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
Visit to download the full and correct content document:
https://textbookfull.com/product/image-analysis-and-recognition-17th-international-co
nference-iciar-2020-povoa-de-varzim-portugal-june-24-26-2020-proceedings-part-ii-a
urelio-campilho/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
Image Analysis and Recognition 17th International

Conference ICIAR 2020 Póvoa de Varzim Portugal June 24
26 2020 Proceedings Part I 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/

Conference ICIAR 2014 Vilamoura Portugal October 22 24
2014 Proceedings Part II 1st Edition Aurélio Campilho
recognition-11th-international-conference-iciar-2014-vilamoura-
portugal-october-22-24-2014-proceedings-part-ii-1st-edition-
aurelio-campilho/

Conference ICIAR 2014 Vilamoura Portugal October 22 24
2014 Proceedings Part I 1st Edition Aurélio Campilho
recognition-11th-international-conference-iciar-2014-vilamoura-
portugal-october-22-24-2014-proceedings-part-i-1st-edition-
aurelio-campilho/

Conference ICIAR 2016 in Memory of Mohamed Kamel Póvoa
de Varzim Portugal July 13 15 2016 Proceedings 1st
Edition Aurélio Campilho
recognition-13th-international-conference-iciar-2016-in-memory-
of-mohamed-kamel-povoa-de-varzim-portugal-
Detection of Intrusions and Malware and Vulnerability
Assessment 17th International Conference DIMVA 2020
Lisbon Portugal June 24 26 2020 Proceedings Clémentine
Maurice
https://textbookfull.com/product/detection-of-intrusions-and-
malware-and-vulnerability-assessment-17th-international-
conference-dimva-2020-lisbon-portugal-
june-24-26-2020-proceedings-clementine-maurice/
HCI International 2020 Posters 22nd International

Conference HCII 2020 Copenhagen Denmark July 19 24 2020
Proceedings Part II Constantine Stephanidis
https://textbookfull.com/product/hci-
international-2020-posters-22nd-international-conference-
hcii-2020-copenhagen-denmark-july-19-24-2020-proceedings-part-ii-
constantine-stephanidis/
Engineering Psychology and Cognitive Ergonomics.

Cognition and Design: 17th International Conference,
EPCE 2020, Held as Part of the 22nd HCI International
Conference, HCII 2020, Copenhagen, Denmark, July 19–24,
2020, Proceedings, Part II Don Harris
https://textbookfull.com/product/engineering-psychology-and-
cognitive-ergonomics-cognition-and-design-17th-international-
conference-epce-2020-held-as-part-of-the-22nd-hci-international-
conference-hcii-2020-copenhagen-denmark/
HCI International 2020 Late Breaking Posters 22nd

International Conference HCII 2020 Copenhagen Denmark
July 19 24 2020 Proceedings Part II Constantine
Stephanidis
https://textbookfull.com/product/hci-international-2020-late-
breaking-posters-22nd-international-conference-
hcii-2020-copenhagen-denmark-july-19-24-2020-proceedings-part-ii-
constantine-stephanidis/
Graph Transformation 13th International Conference ICGT

2020 Held as Part of STAF 2020 Bergen Norway June 25 26
2020 Proceedings Fabio Gadducci
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
Editorial Board Members

Elisa Bertino
Purdue University, West Lafayette, IN, USA
Wen Gao
Peking University, Beijing, China
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Gerhard Woeginger
RWTH Aachen, Aachen, Germany
Moti Yung
Columbia University, New York, NY, USA
More information about this series at http://www.springer.com/series/7412
Aurélio Campilho Fakhri Karray
• •
Zhou Wang (Eds.)
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
ISSN 0302-9743 ISSN 1611-3349 (electronic)

Lecture Notes in Computer Science
ISBN 978-3-030-50515-8 ISBN 978-3-030-50516-5 (eBook)
https://doi.org/10.1007/978-3-030-50516-5
LNCS Sublibrary: SL6 – Image Processing, Computer Vision, Pattern Recognition, and Graphics
© Springer Nature Switzerland AG 2020

The chapter “Deep Learning Models for Segmentation of Mobile-Acquired Dermatological Images” is
licensed under the terms of the Creative Commons Attribution 4.0 International License (http://
creativecommons.org/licenses/by/4.0/). For further details see license information in the chapter.
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now
known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book are
believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors
give a warranty, expressed or implied, with respect to the material contained herein or for any errors or
omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in
published maps and institutional affiliations.
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
Chapter 8 and 9 correspond to two successful parallel events: Special Session on

“Novel Imaging Methods for Diagnosis and Screening of Ophthalmic Diseases”
co-chaired by Ana Mendonça (University of Porto, Portugal) and Koen Vermeer
(Roterdam Eye Hospital, The Netherlands); and “Grand Challenge on Automatic
Lung Cancer Patient Management” organized by João Pedrosa, Carlos Ferreira, and
Guilherme Aresta from Institute for Systems and Computer Engineering, Technology
and Science (INESC TEC), Portugal.
We would like to thank the program area chairs: Armando Pinho (University of
Aveiro, Portugal) and Ed Vrscay (University of Waterloo, Canada), chairs for the area
on Image Processing and Analysis; José Santos Victor (Instituto Superior Técnico,
University of Lisbon, Portugal) and Petia Radeva (University of Barcelona, Spain),
chairs for the area on Computer Vision; Jaime Cardoso (University of Porto, Portugal)
and J. Salvador Sanchez Garreta (University of Jaume I, Spain), chairs for the area on
Machine Learning; and Ana Mendonça (University of Porto, Portugal) and Roberto
Hornero (University of Valladolid, Spain), chairs for the area on Medical Image
Analysis; who have secured a high-quality program. We also would like to thank the
members of the Organizing Committee from INESC TEC, for helping with the local
logistics, and the publications and web chairs, Carlos Ferreira and Khaled Hammouda,
for maintaining the website, interacting with the authors, and preparing the proceed-
ings. We are also grateful to Springer’s editorial staff, for supporting this publication in
the LNCS series. As well, we would like to thank the precious sponsorship and support
of the INESC TEC, the Faculty of Engineering at the University of Porto, Portugal, the
Waterloo Artificial Intelligence Institute, the Faculty of Engineering of the University
of Waterloo, and the Center for Pattern Analysis and Machine Intelligence at the
University of Waterloo. We also appreciate the valuable co-sponsorship of the
IEEE EMB Portugal Chapter, the IEEE Computational Intelligence Society,
Kitchener-Waterloo Chapter, and the Portuguese Association for Pattern Recognition.
We also would like to acknowledge Lurdes Catalino from Abreu Events for managing
the registrations.
We were very pleased to welcome all the participants to ICIAR 2020, a virtual
conference edition. For those who were not able to attend, we hope this publication
provides a good overview into the research presented at the conference.
June 2020 Aurélio Campilho

Fakhri Karray
Zhou Wang
Organization
General Chairs
Aurélio Campilho University of Porto, Portugal
Fakhri Karray University of Waterloo, Canada
Zhou Wang University of Waterloo, Canada
Local Organizing Committee

Catarina Carvalho INESC TEC, Portugal
João Pedrosa INESC TEC, Portugal
Luís Teixeira University of Porto, Portugal
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
Medical Image Analysis

Ana Mendonça University of Porto, Portugal
Roberto Hornero University of Valladolid, Spain
Grand Challenge on Automatic Lung Cancer Patient Management

Carlos Ferreira INESC TEC, Portugal
Guilherme Aresta INESC TEC, Portugal
Novel Imaging Methods for Diagnosis and Screening of Ophthalmic Diseases

Ana Mendonça University of Porto, Portugal
Koen Vermeer Rotterdam Eye Hospital, The Netherlands
viii Organization
Publication and Web Chairs

Carlos Ferreira INESC TEC, Portugal
Khaled Hammouda Shopify, Canada
Supported and Co-sponsored by
AIMI – Association for Image and Machine Intelligence
Center for Biomedical Engineering Research

INESC TEC – Institute for Systems and Computer Engineering,
Technology and Science
Portugal
Department of Electrical and Computer Engineering

Faculty of Engineering
University of Porto
Portugal
Faculty of Engineering
Canada
Organization ix
CPAMI – Centre for Pattern Analysis and Machine Intelligence

Canada
Waterloo AI Institute
Canada
IEEE Engineering in Medicine and Biology Society

Portugal
IEEE Computational Intelligence Society

Kitchener-Waterloo Chapter
APRP - Portuguese Association for Pattern Recognition

Portugal
x Organization
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
Organization xi
João Rodrigues University of the Algarve, Portugal

Johan Debayle École nationale supérieure des Mines de Saint-Étienne,
France
Jonathan Boisvert CNRC, Canada
Jorge Batista University of Coimbra, Portugal
Jorge Marques University of Lisbon, Portugal
Jorge Silva University of Porto, Portugal
José Alba Castro University of Vigo, Spain
José Garreta University of Jaume I, Spain
José Rouco University of Coruña, Spain
José Santos-Victor University of Lisbon, Portugal
Jose-Jesus Fernandez CNB-CSIC, Spain
Juan José Rodríguez Universidad de Burgos, Spain
Juan Lorenzo Ginori Universidad Central Marta Abreu de Las Villas, Cuba
Kaushik Roy North Carolina A&T State University, USA
Kelwin Fernandes NILG.AI, Portugal
Koen Vermeer Rotterdam Eye Hospital, The Netherlands
Linlin Xu University of Waterloo, Canada
Luc Duong École de technologie supérieure, Canada
Luís Alexandre University of Beira Interior, Portugal
Luís Teixeira University of Porto, Portugal
Mahmoud El-Sakka University of Western Ontario, Canada
Mahmoud Hassaballah South Valley University, Egypt
Mahmoud Melkemi Univeristé de Haute-Alsace, France
Manuel Penedo University of Coruña, Spain
María García University of Valladolid, Spain
Marie Muller North Carolina State University, USA
Mariella Dimiccoli Institut de Robòtica i Informàtica Industrial, Spain
Mario Vento Università di Salerno, Italy
Markus Koskela CSC - IT Center for Science, Finland
Mehran Ebrahimi University of Ontario, Canada
Mohammad Shafiee University of Waterloo, Canada
Nicola Strisciuglio University of Groningen, The Netherlands
Oliver Montesdeoca Universitat de Barcelona, Spain
Parthipan Siva Sportlogiq, Canada
Pascal Fallavollita University of Ottawa, Canada
Pavel Zemčík Brno University of Technology, Czech Republic
Pedro Carvalho INESC TEC, Portugal
Pedro Pina University of Lisbon, Portugal
Petia Radeva University of Barcelona, Spain
Philip Morrow Ulster University, UK
Radim Kolář Brno University of Technology, Czech Republic
Reyer Zwiggelaar Aberystwyth University, UK
Robert Fisher University of Edinburgh, UK
Robert Sablatnig TU Wien, Austria
Roberto Hornero University of Valladolid, Spain
xii Organization
Rosa María Valdovinos Universidad Autónoma del Estado de México, Mexico

Rui Bernardes University of Coimbra, Portugal
Sajad Saeedi Imperial College London, UK
Sébai Dorsaf National School of Computer Science, Tunisia
Shamik Sural Indian Institute of Technology, India
Vicente García-Jiménez Universidad Autónoma de Ciudad Juérez, Mexico
Víctor González-Castro Universidad de Leon, Spain
Xosé Pardo CiTIUS, Universidade de Santiago de Compostela,
Spain
Yasuyo Kita National Institute AIST, Japan
Yun-Qian Miao General Motors, Canada
Zhou Wang University of Waterloo, Canada
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
Weighted Fisher Discriminant Analysis in the Input and Feature Spaces . . . . 3

Benyamin Ghojogh, Milad Sikaroudi, H. R. Tizhoosh, Fakhri Karray,
and Mark Crowley
Backprojection for Training Feedforward Neural Networks in the Input

and Feature Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Benyamin Ghojogh, Fakhri Karray, and Mark Crowley
Parallel Implementation of the DRLSE Algorithm . . . . . . . . . . . . . . . . . . . . 25

Daniel Popp Coelho and Sérgio Shiguemi Furuie
A Multiscale Energy-Based Time-Domain Approach for Interference

Detection in Non-stationary Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Vittoria Bruni, Lorenzo Della Cioppa, and Domenico Vitulano
SMAT: Smart Multiple Affinity Metrics for Multiple Object Tracking . . . . . . 48

Nicolas Franco Gonzalez, Andres Ospina, and Philippe Calvez
Combining Mixture Models and Spectral Clustering for Data Partitioning . . . . 63

Julien Muzeau, Maria Oliver-Parera, Patricia Ladret,
and Pascal Bertolino
MSPNet: Multi-level Semantic Pyramid Network for Real-Time

Object Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
Ji Li and Yingdong Ma
Multi-domain Document Layout Understanding Using Few-Shot

Object Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Pranaydeep Singh, Srikrishna Varadarajan, Ankit Narayan Singh,
and Muktabh Mayank Srivastava
Object Tracking Through Residual and Dense LSTMs. . . . . . . . . . . . . . . . . 100

Fabio Garcea, Alessandro Cucco, Lia Morra, and Fabrizio Lamberti
Theoretical Insights into the Use of Structural Similarity Index

in Generative Models and Inferential Autoencoders . . . . . . . . . . . . . . . . . . . 112
Efficient Prediction of Gold Prices Using Hybrid Deep Learning . . . . . . . . . 118

Turner Tobin and Rasha Kashef
xiv Contents – Part II
Exploring Information Theory and Gaussian Markov Random Fields

for Color Texture Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
Cédrick Bamba Nsimba and Alexandre L. M. Levada
Anomaly Detection for Images Using Auto-encoder Based

Sparse Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
Qiang Zhao and Fakhri Karray
Medical Image Analysis
A Framework for Fusion of T1-Weighted and Dynamic MRI Sequences . . . . 157

João F. Teixeira, Sílvia Bessa, Pedro F. Gouveia,
and Hélder P. Oliveira
Contributions to a Quantitative Unsupervised Processing and Analysis

of Tongue in Ultrasound Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
Fábio Barros, Ana Rita Valente, Luciana Albuquerque, Samuel Silva,
António Teixeira, and Catarina Oliveira
Improving Multiple Sclerosis Lesion Boundaries Segmentation

by Convolutional Neural Networks with Focal Learning . . . . . . . . . . . . . . . 182
Gustavo Ulloa, Alejandro Veloz, Héctor Allende-Cid,
and Héctor Allende
B-Mode Ultrasound Breast Anatomy Segmentation . . . . . . . . . . . . . . . . . . . 193

João F. Teixeira, António M. Carreiro, Rute M. Santos,
and Hélder P. Oliveira
Enhancing the Prediction of Lung Cancer Survival Rates Using 2D Features

from 3D Scans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
Tahira Ghani and B. John Oommen
Lesion Localization in Paediatric Epilepsy Using Patch-Based

Convolutional Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Azad Aminpour, Mehran Ebrahimi, and Elysa Widjaja
Deep Learning Models for Segmentation of Mobile-Acquired

Dermatological Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
Catarina Andrade, Luís F. Teixeira, Maria João M. Vasconcelos,
and Luís Rosado
Semi-automatic Tool to Identify Heterogeneity Zones in LGE-CMR

and Incorporate the Result into a 3D Model of the Left Ventricle . . . . . . . . . 238
Maria Narciso, António Ferreira, and Pedro Vieira
Contents – Part II xv
Analysis of Histopathology Images
A Deep Learning Based Pipeline for Efficient Oral Cancer Screening

on Whole Slide Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Jiahao Lu, Nataša Sladoje, Christina Runow Stark, Eva Darai Ramqvist,
Jan-Michaél Hirsch, and Joakim Lindblad
Studying the Effect of Digital Stain Separation of Histopathology Images

on Image Search Performance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
Alison K. Cheeseman, Hamid R. Tizhoosh, and Edward R. Vrscay
Generalized Multiple Instance Learning for Cancer Detection

in Digital Histopathology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
Jan Hering and Jan Kybic
Diagnosis and Screening of Ophthalmic Diseases
A Multi-dataset Approach for DME Risk Detection in Eye

Fundus Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
Catarina Carvalho, João Pedrosa, Carolina Maia, Susana Penas,
Ângela Carneiro, Luís Mendonça, Ana Maria Mendonça,
and Aurélio Campilho
Enhancement of Retinal Fundus Images via Pixel Color Amplification . . . . . 299

Alex Gaudio, Asim Smailagic, and Aurélio Campilho
Wavelet-Based Retinal Image Enhancement . . . . . . . . . . . . . . . . . . . . . . . . 313

Safinaz ElMahmoudy, Lamiaa Abdel-Hamid, Ahmed El-Rafei,
and Salwa El-Ramly
An Interpretable Data-Driven Score for the Assessment of Fundus

Images Quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
Youri Peskine, Marie-Carole Boucher, and Farida Cheriet
Optic Disc and Fovea Detection in Color Eye Fundus Images . . . . . . . . . . . 332
Ana Maria Mendonça, Tânia Melo, Teresa Araújo,
and Aurélio Campilho
The Effect of Menopause on the Sexual Dimorphism in the Human

Retina – Texture Analysis of Optical Coherence Tomography Data . . . . . . . . 344
Ana Nunes, Pedro Serranho, Hugo Quental, Miguel Castelo-Branco,
and Rui Bernardes
Deep Retinal Diseases Detection and Explainability Using OCT Images . . . . 358
Mohamed Chetoui and Moulay A. Akhloufi
xvi Contents – Part II
Grand Challenge on Automatic Lung Cancer Patient Management
An Automated Workflow for Lung Nodule Follow-Up Recommendation

Using Deep Learning. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
Krishna Chaitanya Kaluva, Kiran Vaidhya, Abhijith Chunduru,
Sambit Tarai, Sai Prasad Pranav Nadimpalli, and Suthirth Vaidya
Pulmonary-Nodule Detection Using an Ensemble of 3D SE-ResNet18

and DPN68 Models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378
Or Katz, Dan Presil, Liz Cohen, Yael Schwartzbard, Sarah Hoch,
and Shlomo Kashani
3DCNN for Pulmonary Nodule Segmentation and Classification . . . . . . . . . . 386

Zhenhuan Tian, Yizhuan Jia, Xuejun Men, and Zhongwei Sun
Residual Networks for Pulmonary Nodule Segmentation

and Texture Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
Adrian Galdran and Hamid Bouchachia
Automatic Lung Cancer Follow-Up Recommendation with 3D

Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
Gurraj Atwal and Hady Ahmady Phoulady
Deep Residual 3D U-Net for Joint Segmentation and Texture Classification

of Nodules in Lung . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
Alexandr Rassadin
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 429

Contents – Part I
Image Processing and Analysis
Exploring Workout Repetition Counting and Validation Through

Deep Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Bruno Ferreira, Pedro M. Ferreira, Gil Pinheiro, Nelson Figueiredo,
Filipe Carvalho, Paulo Menezes, and Jorge Batista
FlowChroma - A Deep Recurrent Neural Network for Video Colorization . . . 16

Thejan Wijesinghe, Chamath Abeysinghe, Chanuka Wijayakoon,
Lahiru Jayathilake, and Uthayasanker Thayasivam
Benchmark for Generic Product Detection: A Low Data Baseline

for Dense Object Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
Srikrishna Varadarajan, Sonaal Kant, and Muktabh Mayank Srivastava
Supervised and Unsupervised Detections for Multiple Object Tracking

in Traffic Scenes: A Comparative Study . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Hui-Lee Ooi, Guillaume-Alexandre Bilodeau, and Nicolas Saunier
Variation of Perceived Colour Difference Under Different

Surround Luminance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Thilan Costa, Vincent Gaudet, Edward R. Vrscay, and Zhou Wang
4K or Not? - Automatic Image Resolution Assessment . . . . . . . . . . . . . . . . 61

Vyas Anirudh Akundy and Zhou Wang
Detecting Macroblocking in Images Caused by Transmission Error . . . . . . . . 66

Ganesh Rajasekar and Zhou Wang
Bag of Tricks for Retail Product Image Classification . . . . . . . . . . . . . . . . . 71

Muktabh Mayank Srivastava
Detection and Recognition of Food in Photo Galleries for Analysis

of User Preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
Evgeniy Miasnikov and Andrey Savchenko
Real Time Automatic Urban Traffic Management Framework Based

on Convolutional Neural Network Under Limited Resources Constraint . . . . . 95
Antoine Meicler, Assan Sanogo, Nadiya Shvai, Arcadi Llanza,
Abul Hasnat, Marouan Khata, Ed-Doughmi Younes, Alami Khalil,
Yazid Lachachi, and Amir Nakib
xviii Contents – Part I
Slicing and Dicing Soccer: Automatic Detection of Complex Events

from Spatio-Temporal Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Lia Morra, Francesco Manigrasso, Giuseppe Canto, Claudio Gianfrate,
Enrico Guarino, and Fabrizio Lamberti
Video Analysis
RN-VID: A Feature Fusion Architecture for Video Object Detection . . . . . . . 125

Hughes Perreault, Maguelonne Heritier, Pierre Gravel,
Guillaume-Alexandre Bilodeau, and Nicolas Saunier
Color Inference from Semantic Labeling for Person Search in Videos . . . . . . 139
Jules Simon, Guillaume-Alexandre Bilodeau, David Steele,
and Harshad Mahadik
2D Bidirectional Gated Recurrent Unit Convolutional Neural Networks

for End-to-End Violence Detection in Videos . . . . . . . . . . . . . . . . . . . . . . . 152
Abdarahmane Traoré and Moulay A. Akhloufi
Video Based Live Tracking of Fishes in Tanks . . . . . . . . . . . . . . . . . . . . . . 161

José Castelo, H. Sofia Pinto, Alexandre Bernardino, and Núria Baylina
Using External Knowledge to Improve Zero-Shot Action Recognition

in Egocentric Videos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
Adrián Núñez-Marcos, Gorka Azkune, Eneko Agirre,
Diego López-de-Ipiña, and Ignacio Arganda-Carreras
A Semantics-Guided Warping for Semi-supervised Video Object

Instance Segmentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
Qiong Wang, Lu Zhang, and Kidiyo Kpalma
Two-Stream Framework for Activity Recognition with 2D Human

Pose Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
Wei Chang, Chunyang Ye, and Hui Zhou
Video Object Segmentation Using Convex Optimization of Foreground

and Background Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Jia-Wei Chen and Jin-Jang Leou
Computer Vision
Deep Learning for Partial Fingerprint Inpainting and Recognition . . . . . . . . . 223

Marc-André Blais, Andy Couturier, and Moulay A. Akhloufi
Contents – Part I xix
A Visual Perception Framework to Analyse Neonatal Pain in Face Images. . . 233

Lucas Pereira Carlini, Juliana C. A. Soares, Giselle V. T. Silva,
Tatiany M. Heideirich, Rita C. X. Balda, Marina C. M. Barros,
Ruth Guinsburg, and Carlos Eduardo Thomaz
Combining Asynchronous Events and Traditional Frames for Steering

Angle Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
Abdoulaye O. Ly and Moulay A. Akhloufi
Survey of Preprocessing Techniques and Classification Approaches

in Online Signature Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
Mohammad Saleem and Bence Kovari
SSIM Based Signature of Facial Micro-Expressions. . . . . . . . . . . . . . . . . . . 267

Vittoria Bruni and Domenico Vitulano
Learning to Search for Objects in Images from Human Gaze Sequences . . . . 280
Afonso Nunes, Rui Figueiredo, and Plinio Moreno
Detecting Defects in Materials Using Deep Convolutional Neural Networks . . . 293

Quentin Boyadjian, Nicolas Vanderesse, Matthew Toews,
and Philippe Bocher
Visual Perception Ranking of Chess Players . . . . . . . . . . . . . . . . . . . . . . . . 307

Laercio R. Silva Junior and Carlos E. Thomaz
Video Tampering Detection for Decentralized Video

Transcoding Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
Rabindranath Andujar, Ignacio Peletier, Jesus Oliva,
Marc Cymontkowski, Yondon Fu, Eric Tang, and Josh Allman
Generalized Subspace Learning by Roweis Discriminant Analysis . . . . . . . . . 328

Understanding Public Speakers’ Performance: First Contributions

to Support a Computational Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
Fábio Barros, Ângelo Conde, Sandra C. Soares, António J. R. Neves,
and Samuel Silva
Open Source Multipurpose Multimedia Annotation Tool . . . . . . . . . . . . . . . 356

Joed Lopes da Silva, Alan Naoto Tabata, Lucas Cardoso Broto,
Marta Pereira Cocron, Alessandro Zimmer, and Thomas Brandmeier
SLAM-Based Multistate Tracking System for Mobile

Human-Robot Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
Thorsten Hempel and Ayoub Al-Hamadi
xx Contents – Part I
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
Exploitation of Dense MLS City Maps for 3D Object Detection . . . . . . . . . . 393

Örkény Zováthi, Balázs Nagy, and Csaba Benedek
Automatic Stereo Disparity Search Range Detection on Parallel

Computing Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
Ruveen Perera and Tobias Low
Multi-camera Motion Estimation with Affine Correspondences . . . . . . . . . . . 417

Khaled Alyousefi and Jonathan Ventura
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433

Machine Learning
Weighted Fisher Discriminant Analysis
in the Input and Feature Spaces
Benyamin Ghojogh1(B) , Milad Sikaroudi2 , H. R. Tizhoosh2 , Fakhri

Karray1 , and Mark Crowley1
1
Department of Electrical and Computer Engineering,
University of Waterloo, Waterloo, ON, Canada
{bghojogh,karray,mcrowley}@uwaterloo.ca
2
KIMIA Lab, University of Waterloo, Waterloo, ON, Canada
{msikaroudi,tizhoosh}@uwaterloo.ca
Abstract. Fisher Discriminant Analysis (FDA) is a subspace learning

method which minimizes and maximizes the intra- and inter-class scat-
ters of data, respectively. Although, in FDA, all the pairs of classes are
treated the same way, some classes are closer than the others. Weighted
FDA assigns weights to the pairs of classes to address this shortcoming
of FDA. In this paper, we propose a cosine-weighted FDA as well as an
automatically weighted FDA in which weights are found automatically.
We also propose a weighted FDA in the feature space to establish a
weighted kernel FDA for both existing and newly proposed weights. Our
experiments on the ORL face recognition dataset show the effectiveness
of the proposed weighting schemes.
Keywords: Fisher Discriminant Analysis (FDA) · Kernel FDA ·

Cosine-weighted FDA · Automatically weighted FDA · Manually
weighted FDA
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.
FDA (CW-FDA) as a new modification of FDA, (2) proposing Automatically

Weighted FDA (AW-FDA) as a new version of FDA in which the weights are set
automatically, and (3) proposing Weighted KFDA (W-KFDA) to have weighting
procedures in the feature space, where both the existing and the newly proposed
weighting methods can be used in the feature space.
The paper is organized as follows: In Sect. 2, we briefly review the theory
of FDA and KFDA. In Sect. 3, we formulate the weighted FDA, review the
existing weighting methods, and then propose CW-FDA and AW-FDA. Section 4
proposes weighted KFDA in the feature space. In addition to using the existing
methods for weighted KFDA, two versions of CW-KFDA and also AW-KFDA
are proposed. Section 5 reports the experiments. Finally, Sect. 6 concludes the
paper.
2 Fisher and Kernel Discriminant Analysis

2.1 Fisher Discriminant Analysis
(r)
Let {xi ∈ Rd }ni=1
r
denote the samples of the r-th class where nr is the class’s
sample size. Suppose μ(r) ∈ Rd , c, n, and U ∈ Rd×d denote the mean of r-th
class, the number of classes, the total sample size, and the projection matrix
in FDA, respectively. Although some methods solve FDA using least squares
problem [5,6], the regular FDA [2] maximizes the Fisher criterion [7]:
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].
2.2 Kernel Fisher Discriminant Analysis

Let the scalar and matrix kernels be denoted by k(xi , xj ) := φ(xi ) φ(xj ) and
K(X 1 , X 2 ) := Φ(X 1 ) Φ(X 2 ), respectively, where φ(.) and Φ(.) are the pulling
functions. According to the representation theory [12], any solution must lie in
the span of all the training vectors, hence, Φ(U ) = Φ(X) Y where Y ∈ Rn×d
contains the coefficients. The optimization of kernel FDA is [3,9]:
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].
3 Weighted Fisher Discriminant Analysis

The optimization of Weighted FDA (W-FDA) is as follows:
maximize B U ),
tr(U S
U (9)
subject to U S W U = I,
6 B. Ghojogh et al.
B ∈ Rd×d , is defined as:

where the weighted inter-class scatter, S

c
c
c
B :=
S αr nr n (μ(r) − μ() )(μ(r) − μ() ) = nr M r Ar N M
r ,
r=1 =1 r=1
(10)
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.
3.1 Existing Manual Methods

In the following, we review some of the existing weights for W-FDA.
Approximate Pairwise Accuracy Criterion: The Approximate Pairwise
Accuracy Criterion (APAC) method [13] has the weight function:
1 d
r
αr := erf √ , (11)
2 d2r 2 2
where erf(x) is the error function:

x
2 2
[−1, 1] erf(x) := √ e−t dt. (12)
π 0
This method approximates the Bayes error for class pairs.

Powered Distance Weighting: The powered distance (POW) method [14]
uses the following weight function:
1
αr := , (13)
dm
r
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
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.
3.2 Cosine Weighted Fisher Discriminant Analysis
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.
3.3 Automatically Weighted Fisher Discriminant Analysis
In AW-FDA, there are c + 1 matrix optimization variables which are V and

Ak ∈ Rc×c , ∀k ∈ {1, . . . , c} because at the same time where we want to maximize
the Fisher criterion, the optimal weights are found. Moreover, to use the betting
on sparsity principle [1,18], we can make the weight matrix sparse, so we use
“0 ” norm for the weights to be sparse. The optimization problem is as follows
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.
We use alternating optimization [22] to solve this problem:

U (τ +1) := arg max tr(U S (τ ) U ) U S W U = I , (18)
B
U

(τ +1)
A(τr
+1)
:= arg min − tr(U S B U (τ +1)
) ||A r ||0 ≤ k , ∀r, (19)
Ar
where τ denotes the iteration.

Since we use an iterative solution for the optimization, it is better to nor-
malize the weights in the weighted inter-class scatter; otherwise, the weights
gradually explode to maximize the objective function. We use 2 (or Frobenius)
norm for normalization for ease of taking derivatives. Hence, for OW-FDA, we
slightly modify the weighted inter-class scatter as

c
c
αr
B :=
S c (r)
− μ() )(μ(r) − μ() )
2 nr n (μ
αr
(20)
r=1 =1 =1
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.
4 Weighted Kernel Fisher Discriminant Analysis

We define the optimization for Weighted Kernel FDA (W-KFDA) as:
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.
4.1 Manually Weighted Methods in the Feature Space

All the existing weighting methods in the literature for W-FDA can be used
as weights in W-KFDA to have W-FDA in the feature space. Therefore, Eqs.
(11), (13), (14), and (15) can be used as weights in Eq. (26) to have W-KFDA
with APAC, POW, CDM, and kNN weights, respectively. To the best of our
knowledge, W-KFDA is novel and has not appeared in the literature. Note that
there is a weighted KFDA in the literature [24], but that is for data integration,
which is for another purpose and has an entirely different approach.
The CW-FDA can be used in the feature space to have CW-KFDA. For
this, we propose two versions of CW-KFDA: (I) In the first version, we use
Eq. (16) or Ar := diag(αr , ∀) in the Eq. (26). (II) In the second ver-
sion, we notice that cosine is based on inner product so the normalized ker-
nel matrix between the means of classes can be used instead to use the sim-
ilarity/dissimilarity in the feature space rather than in the input space. Let

i,j := K i,j / K i,i K j,j be the normalized
Rd×c M := [μ1 , . . . , μc ]. Let K
kernel matrix [25] where K i,j denotes the (i, j)-th element of the kernel matrix
Rc×c K(M , M ) = Φ(M ) Φ(M ). The weights are [0, 1] αr := K r, or
Ar := diag(K r, , ∀). We set αr,r = 0.
4.2 Automatically Weighted Kernel Fisher Discriminant Analysis

Similar to before, the optimization in AW-KFDA is:
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.
The Norwegian violinist, Ole Bull (1810-1880), who achieved an

international fame, should be mentioned in this connection. His
compositions, in slight forms or transcriptions, enjoyed considerable
popularity.
On the whole the technique of violin playing has hardly advanced

beyond Paganini. Practically little or no advance has been possible.
But undoubtedly this once miraculous technique is now within the
grasp of all the great virtuosi of the present day. To mention these
would go beyond the purpose of this chapter, which has been, in so
far as possible, to select from the list of hundreds a few men that
have united, so to speak, the technique of the violin to the general
progress of music, through their influence as players, as teachers, as
composers, or as mentors, so far as violin music is concerned, to
greater composers.
The mass of music composed by the great violinists of the

nineteenth century is immense. The works of large proportions as
well as those of small were composed with perhaps the chief aim of
revealing the scope of the instrument; and as for the concertos it is
hardly unfair to say that they were composed with the additional
purpose of offering to the composer the best chance to display his
individual style as a player. Certainly of these many composers
Spohr and Vieuxtemps were the most capable as musicians in a
general way; and as it must be granted that both were at their best in
the performance of their own concertos, so it may be said that their
concertos rose to their highest value under the fingers of their
creators. To that same value they have not otherwise risen.
The concerto is, after all, a long piece of music in symphonic

proportions, and time seems to have proved that it must justify itself
by more than display of the special qualities of a certain instrument.
There must be in addition to this something of genuine musical
value. The thoughts which it expresses—for so we must name the
outpourings of a musical inspiration which have no substance but
sound—must be first worthy of expression. There must be melody
and harmony of distinct and vivid character. These the concertos of
the violin-composers oftenest lack; and therefore from the point of
view of pure music, one finds in them a lack not only of originality but
of strength.
Their short pieces stand a better chance of a longer life, because in

them a slender idea is not stretched to fill a broad form, and because
for a short time sheer beauty of sound, such as the violin is capable
of, and dexterity of fingers are a sufficient delight to the ear.
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.
The sonata in A, opus 47, is the ninth of the violin sonatas of

Beethoven. It was written especially for the English violinist, George
Bridgetower, with whom Beethoven played it for the first time on the
17th or 24th of May, 1803. According to the violinist himself, who
was, by the way, a mulatto and exceedingly mannered, he altered a
passage in this performance of the work which greatly pleased
Beethoven. However this may be, Beethoven later fell out with him,
and subsequently dedicated the sonata to the great violinist
Rodolphe Kreutzer, who came to Vienna in the suite of General
Bernadotte. It has since been known as the Kreutzer Sonata. It is an
imposing and brilliant work, but it may be fairly said that it owes its
general popularity to the favor of virtuosi to whom it offers a grateful
test of technical ability. Emotionally the first movement alone is of
sustained and impressive meaning. The theme of the Andante is of
great sweetness, but the variations are hardly more than a series of
more and more elaborate ornamentations, designed for the benefit of
the players. The brilliant last movement seems to have been first
conceived for the preceding sonata in A major, opus 30, No. 1.
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 two violin sonatas by Schumann, in A minor, opus 105,

and in D minor, opus 121. Both are works belonging to the last years
of his life, and both reflect a sad and gloomy spirit; but both contain
much that is rarely beautiful. They will strike the ear at once as more
modern than those of Beethoven, mostly of course because of the
treatment of the pianoforte. Here it may well be mentioned that
improvements in the pianoforte rather changed the problem of
writing duet sonatas such as these. The new power of the instrument
might easily threaten the violin with extinction. On the whole
Schumann’s handling of the combination is remarkably successful.
He is inclined now and then to treat the pair of instruments in unison
—as in the first movement of the sonata in A minor—which is a rank
waste of the beauties which the diversity in the natures of pianoforte
and violin makes possible. On the other hand, such a movement as
that in G major in the second sonata, its unusual beginning with a
melody given by the violin in pizzicato chords, and its third statement
of the melody in rich double-stops, is a masterpiece.[53]
The only considerable contribution by Mendelssohn to the literature

of the violin is the concerto written for and first performed by
Ferdinand David. A sonata in F minor, opus 4, is without distinction.
But the concerto must be reckoned as one of Mendelssohn’s
greatest works. Certainly, standing as it does between the concerto
of Beethoven, on the one hand, and that of Brahms, on the other, it
cannot but appear small in size and slight in content. But the themes,
especially the chief theme of the first movement, are well chosen,
the orchestral part exquisitely and thoroughly finished, and the
treatment of the violin, thanks to David, smoothly effective. The
cadenza—is it Mendelssohn or David?—is of sterling worth, and it is
happily arranged in the movement as a whole before the third
section, so that the hearer has not the shock which accompanies the
enforced dragging in of virtuoso stuff in most cadenzas. It glides
naturally out of what came before, and slowly flows back into the
course of the movement.
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 workmanship is, of course, highly involved and recondite. There

is a thematic relationship between the first and last movements,[54]
and the themes and even the accompaniment are put to learned
uses. But the style is gracious and charming, the treatment of the
violin wholly satisfactory, and the combination of the two instruments
close and interesting.
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.
No violin sonatas show more ingenuity in the combining of the two

instruments than those of Brahms. Mr. Thomas F. Dunhill in his book
on Chamber Music,[55] chooses from each of them a passage which
really represents a new effect in this field of which one would have
thought all the effects discovered.
The concerto for violin and orchestra stands among Brahms’
supreme achievements, a giant among concertos matched only by
that of Beethoven. It is not a matter for surprise that Brahms, who in
many ways deliberately tried to follow Beethoven, and who even
here chose the same key (D major) that Beethoven chose for his
concerto, chose likewise the old-fashioned form of concerto. The
work gains ponderance by reason of the long orchestral introduction
in both the first and second movements. There is, likewise, as in the
pianoforte concertos, too conscious a suppression of superficial
brilliance. But what is this slight heaviness compared to the soaring
power of its glorious themes? Truly the violin rises high above the
orchestra as on wings of light.
The treatment of the violin relates the concerto to Joachim even

more definitely than the dedication. It is full of the most exacting
difficulties, some of which in the last movement gave even Joachim
pause. The double-stops, however, and the frequent passages in
two voices were, after all, effects in which Joachim was especially
successful. Some of the close co-operation of the two great masters
on this single great masterpiece is revealed in the correspondence
which passed between Joachim and Brahms and happily has been
preserved.
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.
This is by no means the only brilliant accomplishment of the French

composers in violin music. Lalo’s Concerto in F minor, opus 20, and
his Spanish Symphony for violin and orchestra, opus 21, must be
given a place among the most successful of modern compositions.
They were both composed between 1873 and the beginning of 1875.
Both were dedicated to Sarasate, whose influence contributed not a
little to their perfection of style, and who was the first to play them in
public. The ‘Spanish Symphony’ was greatly admired by
Tschaikowsky and apparently put the thought of writing his own
concerto into his head. In a letter to Mme. von Meck, written in
March, 1878, he showed a positive enthusiasm for Lalo’s work which
had recently become known to him through the performance by the
‘very modern’ violinist Sarasate. And of Lalo he wrote that, like Léo
Delibes and Bizet, he shunned studiously all routine commonplaces,
sought new forms without wishing to appear profound, and, unlike
the Germans, cared more for musical beauty than for mere respect
of the old traditions. Besides these two concertos Lalo wrote within
the next few years a ‘Romance-Serenade,’ a ‘Norwegian Fantasia,’
and a Concerto Russe, for violin and orchestra.
Sarasate seems to have stimulated almost all of the composers with

whom he came in contact. Saint-Saëns wrote three concertos for
violin and orchestra, opus 20, in A major, opus 58, in C major, and
opus 61, in B minor, and dedicated all to Sarasate. Of these the third
is the broadest in form and the most impressing, and is a favorite
among its fellows as the second concerto for pianoforte, opus 22, is
among the five works in that form. It was composed in 1880 and
played for the first time by Sarasate. Saint-Saëns wrote besides
these three concertos an ‘Introduction and Rondo Capriccioso,’ opus
28, a ‘Romanze,’ opus 48, and a ‘Concert Piece,’ opus 62, for violin
and orchestra, and two sonatas—opus 75, in D minor, and opus 102,
in E-flat major—for violin and pianoforte. There is also a brilliant
Havanaise, opus 83, for violin and orchestra.
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.
Among the Scandinavian composers Grieg holds the highest rank,

and his three sonatas for violin and pianoforte are among the favorite
compositions for this combination. Their charm is like that of his
other works, and consists not a little in the presence of a distinct
national idiom which, until one becomes thoroughly used to it, strikes
the ear with delightful freshness. The three sonatas are respectively
opus 8, in F major, opus 13, in G major, and opus 45, in C minor.
The last is a fiery, dramatic work. The two earlier ones are
characterized by grace and charm. With the exception of the
pianoforte concerto in A minor, Grieg showed himself nowhere more
successful than in these sonatas in the treatment of form. His ideas
are generally slight, and his workmanship delicate and refined.
Hence he is at his best in short pieces. But the violin sonatas are on
the whole well sustained, and the themes in the last of them, and
particularly the chief theme of the first movement, have a breadth
quite unusual in the great part of his music.
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 the Russians, Tschaikowsky’s concerto for violin in D major,

opus 35, is one of the greatest written for the instrument. Of
Tschaikowsky’s admiration for the Spanish Symphony of Lalo,
mention has already been made. After this had prompted him to
write a concerto of his own, the work went on with astonishing
rapidity; was, in fact, roughly on paper within the space of a month. It
was first performed on December 4, 1884, at a Philharmonic concert
in Vienna by Adolf Brodsky (b. 1851). It was originally dedicated to
Leopold Auer (b. 1845), but Tschaikowsky later re-dedicated it to
Brodsky, having heard that Auer had dissuaded Émile Sauret from
playing it in Petrograd. As to the difficulties of the work much may be
gleaned from a letter written by Brodsky to Tschaikowsky after the
first performance. Among other things he wrote: 'I had the wish to
play the concerto in public ever since I first looked it through. * * * I
often took it up and often put it down, because my laziness was
stronger than my wish to reach the goal. You have, indeed, crammed
too many difficulties into it. * * * One can play it again and again and
never be bored; and this is a most important circumstance for the
conquering of its difficulties.’[57]
Of the three movements only the last (allegro vivacissimo, 2-4, D

major) has a distinctly Russian flavor. This comes to it not only from
the nature of the two chief themes, which are in the character of
Russian folk-songs, but from the gorgeous coloring, both harmonic
and orchestral, the wildness of climaxes, and the Slavic idiom of
repeating a single phrase over and over again. It is a riotous piece of
music, this last movement, full of an animation, almost a madness
which is intoxicating. Hanslick heard in it only the brutal and
wretched jollity of a Russian Kermesse; but his fierce judgment has
not been supported by the public or by the profession.
There is a concerto for violin in A minor, opus 82, by Alexander

Glazounoff, composed in 1904 and first performed at a Queen’s Hall
concert in London, by Mischa Elman, on October 17, 1905. The work
is dedicated to Leopold Auer, to whom, as has just been mentioned,
Tschaikowsky originally dedicated his concerto for violin. It is a work
without distinction.
Modern Violinists. From top left to bottom right: Pablo Sarasate,
Fritz Kreisler,
Eugène Ysäye. Jacques Thibaud.
The violin concerto of Sibelius in D minor, opus 47, was composed in
1905 and first played by Karl Halir in Berlin, October 19, 1905. It is a
work of far greater power than that of Glazounoff. Mrs. Rosa
Newmarch in her monograph on Sibelius,[58] likens the difficulties in
it to those of the Tschaikowsky concerto, which were for a while
considered insurmountable. The concerto is in three movements of
which the first is gloomy and forbidding, though poignant in the
extreme, the second noble and more classic, the last—the coda of
which was added by Pietro Floridia—savagely effective.
In Germany we meet with Sarasate again in the second concerto

and Scottish Fantasy by Max Bruch. These are the best known of
Bruch’s works for violin and orchestra, among which may be
mentioned a first concerto, opus 26, in G minor, a Romance, opus
42, an Adagio Appassionato, opus 57, and a Serenade, opus 75.
The second concerto, opus 44, was, according to Bruch, inspired by
stories of the Carlist wars in Spain, told by Sarasate. It was
composed in Bonn in 1877, ten years after the first, and was first
publicly performed by Sarasate, in London, during the fall of that
year. In form it is free and rhapsodical, consisting of an adagio
movement, then a movement in recitative style, and a final rondo. All
through the work the solo violin predominates. The Scottish
Fantasia, composed a year or two later, was dedicated to Sarasate.
The use of Scotch songs in the five movements is so free that
English critics could hardly recognize them, and were angry.
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.
The Italian Leone Sinigaglia became known to the world by his

concerto for violin, opus 20, in A major, played in Berlin in 1901 by
his countryman, Arrigo Serrato. Later works include a Rapsodia
piemontese for violin and orchestra, and a Romance for the same
combination, opus 29. The violin music of Emanuel Móor, including a
concerto and a remarkably fine suite for violin unaccompanied, has
yet to be better known. Georges Enescou first attracted attention by
compositions for the violin. On the whole, however, it may be said
that the violin is awaiting a new contribution to its literature. This
contribution is doubtless delayed by the great attention given at the
present day to the piano, the orchestra, or other combinations of
instruments, by which the modern growth in harmony and the
change in ideas of polyphony may be given a full expression. Until
these various ideas have become firmly rooted and well-grown, the
violin will profit but vicariously by them.
FOOTNOTES:
[51] This famous arrangement was published by the Maison Richault in Paris as
Thème de Rode, chanté avec variations dans le Barbier de Séville en Italien par
Mmes. Sontag, Alboni, Trebelli; en français par Mlle. Maria Bailly; paroles
françaises d’Adolph Larmande, avec accompagnement de piano par L. Moreau.
See Notice sur Rode, by F. A. A. Paroisse-Pougin (Paris, 1874).
[52] See Paul Bekker: ‘Beethoven.’ Berlin, 1913.
[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.
[55] ‘Chamber Music.’ London, 1913.
[56] See Song Journal, November 10, 1895.
[57] See Modest Tschaikowsky: ‘Life of Peter Ilyitch Tschaikowsky.’
[58] ‘Jean Sibelius, a Finnish Composer.’

CHAPTER XIV
THE BEGINNINGS OF CHAMBER
MUSIC
The term ‘chamber music’; fifteenth-century dances; lute music,
early suites; vocal ‘chamber music’—Early ‘sonatas’: Gabrieli;
Rossi; Marini; etc.—Vitali, Veracini, Bassani and Corelli; Corelli’s
pupils; Vivaldi; Bach and Handel.
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.
The fifteenth century dances such as the Pawirschwantz, the

Fochsschwantz, and others, employed the polyphonic style peculiar
to the vocal compositions of the time. They lacked inspiration and
were of a restless character because of frequent changes of rhythm.
There was little to distinguish them from each other; they were in
fact, in the words of Michael Prætorius, ‘as like as eggs,’ and their
general character was not different from that of the vocal
compositions of the same period. Probably no modern ear could
listen to them with enjoyment.
Presumably this music was to be played on any instrument, without

differentiation. No single instrument was especially favored until the
following century, when the perfection and the popularity of the lute
helped to bring chamber music into existence. This instrument was
indeed so highly perfected and the players so skilled that they were
able to perform upon it even difficult polyphonic works. This gave an
opportunity to the people to become acquainted, through private
performances, with a great number of musical compositions. To
satisfy the demands of their friends lutenists arranged and
transcribed for their instruments all kinds of compositions, including
even entire six-part masses. While these arrangements served their
purpose they were probably not more satisfactory than the pianoforte
arrangement of orchestral scores today. Pieces of polyphonic
character were also composed directly for the lute, and bore such
names as Ricercar, Fantasia, Præludium, Preambel, Trio, Trium,
Toccata, Tartar le corde, etc. Besides this the lutenists produced a
large amount of music in a more popular vein, popular tunes,
dances, and descriptive pieces including ‘battles,’ ‘echoes,’ ‘bird-
songs,’ in which the composer’s intention was often not self-evident.

Image Analysis and Recognition 17th International Conference ICIAR 2020 Póvoa de Varzim Portugal June 24 26 2020 Proceedings Part II Aurélio Campilho

Uploaded by

Copyright:

Available Formats

You might also like

Image Analysis and Recognition 17th International Conference ICIAR 2020 Póvoa de Varzim Portugal June 24 26 2020 Proceedings Part II Aurélio Campilho

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Image Analysis and Recognition 17th International Conference ICIAR 2020 Póvoa de Varzim Portugal June 24 26 2020 Proceedings Part II Aurélio Campilho

Uploaded by

Copyright:

Available Formats

Image Analysis and Recognition 17th

International Conference ICIAR 2020

Image Analysis and Recognition 17th International

Image Analysis and Recognition 11th International

Image Analysis and Recognition 11th International

Image Analysis and Recognition 13th International

HCI International 2020 Posters 22nd International

Engineering Psychology and Cognitive Ergonomics.

HCI International 2020 Late Breaking Posters 22nd

Graph Transformation 13th International Conference ICGT

Editorial Board Members

Zhou Wang (Eds.)

ISSN 0302-9743 ISSN 1611-3349 (electronic)

© Springer Nature Switzerland AG 2020

Chapter 8 and 9 correspond to two successful parallel events: Special Session on

June 2020 Aurélio Campilho

Local Organizing Committee

Medical Image Analysis

Grand Challenge on Automatic Lung Cancer Patient Management

Novel Imaging Methods for Diagnosis and Screening of Ophthalmic Diseases

Publication and Web Chairs

Supported and Co-sponsored by

AIMI – Association for Image and Machine Intelligence

Center for Biomedical Engineering Research

Department of Electrical and Computer Engineering

CPAMI – Centre for Pattern Analysis and Machine Intelligence

IEEE Engineering in Medicine and Biology Society

IEEE Computational Intelligence Society

APRP - Portuguese Association for Pattern Recognition

João Rodrigues University of the Algarve, Portugal

Rosa María Valdovinos Universidad Autónoma del Estado de México, Mexico

Weighted Fisher Discriminant Analysis in the Input and Feature Spaces . . . . 3

Backprojection for Training Feedforward Neural Networks in the Input

Parallel Implementation of the DRLSE Algorithm . . . . . . . . . . . . . . . . . . . . 25

A Multiscale Energy-Based Time-Domain Approach for Interference

SMAT: Smart Multiple Affinity Metrics for Multiple Object Tracking . . . . . . 48

Combining Mixture Models and Spectral Clustering for Data Partitioning . . . . 63

MSPNet: Multi-level Semantic Pyramid Network for Real-Time

Multi-domain Document Layout Understanding Using Few-Shot

Object Tracking Through Residual and Dense LSTMs. . . . . . . . . . . . . . . . . 100

Theoretical Insights into the Use of Structural Similarity Index

Efficient Prediction of Gold Prices Using Hybrid Deep Learning . . . . . . . . . 118

Exploring Information Theory and Gaussian Markov Random Fields

Anomaly Detection for Images Using Auto-encoder Based

Medical Image Analysis

A Framework for Fusion of T1-Weighted and Dynamic MRI Sequences . . . . 157

Contributions to a Quantitative Unsupervised Processing and Analysis

Improving Multiple Sclerosis Lesion Boundaries Segmentation

B-Mode Ultrasound Breast Anatomy Segmentation . . . . . . . . . . . . . . . . . . . 193

Enhancing the Prediction of Lung Cancer Survival Rates Using 2D Features

Lesion Localization in Paediatric Epilepsy Using Patch-Based

Deep Learning Models for Segmentation of Mobile-Acquired

Semi-automatic Tool to Identify Heterogeneity Zones in LGE-CMR

Analysis of Histopathology Images

A Deep Learning Based Pipeline for Efficient Oral Cancer Screening

Studying the Effect of Digital Stain Separation of Histopathology Images

Generalized Multiple Instance Learning for Cancer Detection

Diagnosis and Screening of Ophthalmic Diseases

B ∈ Rd×d , is deﬁned as: