S W ) and the p leading columns of U span
the subspace.
1 d
r
αr := erf √ , (11)
2 d2r 2 2
where m > 0 is an integer. As αr is supposed to drop faster than the increase
of dk , we should have m ≥ 3 (we use m = 3 in the experiments).
Confused Distance Maximization: The Confused Distance Maximization
(CDM) [15] method uses the confusion probability among the classes as the
weight function:
n|r
nr if k = ,
αr := (14)
0 if r = ,
where n|r is the number of points of class r classified as class by a classifier such
as quadratic discriminant analysis [15,16]. One problem of the CDM method is
Weighted Fisher Discriminant Analysis in the Input and Feature Spaces 7
that if the classes are classified perfectly, all weights become zero. Conditioning
the performance of a classifier is also another flaw of this method.
k-Nearest Neighbors Weighting: The k-Nearest Neighbor (kNN) method
[17] tries to put every class away from its k-nearest neighbor classes by defining
the weight function as
1 if μ() ∈ kNN(μ(r) ),
αr := (15)
0 otherwise.
The kNN and CDM methods are sparse to make use of the betting on sparsity
principle [1,18]. However, these methods have some shortcomings. For example,
if two classes are far from one another in the input space, they are not considered
in kNN or CDM, but in the obtained subspace, they may fall close to each other,
which is not desirable. Another flaw of kNN method is the assignment of 1 to
all kNN pairs, but in the kNN, some pairs might be comparably closer.
Literature has shown that cosine similarity works very well with the FDA, espe-
cially for face recognition [19,20]. Moreover, according to the opposition-based
learning [21], capturing similarity and dissimilarity of data points can improve
the performance of learning. A promising operator for capturing similarity and
dissimilarity (opposition) is cosine. Hence, we propose CW-FDA, as a manually
weighted method, with cosine to be the weight defined as
μ(r) μ()
αr := 0.5 × 1 + cos (μ(r) , μ() ) = 0.5 × 1 + , (16)
||μ(r) ||2 ||μ() ||2
to have αr ∈ [0, 1]. Hence, the r-th weight matrix is Ar := diag(αr , ∀), which
is used in Eq. (10). Note that as we do not care about αr,r , because inter-class
scatter for r = is zero, we can set αrr = 0.
B U ),
maximize tr(U S
U , Ar
subject to U S W U = I, (17)
||Ar ||0 ≤ k, ∀r ∈ {1, . . . , c}.
8 B. Ghojogh et al.
c
= nr M r Ăr N M
r , (21)
r=1
where Ăr := Ar /||Ar ||2F because Ak is diagonal, and ||.||F is Frobenius norm.
As discussed before, the solution to Eq. (18) is the generalized eigenvalue
problem (S (τ ) , S W ). We use a step of gradient descent [23] to solve Eq. (19)
B
followed by satisfying the “0 ” norm constraint [22]. The gradient is calculated
as follows. Let R f (U , Ak ) := −tr(U S B U ). Using the chain rule, we have:
∂ Ă ∂f
∂f r ∂SB
Rc×c = vec−1
c×c ( ) ( ) vec( ) , (22)
∂Ar ∂Ar ∂ Ăr B
∂S
where we use the Magnus-Neudecker convention in which matrices are vectorized,
vec(.) vectorizes the matrix, and vec−1
c×c is de-vectorization to c × c matrix. We
have Rd×d ∂f /∂ S B = −U U whose vectorization has dimensionality d2 . For
the second derivative, we have:
2 B
∂S
×c2
Rd = nr (M r N ) ⊗ M r , (23)
∂ Ăr
where ⊗ denotes the Kronecker product. The third derivative is:
2
×c2 ∂ Ăr 1 −2
Rc = (Ar ⊗ Ar ) + I c2 . (24)
∂Ar ||Ar ||F ||Ar ||F
2 2
The learning rate of gradient descent is calculated using line search [23].
After the gradient descent step, to satisfy the condition ||Ar ||0 ≤ k, the solu-
tion is projected onto the set of this condition. Because −f should be maximized,
this projection is to set the (c − k) smallest diagonal entries of Ar to zero [22]. In
case k = c, the projection of the solution is itself, and all the weights are kept.
After solving the optimization, the p leading columns of U are the OW-FDA
projection directions that span the subspace.
Weighted Fisher Discriminant Analysis in the Input and Feature Spaces 9
maximize B Y ),
tr(Y Δ
Y (25)
subject to Y ΔW Y = I,
B ∈ Rn×n , is
where the weighted inter-class scatter in the feature space, Δ
defined as:
c
c
c
B :=
Δ αr nr n (ξ (r) − ξ () )(ξ (r) − ξ () ) = nr Ξ r Ar N Ξ
r . (26)
r=1 =1 r=1
B , ΔW ) and
The solution to Eq. (25) is the generalized eigenvalue problem (Δ
the p leading columns of Y span the subspace.
B Y ),
maximize tr(Y Δ
Y , Ar
subject to Y ΔW Y = I, (27)
||Ar ||0 ≤ k, ∀r ∈ {1, . . . , c},
Another random document with
no related content on Scribd:
him for four years holding the place of konzertmeister in Liszt’s
orchestra at Weimar. Then he is konzertmeister in Hannover, where
he married Amalie Weiss, a singer of unrivalled art. Still later he went
to Berlin, where, as teacher and quartet leader, he stood for the very
highest ideals of his art. The famous Joachim quartet, which his spirit
may be said almost to have created, consisted of Joachim, De Ahna
(1835-1892), once a pupil of Mayseder, Emanuel Wirth, violist, who
succeeded Rappoldi in 1877, and Robert Hausmann (1852-1909).
De Ahna was succeeded by J. C. Kruse (b. 1859), and Kruse in
1897 by Karl Halir. Joachim gave himself with deepest devotion to
the study of Beethoven’s works; and probably his performances of
the last quartets of Beethoven have established a standard of
excellence in chamber music which may never be exalted further.
Brahms wrote his violin concerto especially for Joachim, who alone
for many years was able to play it. Here is but another case where
the great virtuoso stands behind the great composer. Kreutzer,
Clement, and Rode all have entered in spirit into the immortality of
great music through Beethoven. David stands behind the concerto of
Mendelssohn, Joachim behind that of Brahms.
So, too, there is a great virtuoso just behind three of the most
successful of modern concertos: Sarasate behind the first concerto
of Lalo, the very substance of Bruch’s second concerto and his
Scottish Fantasia. Pablo de Sarasate (1844-1908) came from his
native land of Spain to Paris in 1856. Already as a boy of ten he had
astonished the Spanish court. Into his small hands had already come
a priceless Stradivari, gift of the queen of Spain. After three years’
study under Alard in Paris he entered upon his career of virtuoso,
which took him well over the face of the world, from the Orient to the
United States. The numerous short pieces which he has composed
are tinged with Spanish color. There are gypsy dances, Spanish
dances, the Jota Aragonesa, romances and fantasias, all of which
are brilliant and many of which are at present among the favorite
solos of all violinists.
VII
In turning to the violin pieces of the great masters of music one finds
first and foremost ideas, great or charming, which are wholly worthy
of expression. As these find their outlet in music in melody, harmony,
and rhythm, and take their shape in form, melody becomes
intensified and suggests as well as sings, harmony is enriched, form
developed and sustained. Only the solo sonatas of Bach have
demanded such manifold activity from the violin alone. Other
composers have called to the aid of their ideas some other
instrument—pianoforte, organ, or orchestra. The great masters have
indeed placed no small burden of the frame and substance of such
compositions on the shoulders of this second instrument, usually the
pianoforte. Hence we have music which is no longer solo music for
the violin, but duets in which both instruments play an obbligato part.
Such are the violin sonatas of Beethoven, Brahms, César Franck
and others, thoroughly developed, well-articulated and often truly
great music.
Beethoven wrote ten sonatas for pianoforte and violin, all but one
between the years 1798 and 1803. This was a time when his own
fame as a virtuoso was at its height, and the pianoforte part in all the
sonatas calls for technical skill and musicianship from the pianist.
Upon the violinist, too, they make no less claim. In fact Beethoven’s
idea of this duet sonata as revealed in all but the last, that in G
major, opus 96, is the idea of a double concerto, both performers
displaying the best qualities and the most brilliant of their
instruments, the pianist at the same time adding the harmonic
background and structural coherence which may well be conceived
as orchestral. It is not surprising then to find in these works
something less of the ‘poetic idea’ than may be discovered, or has
been, in the sonatas for pianoforte alone, the string quartets, and the
symphonies. Beethoven is not concerned solely with poetic
expression in music. And not only many of the violin sonatas, but the
horn sonata and the 'cello sonatas, were written for a certain player,
and even for a special occasion.
Of the three sonatas, opus 12, written not later than 1798 and
dedicated to the famous Italian Salieri, then resident in Vienna, little
need be said. On the whole they are without conspicuous distinction
in style, treatment, or material; though certain movements, especially
the slow movements of the second and third sonatas, are full of deep
feeling. Likewise the next two sonatas, that in A minor, opus 23, and
that in F major, opus 24, are not of great significance in the list of
Beethoven’s works, though the former speaks in a highly
impassioned vein, and the latter is so frankly charming as to have
won for itself something of the favor of the springtime.
Shortly after these Beethoven composed the three sonatas, opus 30,
dedicated to the Czar of Russia, in which there is at once a more
pronounced element of virtuosity and likewise a more definite poetic
significance. The first and last of this set are in A major and G major,
and show very clearly the characteristics which are generally
associated with these keys. The former is vigorous, the latter
cheerful. Both works are finely developed and carefully finished in
style, and the Tempo di minuetto in the latter is one of the most
charming of Beethoven’s compositions. The sonata in C minor which
stands between these two is at once more rough-hewn and
emotionally more powerful.
Toward the end of 1812 the French violinist, Pierre Rode, came to
Vienna, and to this event alone is probably due the last of
Beethoven’s sonatas for pianoforte and violin. If he had set out to
exhaust the possibilities of brilliant effect in the combination of the
two instruments, he achieved his goal, as far as it was attainable
within the limits of technique at that time, in the Kreutzer Sonata.
Then for a period of nine years he lost interest in the combination.
When he turned to it again, for this sonata in G, opus 96, it was with
far deeper purpose. The result is a work of a fineness and reserve,
of a pointed style, and cool meaning. It recalls in some measure the
Eighth Symphony, and like that symphony has been somewhat
eclipsed by fellow works of more obvious and striking character. Yet
from the point of view of pure and finely-wrought music it is the best
of the sonatas for pianoforte and violin. Mention has already been
made of the first performance of the work, given on the 29th of
December, 1812, by Rode and Beethoven’s pupil, the Archduke
Rudolph.
The concerto for violin and orchestra, opus 61, must be given a
place among his masterpieces. It belongs in point of time between
the two great pianoforte concertos, in G major and E-flat major; and
was first performed by the violinist Franz Clement, to whom it was
dedicated, at a concert in the Theater an der Wien, on December 23,
1806. Difficult as the concerto is for the violinist, Beethoven has
actually drawn upon only a few of the characteristics of the
instrument, and chiefly upon its power over broad, soaring melody.
He had written a few years earlier two Romances, opus 40 and opus
50, for violin and orchestra, which may be taken as preliminary
experiments in weaving a solo-violin melody with the many strands
of the orchestra. The violin part in the concerto is of noble and
exalted character, and yet at the same time gives to the instrument
the chance to express the best that lies within it.
The plan of the work is suggestively different from the plan of the last
two concertos for pianoforte. In these Beethoven treats the solo
instrument as a partner or at times as an opponent of the orchestra,
realizing its wholly different and independent individuality. At the very
beginning of both the G major and the E-flat major concertos, the
piano asserts itself with weight and power equal to the orchestra’s,
and the ensuing music results as it were from the conflict or the
union of these two naturally contrasting forces. The violin has no
such independence from the orchestra, of which, in fact, it is an
organic member. The violin concerto begins with a long orchestral
prelude, out of which the solo instrument later frees itself, as it were,
and rises, to pursue its course often as leader, but never as
opponent.[52]
The few works by Schubert for pianoforte and violin belong to the
winter of 1816 and 1817, and, though they have a charm of melody,
they are of relatively slight importance either in his own work or in
the literature for the instrument. There are a concerto in D major;
three sonatinas, in D, A minor, and G minor, opus 137, Nos. 1, 2, 3;
and a sonata in A, opus 162.
There are three violin sonatas by Brahms which hold a very high
place in music. The first, opus 78, in G major, was written after the
first and second symphonies and even the violin concerto had been
made public (Jan. 1, 1879). It has, perhaps, more than any of his
earlier works, something of grace and pleasant warmth, of those
qualities which made the second symphony acceptable to more than
his prejudiced friends. Certainly this sonata, which was played with
enthusiasm by Joachim all over Europe, made Brahms’ circle of
admirers vastly broader than it had been before.
The second sonata, opus 100, did not appear until seven years after
the first. Here again there is warmth and grace of style, though the
impression the work makes as a whole is rather more serious than
that made by the earlier sonata. Of course at a time when Brahms
and Wagner were being almost driven at each other by their ardent
friends and backers the resemblance between the first theme of this
sonata in A major and the melody of the Prize Song in the
Meistersinger did not pass unnoticed. The resemblance is for an
instant startling, but ceases to exist after the first four notes.
The third sonata, that in D minor, opus 108, appeared two years
later. On the whole it has more of the sternness one cannot but
associate with Brahms than either of those which precede it. There
are grotesque accents in the first movement, and also a passage of
forty-six measures over a dominant pedal point, and even the
delightful movement in F-sharp minor (un poco presto e con
sentimento) has a touch of deliberateness. The slow movement on
the other hand is direct, and the last movement has a strong, broad
swing.
VIII
Turning now to music in its more recent developments, we shall find
that each nation has contributed something of enduring worth to the
literature of the violin. Certainly, high above all modern sonatas, and
perhaps above all sonatas for pianoforte and violin, stands that by
César Franck, dedicated to M. Eugène Ysäye. By all the standards
we have, this work is immortally great. From the point of view of style
it presents at their best all the qualities for which Franck’s music is
valued. There are the fineness in detail and the seemingly
spontaneous polyphonic skill, the experiments, or rather the
achievements in binding the four movements into a unified whole by
employing the same or cognate thematic material in all, the
chromatic alterations of harmonies and the almost unlimited
modulations. Besides these more or less general qualities, the
pianoforte and the violin are most sympathetically combined, and the
treatment of both instruments is varied and interesting. Franck’s
habit of short phrases here seems wholly proper, and never
suggests as it does in some of his other works a too intensive
development of musical substance. In short this sonata, full of
mystical poetry, is a flawless masterpiece, from the opening
movement that seems like a dreamy improvisation, to the sunny
canon at the end of the work.
There is a sonata for violin and piano by Gabriel Fauré, opus 13,
which has won favor, and which Saint-Saëns characterized as
géniale. The year 1905 heard the first performance of the admirable
violin sonata in C major of M. Vincent d’Indy.
Of far broader conception, however, than the sonatas, are the two
brilliant concertos by Christian Sinding, the first in A major, opus 45,
the second in D major, opus 60. Concerning his music in general M.
Henry Marteau, the eminent French violinist who introduced the first
concerto to the public and who is a close friend of Sinding, has
written: 'He is very Norwegian in his music, but less so than Grieg,
because his works are of far broader conception and would find
themselves cramped in the forms that are so dear to Grieg.’[56]
Among more recent works for the violin by German composers the
sonata by Richard Strauss stands conspicuous. This is an early work
—opus 18—and its popularity is already on the wane. There is a
concerto in A major, opus 101, by Max Reger, and a Suite im alten
Stil for violin and piano, opus 93. There are concertos by Gernsheim,
as well: but on the whole there has been no remarkable output of
music for the violin in Germany since that of Brahms and of Max
Bruch.
Karl Goldmark, the Bohemian composer, has written two concertos,
of which the first, opus 28, in A minor, offers an excellent example of
the composer’s finished and highly pleasing style. The second
concerto, without opus number, is among his later works. Two suites
for piano and violin, opus 11 and opus 43, were made familiar by
Sarasate. Dvořák’s concerto, opus 53, has been frequently played.
He composed as well a Romance, opus 11, for violin and orchestra,
and a sonatina, opus 100, for violin and pianoforte. The works of
Jenö Hubay are of distinctly virtuoso character.
[53] Joachim had in his possession a concerto for violin by Schumann, written
likewise near the end of his life.
[54] The theme of the last movement can be found in two songs, Regenlied and
Nachklang, opus 59, published seven years earlier.
I
In giving an account of early chamber music we may confine
ourselves to the consideration of early instrumental music of certain
kinds, although the term at first did not apply to pure instrumental
music alone. Chamber music in the sixteenth century meant
instrumental or vocal music for social and private purposes as
distinguished from public musical performances in churches or in
theatres. In its modern sense chamber music applies, of course, only
to instrumental ensembles, and it is therefore not necessary to dwell
upon the vocal side of chamber music beginnings, except where, as
in its incipient stages, music was written for both kinds of
performances.[59] In searching for examples of early chamber music,
therefore, we must above all consider all such music, vocal or
instrumental, as was not composed for the use of the church or
theatre. Properly speaking the accompanied art-songs of the
fourteenth and fifteenth centuries, which were discussed in Vol. I,
Chapter IX, of our narrative history, represent the very beginnings of
artistic instrumental music that during the following three centuries
developed into pure instrumental chamber music. In forwarding this
development the dance music of the period and other instrumental
compositions of the fifteenth century were important factors.