Image Analysis and Recognition 11th

International Conference ICIAR 2014

Vilamoura Portugal October 22 24 2014
Proceedings Part I 1st Edition Aurélio
Campilho
Visit to download the full and correct content document:
https://textbookfull.com/product/image-analysis-and-recognition-11th-international-co
nference-iciar-2014-vilamoura-portugal-october-22-24-2014-proceedings-part-i-1st-ed
ition-aurelio-campilho/
More products digital (pdf, epub, mobi) instant
download maybe you interests ...

Image Analysis and Recognition 11th International

Conference ICIAR 2014 Vilamoura Portugal October 22 24
2014 Proceedings Part II 1st Edition Aurélio 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/

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/

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-ii-aurelio-
campilho/

Image Analysis and Recognition 13th International

Conference ICIAR 2016 in Memory of Mohamed Kamel Póvoa
de Varzim Portugal July 13 15 2016 Proceedings 1st
Edition Aurélio Campilho
https://textbookfull.com/product/image-analysis-and-
recognition-13th-international-conference-iciar-2016-in-memory-
of-mohamed-kamel-povoa-de-varzim-portugal-
Computational Intelligence International Joint
Conference IJCCI 2014 Rome Italy October 22 24 2014
Revised Selected Papers 1st Edition Juan Julian Merelo

https://textbookfull.com/product/computational-intelligence-
international-joint-conference-ijcci-2014-rome-italy-
october-22-24-2014-revised-selected-papers-1st-edition-juan-
julian-merelo/

Web Information Systems Engineering WISE 2014 15th

International Conference Thessaloniki Greece October 12
14 2014 Proceedings Part I 1st Edition Boualem
Benatallah
https://textbookfull.com/product/web-information-systems-
engineering-wise-2014-15th-international-conference-thessaloniki-
greece-october-12-14-2014-proceedings-part-i-1st-edition-boualem-
benatallah/

Internet and Distributed Computing Systems 7th

International Conference IDCS 2014 Calabria Italy
September 22 24 2014 Proceedings 1st Edition Giancarlo
Fortino
https://textbookfull.com/product/internet-and-distributed-
computing-systems-7th-international-conference-
idcs-2014-calabria-italy-september-22-24-2014-proceedings-1st-
edition-giancarlo-fortino/

Algorithms and Architectures for Parallel Processing

14th International Conference ICA3PP 2014 Dalian China
August 24 27 2014 Proceedings Part I 1st Edition Xian-
He Sun
https://textbookfull.com/product/algorithms-and-architectures-
for-parallel-processing-14th-international-conference-
ica3pp-2014-dalian-china-august-24-27-2014-proceedings-
part-i-1st-edition-xian-he-sun/

HCI International 2014 Posters Extended Abstracts

International Conference HCI International 2014
Heraklion Crete Greece June 22 27 2014 Proceedings Part
II 1st Edition Constantine Stephanidis (Eds.)
https://textbookfull.com/product/hci-international-2014-posters-
extended-abstracts-international-conference-hci-
international-2014-heraklion-crete-greece-
 Aurélio Campilho
Mohamed Kamel (Eds.)
LNCS 8814

Image Analysis
and Recognition
11th International Conference, ICIAR 2014
Vilamoura, Portugal, October 22–24, 2014
Proceedings, Part I

123
Lecture Notes in Computer Science 8814
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen

Editorial Board
David Hutchison
Lancaster University, Lancaster, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Alfred Kobsa
University of California, Irvine, CA, USA
Friedemann Mattern
ETH Zurich, Zürich, Switzerland
John C. Mitchell
Stanford University, Stanford, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
Oscar Nierstrasz
University of Bern, Bern, Switzerland
C. Pandu Rangan
Indian Institute of Technology, Madras, India
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbruecken, Germany
More information about this series at http://www.springer.com/series/7412
Aurélio Campilho Mohamed Kamel (Eds.)
•

Image Analysis
and Recognition
11th International Conference, ICIAR 2014
Vilamoura, Portugal, October 22–24, 2014
Proceedings, Part I

123
Editors
Aurélio Campilho Mohamed Kamel
Faculty of Engineering Department of Electrical and Computer
University of Porto Engineering
Porto University of Waterloo
Portugal Waterloo, ON
Canada

ISSN 0302-9743 ISSN 1611-3349 (electronic)

ISBN 978-3-319-11757-7 ISBN 978-3-319-11758-4 (eBook)
DOI 10.1007/978-3-319-11758-4

Library of Congress Control Number: 2014950801

LNCS Sublibrary: SL6 – Image Processing, Computer Vision, Pattern Recognition, and Graphics

Springer Cham Heidelberg New York Dordrecht London

© Springer International Publishing Switzerland 2014

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. Exempted from this legal reservation are brief excerpts in connection with
reviews or scholarly analysis or material supplied specifically for the purpose of being entered and executed
on a computer system, for exclusive use by the purchaser of the work. Duplication of this publication or
parts thereof is permitted only under the provisions of the Copyright Law of the Publisher’s location, in its
current version, and permission for use must always be obtained from Springer. Permissions for use may be
obtained through RightsLink at the Copyright Clearance Center. Violations are liable to prosecution under
the respective Copyright Law.
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.
While the advice and information in this book are believed to be true and accurate at the date of publication,
neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or
omissions that may be made. The publisher makes no warranty, express or implied, with respect to the
material contained herein.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

 Preface

This is the 11th edition of the ICIAR series of annual conferences offering an
opportunity for the participants to interact and present their latest research in theory,
methodology, and applications of image analysis and recognition. ICIAR 2014, the
International Conference on Image Analysis and Recognition, was held in Vila Moura,
Portugal, October 22–24, 2014. ICIAR is organized by AIMI – Association for Image
and Machine Intelligence, a not-for-profit organization registered in Ontario, Canada.
For ICIAR 2014, we received a total of 177 full papers from 39 countries. Before
the review process all the papers were checked for similarity using a comparison
database of scholarly work. The review process was carried out by members of the
Program Committee and other reviewers. Each paper was reviewed by at least two
reviewers, and checked by the conference chairs. A total of 107 papers were finally
accepted and appear in the two volumes of this proceedings. We would like to sincerely
thank the authors for responding to our call, and we thank the reviewers for the careful
evaluation and feedback provided to the authors. It is this collective effort that resulted
in the strong conference program and high-quality proceedings.
Each year we attempt to focus on a specific topic for the keynote speeches and
conduct a panel discussion on the topic.
This year, the conference theme was focused on the topic “Sparse Representations
for Image Analysis and Recognition.” We were very pleased to include three out-
standing keynote talks on this topic: “Optimization Algorithms for Sparse Represen-
tations: Some History and Recent Developments” by Mário Figueiredo, Instituto
Superior Técnico Portugal; “Morphological Diversities in Astrophysics” by Jean-Luc
Starck, CosmoStat Laboratory, France; and “Sparse Stochastic Processes with Appli-
cation to Biomedical Imaging” by Michael Unser, Ecole Polytechnique Fédérale de
Lausanne, Switzerland. The keynote speakers also participated in the panel “Sparse
Representation for Image Analysis and Recognition: Trends and Applications.” We
would like to express our gratitude to the keynote speakers for accepting our invitation
to share their vision and recent advances in their areas of expertise, which are at the
core of the topics of the conference.
We would like to thank Khaled Hammouda, the webmaster of the conference, for
maintaining the Web pages, interacting with the authors, and preparing the
proceedings.
As all conferences, the success of ICIAR 2014 is attributed to the effort and work of
many people, including members of the Organizing Committee, staff, and volunteers.
We gratefully acknowledge their support and efforts.
We are also grateful to Springer’s editorial staff for supporting this publication in the
LNCS series. We also would like to acknowledge the professional service of Viagens
Abreu in taking care of the registration process and the special events of the
conference.
VI Preface

Finally, we are very pleased to welcome all the participants to ICIAR 2014. For
those who were not able to attend, we hope this publication provides a good view into
the research presented at the conference, and we look forward to meeting you at the
next ICIAR conference.

October 2014 Aurélio Campilho

Mohamed Kamel
 ICIAR 2014 – International Conference
on Image Analysis and Recognition

General Chairs
Aurélio Campilho University of Porto, Portugal
Mohamed Kamel University of Waterloo, Canada

Local Organizing Committee

Ana Maria Mendonça University of Porto, Portugal
Jorge Alves Silva University of Porto, Portugal
João Rodrigues University of the Algarve, Portugal
José Rouco Maseda Biomedical Engineering Institute, Portugal
Jorge Novo Buján Biomedical Engineering Institute, Portugal

Conference Secretariat
Viagens Abreu SA, Portugal

Webmaster
Khaled Hammouda Waterloo, Ontario, Canada

Advisory Committee
M. Ahmadi University of Windsor, Canada
P. Bhattacharya Concordia University, Canada
T.D. Bui Concordia University, Canada
M. Cheriet University of Quebec, Canada
E. Dubois University of Ottawa, Canada
Z. Duric George Mason University, USA
G. Granlund Linköping University, Sweden
L. Guan Ryerson University, Canada
M. Haindl Institute of Information Theory and Automation,
Czech Republic
E. Hancock University of York, UK
J. Kovacevic Carnegie Mellon University, USA
M. Kunt Swiss Federal Institute of Technology (EPFL),
Switzerland
VIII ICIAR 2014

J. Padilha University of Porto, Portugal

K.N. Plataniotis University of Toronto, Canada
A. Sanfeliu Technical University of Catalonia, Spain
M. Shah University of Central Florida, USA
M. Sid-Ahmed University of Windsor, Canada
C.Y. Suen Concordia University, Canada
A.N. Venetsanopoulos University of Toronto, Canada
M. Viergever Utrecht University, Netherlands
B. Vijayakumar Carnegie Mellon University, USA
R. Ward University of British Columbia, Canada
D. Zhang Hong Kong Polytechnic University, Hong Kong

Program Committee
A. Abate University of Salerno, Italy
M. Ahmed Wilfrid Laurier University, Canada
L. Alexandre University of Beira Interior, Portugal
J. Alirezaie Ryerson University, Canada
G. Andreu-Garcia Universitat Politècnica de València, Spain
H. Araújo University of Coimbra, Portugal
Emilio Balaguer-Ballester Bournemouth University, UK
T. Barata University of Coimbra, Portugal
J. Barbosa University of Porto, Portugal
J. Batista University of Coimbra, Portugal
R. Bernardes University of Coimbra, Portugal
A. Bezerianos National University of Singapore, Singapore
J. Bioucas Technical University of Lisbon, Portugal
I. Bloch Télécom ParisTech, France
T.D. Bui Concordia University, Canada
C. Busch Gjøvik University College, Norway
F. Camastra University of Naples Parthenope, Italy
J. Cardoso University of Porto, Portugal
G. Carneiro University of Adelaide, Australia
M. Coimbra University of Porto, Portugal
M. Correia University of Porto, Portugal
J. Debayle Ecole Nationale Supérieure des Mines
de Saint-Étienne, France
J. Dias University of Coimbra, Portugal
G. Doretto West Virginia University, USA
H. du Buf University of the Algarve, Portugal
J. Fernandez Centro Nacional de Biotecnología – CSIC, Spain
I. Fondón University of Seville, Spain
A. Fred Technical University of Lisbon, Portugal
G. Freeman University of Waterloo, Canada
 ICIAR 2014 IX

D. Frejlichowski West Pomeranian University of Technology,

Poland
G. Giacinto University of Cagliari, Italy
M. Giger University of Chicago, USA
B. Gosselin University of Mons, Belgium
G. Grossi University of Milan, Italy
M. Grzegorzek University of Siegen, Germany
M. Haindl Institute of Information Theory and Automation,
Czech Republic
A. Hernandez Universitat Autònoma de Barcelona, Spain
L. Heutte Université de Rouen, France
C. Hong Hong Kong Polytechnic University, Hong Kong
L. Igual University of Barcelona, Spain
M. Khan King Saud University, Saudi Arabia
A. Kong Nanyang Technological University, Singapore
M. Koskela Aalto University, Finland
A. Kuijper Fraunhofer IGD and TU Darmstadt, Germany
J. Liang Simon Fraser University, Canada
L. Liu McGill University, Canada
N. Lomenie Paris Descartes University, France
L. Lopes University of Aveiro, Portugal
J. Lorenzo-Ginori Universidad Central “Marta Abreu” de Las
Villas, Cuba
R. Lukac Foveon, Inc., USA
A. Marcal University of Porto, Portugal
F. Marcelloni University of Pisa, Italy
U. Markowska-Kaczmar Wroclaw University of Technology, Poland
J. Marques Technical University of Lisbon, Portugal
M. Melkemi Univeristé de Haute Alsace, France
A. Mendonça University of Porto, Portugal
J. Meunier University of Montreal, Canada
M. Mignotte University of Montreal, Canada
M. Mirmehdi University of Bristol, UK
A. Mohammed Imam Muhammad Ibn Saud Islamic University,
Saudi Arabia
A. Monteiro University of Porto, Portugal
M. Nappi University of Salerno, Italy
M. Nixon University of Southampton, UK
H. Ogul Başkent University, Turkey
M. Pelillo University of Venice, Italy
M. Penedo Universidade da Coruña, Spain
F. Pereira Technical University of Lisbon, Portugal
E. Petrakis Technical University of Crete, Greece
P. Pina Technical University of Lisbon, Portugal
A. Pinho University of Aveiro, Portugal
X ICIAR 2014

L. Piras University of Cagliari, Italy

L. Prevost University of the French West Indies and Guiana,
France
H. Proença University of Beira Interior, Portugal
P. Quelhas Biomedical Engineering Institute, Portugal
M. Queluz Technical University of Lisbon, Portugal
P. Radeva Autonomous University of Barcelona, Spain
B. Raducanu Computer Vision Center, Spain
E. Ribeiro Florida Institute of Technology, USA
E. Ricci University of Perugia, Italy
S. Rota Bulò Fondazione Bruno Kessler, Italy
A. Ruano University of the Algarve, Portugal
G. Ruano University of the Algarve, Portugal
J. Sanches Technical University of Lisbon, Portugal
B. Santos University of Aveiro, Portugal
A. Sappa Compute Vision Center, Spain
F. Sattar University of Waterloo, Canada
G. Schaefer Loughborough University, UK
P. Scheunders University of Antwerp, Belgium
J. Silva University of Porto, Portugal
B. Smolka Silesian University of Technology, Poland
Z. Sun Institute of Automation, Chinese Academy of
Sciences (CASIA), China
S. Sural Indian Institute of Technology, India
A. Taboada-Crispí Universidad Central “Marta Abreu” de las Villas,
Cuba
X. Tan Nanjing University of Aeronautics and
Astronautics, China
J. Tavares University of Porto, Portugal
O. Terrades Computer Vision Center, Spain
R. Torres University of Campinas (UNICAMP), Brazil
A. Torsello Università Ca’ Foscari Venezia, Italy
A. Uhl University of Salzburg, Austria
M. Vento University of Salerno, Italy
R. Vigário Aalto University, Finland
Y. Voisin Université de Bourgogne, France
E. Vrscay University of Waterloo, Canada
Z. Wang University of Waterloo, Canada
M. Wirth University of Guelph, Canada
J. Wu University of Windsor, Canada
P. Yan Philips Research, USA
P. Zemcik Brno University of Technology, Czech Republic
Q. Zhang Waseda University, Japan
H. Zhou Queen's University Belfast, UK
R. Zwiggelaar Aberystwyth University, UK
 ICIAR 2014 XI

Reviewers

M. Al-Rawi University of Aveiro, Portugal

R. Araujo University of Waterloo, Canada
E. Bhullar South Asian University, India
M. Camplani University of Bristol, UK
C. Caridade Instituto Superior de Engenharia de Coimbra,
Portugal
J. Chen Lehigh University, USA
L. Fernandez University of León, Spain
J. Ferreira University of Porto, Portugal
E. Fidalgo University of León, Spain
M. Gangeh University of Toronto, Canada
M. Garcia University of León, Spain
V. Gonzalez Ecole Nationale Supérieure des Mines
de Saint-Étienne, France
H. Haberdar University of Houston, USA
M. Hortas Universidade da Coruña, Spain
N. Lori University of Coimbra, Portugal
S. Mahmoud University of Waterloo, Canada
J. Marcos Spanish National Research Council, Spain
Y. Miao University of Waterloo, Canada
F. Monteiro IPB – Instituto Politécnico de Bragança, Portugal
P. Moreno Instituto Superior Técnico, Portugal
J. Novo INESC TEC – INESC Technology and Science,
Portugal
H. Oliveira INESC TEC, Portugal
A. Ragab University of Waterloo, Canada
L. Reis University of Minho, Portugal
R. Rocha INESC TEC – INESC Technology and Science,
Portugal
J. Rodrigues University of the Algarve, Portugal
N. Rodriguez Universidade da Coruña, Spain
J. Rouco INESC TEC – INESC Technology and Science,
Portugal
P. Trigueiros Polytechnic Institute of Porto, Portugal
XII ICIAR 2014

Supported by

AIMI – Association for Image and Machine Intelligence

Center for Biomedical Engineering Research

INESC TEC – INESC Technology and Science
Portugal

Department of Electrical and Computer Engineering

Faculty of Engineering
University of Porto
Portugal

CPAMI – Centre for Pattern Analysis and Machine Intelligence

University of Waterloo
Canada
 Contents – Part I

Image Representation and Models

Path Descriptors for Geometric Graph Matching and Registration . . . . . . . . 3

Miguel Am˜vel Pinheiro and Jan Kybic

A Method to Detect Repeated Unknown Patterns in an Image . . . . . . . . . . . 12

Paulo J.S.G. Ferreira and Armando J. Pinho

Some ‘‘Weberized’’ L2 -Based Methods of Signal/Image Approximation . . . . 20

Ilona A. Kowalik-Urbaniak, Davide La Torre, Edward R. Vrscay,
and Zhou Wang

A New Compressor for Measuring Distances among Images . . . . . . . . . . . . 30

Armando J. Pinho, Diogo Pratas, and Paulo J.S.G. Ferreira

Perceptual Evaluation of Demosaicing Artefacts. . . . . . . . . . . . . . . . . . . . . 38

Tomasz Sergej and Radosðaw Mantiuk

Multiscale Shape Description with Laplacian Profile and Fourier Transform. . . 46

Evanthia Mavridou, James L. Crowley, and Augustin Lux

Structural Similarity-Based Approximation over Orthogonal Bases:

Investigating the Use of Individual Component Functions Sk ðx; yÞ . . . . . . . . 55
Paul Bendevis and Edward R. Vrscay

2D Thinning Algorithms with Revised Endpixel Preservation . . . . . . . . . . . 65

G˜bor N¯meth, P¯ter Kardos, and K˜lm˜n Pal˜gyi

Sparse Representation

A New Landmark-Independent Tool for Quantifying and Characterizing

Morphologic Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
S.M. Rolfe, L.L. Cox, L.G. Shapiro, and T.C. Cox

Low Light Image Enhancement via Sparse Representations . . . . . . . . . . . . . 84

Konstantina Fotiadou, Grigorios Tsagkatakis, and Panagiotis Tsakalides

Incremental and Multi-feature Tensor Subspace Learning Applied

for Background Modeling and Subtraction . . . . . . . . . . . . . . . . . . . . . . . . . 94
Andrews Sobral, Christopher G. Baker, Thierry Bouwmans,
and El-hadi Zahzah
XIV Contents – Part I

Face Image Super-Resolution Based on Topology ICA and Sparse

Representation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Yongtao Liu, Hua Yan, Xiushan Nie, and Zhen Liu

Iterative Sparse Coding for Colorization Based Compression . . . . . . . . . . . . 112

Suk-Ho Lee, Paul Oh, and Moon Gi Kang

Noise Modelling in Parallel Magnetic Resonance Imaging:

A Variational Approach. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Adri˜n Martín and Emanuele Schiavi

Image Restoration and Enhancement

An Examination of Several Methods of Hyperspectral Image Denoising:

Over Channels, Spectral Functions and Both Domains . . . . . . . . . . . . . . . . 131
Daniel Otero, Oleg V. Michailovich, and Edward R. Vrscay

Towards a Comprehensive Evaluation of Ultrasound Speckle Reduction . . . . 141

Fernando C. Monteiro, Jos¯ Rufino, and Vasco Cadavez

An Evaluation of Potential Functions for Regularized Image Deblurring . . . . 150

Buda Bajic¤, Joakim Lindblad, and NataÞa Sladoje

Drawing Parrots with Charcoal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159

A. Alsam and H.J. Rivertz

Unconstrained Structural Similarity-Based Optimization . . . . . . . . . . . . . . . 167

Daniel Otero and Edward R. Vrscay

Feature Detection and Image Segmentation

Reflectance-Based Segmentation Using Photometric and Illumination

Invariants. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
Jose-Antonio P¯rez-Carrasco, Bego•a Acha-Pi•ero,
Carmen Serrano-Gotarredona, and Theo Gevers

Meta-learning for Adaptive Image Segmentation . . . . . . . . . . . . . . . . . . . . 187

Aymen Sellaouti, Yasmina Jaâfra, and Atef Hamouda

Dynamic Multiple View Geometry with Affine Cameras . . . . . . . . . . . . . . . 198

Cheng Wan, Yiquan Wu, and Jun Sato

Energy Minimization by a-Erosion for Supervised Texture Segmentation . . . 207

Karl Skretting and Kjersti Engan

ALOE: Augmented Local Operator for Edge Detection . . . . . . . . . . . . . . . . 215

Maria De Marsico, Michele Nappi, and Daniel Riccio
 Contents – Part I XV

Multiple Object Detection with Occlusion Using Active Contour Model

and Fuzzy C-Mean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
Sara Memar, Riadh Ksantini, and Boubakeur Boufama

Classiffcation and Learning Methods

Conversational Interaction Recognition Based on Bodily

and Facial Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
Jingjing Deng, Xianghua Xie, and Shangming Zhou

Handwritten Digit Recognition Using SVM Binary Classifiers

and Unbalanced Decision Trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
Adriano Mendes Gil, Cícero Ferreira Fernandes Costa Filho,
and Marly Guimarães Fernandes Costa

A Visual-Based Driver Distraction Recognition and Detection

Using Random Forest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
Amira Ragab, Celine Craye, Mohamed S. Kamel, and Fakhri Karray

Improving Representation of the Positive Class in Imbalanced

Multiple-Instance Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
Carlos Mera, Mauricio Orozco-Alzate, and John Branch

Restricted Boltzmann Machines for Gender Classification . . . . . . . . . . . . . . 274

Jordi Mansanet, Alberto Albiol, Roberto Paredes, Mauricio Villegas,
and Antonio Albiol

DropAll: Generalization of Two Convolutional Neural Network

Regularization Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
Xavier Frazão and Luís A. Alexandre

Transfer Learning Using Rotated Image Data to Improve Deep Neural

Network Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
Telmo Amaral, Luís M. Silva, Luís A. Alexandre, Chetak Kandaswamy,
Joaquim Marques de S˜, and Jorge M. Santos

Catalogue-Based Traffic Sign Asset Management: Towards User’s

Effort Minimisation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
Kelwin Fernandes, Pedro F.B. Silva, Lucian Ciobanu, and Paulo Fonseca

Scalable Prototype Learning Using GPUs . . . . . . . . . . . . . . . . . . . . . . . . . 309

Tonghua Su, Songze Li, Peijun Ma, Shengchun Deng, and Guangsheng Liang

Fuzzy Integral Combination of One-Class Classifiers Designed

for Multi-class Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
Bilal Hadjadji, Youcef Chibani, and Hassiba Nemmour
XVI Contents – Part I

Automatic Classification of Human Body Postures Based on Curvelet

Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329
N. Zerrouki and A. Houacine

QR Code Localization Using Boosted Cascade of Weak Classifiers . . . . . . . 338

P¯ter Bodn˜r and L˜szl˘ G. Ny˙l

Document Image Analysis

Using Scale-Space Anisotropic Smoothing for Text Line Extraction

in Historical Documents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
Rafi Cohen, Itshak Dinstein, Jihad El-Sana, and Klara Kedem

Multi-script Identification from Printed Words . . . . . . . . . . . . . . . . . . . . . . 359

Saumya Jetley, Kapil Mehrotra, Atish Vaze, and Swapnil Belhe

Segmentation-Free Keyword Retrieval in Historical Document Images . . . . . 369

Irina Rabaev, Itshak Dinstein, Jihad El-Sana, and Klara Kedem

Character-Level Alignment Using WFST and LSTM for Post-processing

in Multi-script Recognition Systems - A Comparative Study . . . . . . . . . . . . 379
Mayce Al Azawi, Adnan Ul Hasan, Marcus Liwicki, and Thomas M. Breuel

Handwritten and Printed Text Separation: Linearity

and Regularity Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
Sameh Hamrouni, Florence Cloppet, and Nicole Vincent

Parallel Layer Scanning Based Fast Dot/Dash Line Detection Algorithm

for Large Scale Binary Document Images . . . . . . . . . . . . . . . . . . . . . . . . . 395
Chinthaka Premachandra, H. Waruna H. Premachandra,
Chandana D. Parape, and Hiroharu Kawanaka

A Hybrid CRF/HMM Approach for Handwriting Recognition . . . . . . . . . . . 403

Gautier Bideault, Luc Mioulet, Cl¯ment Chatelain, and Thierry Paquet

Image and Video Retrieval

Exploring the Impact of Inter-query Variability on the Performance

of Retrieval Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
Francesco Brughi, Debora Gil, Llorenç Badiella, Eva Jove Casabella,
and Oriol Ramos Terrades

Relevance Assessment for Visual Video Re-ranking . . . . . . . . . . . . . . . . . . 421

Javier Aldana-Iuit, Ondrıej Chum, and Jirıi Matas
 Contents – Part I XVII

Remote Sensing

Delineation of Martian Craters Based on Edge Maps and Dynamic

Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
Jorge S. Marques and Pedro Pina

Automatic Interpretation of Remotely Sensed Images for Urban

Form Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
John Mashford, Felix Lipkin, Charlelie Olie, Mailys Cuchennec,
and Yong Song

Image Mosaicing by Camera Pose Estimation Based on Extended

Kalman Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
Alper Yildirim and Mustafa Unel

Applications

A Fast Plain Copy-Move Detection Algorithm Based on Structural Pattern

and 2D Rabin-Karp Rolling Hash . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
Kuznetsov Andrey Vladimirovich and Myasnikov Vladislav Valerievich

Automatic Annotation of an Ultrasound Corpus for Studying

Tongue Movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
Samuel Silva and Ant˘nio Teixeira

Improving Fire Detection Reliability by a Combination of Videoanalytics . . . 477

Rosario Di Lascio, Antonio Greco, Alessia Saggese, and Mario Vento

Automatic Method for Visual Grading of Seed Food Products . . . . . . . . . . . 485

Pierre Dubosclard, Stanislas Larnier, Hubert Konik, Ariane Herbulot,
and Michel Devy

Weight Estimation of Pigs Using Top-View Image Processing . . . . . . . . . . . 496

Mohammadamin Kashiha, Claudia Bahr, Sanne Ott, Christel P.H. Moons,
Theo A. Niewold, Frank O. Ödberg, and Daniel Berckmans

An Efficient Image Self-recovery and Tamper Detection

Using Fragile Watermarking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504
Sajjad Dadkhah, Azizah Abd Manaf, and Somayeh Sadeghi

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515

 Contents – Part II

Action, Gestures and Audio-Visual Recognition

Audio-Visual Emotion Analysis Using Semi-Supervised Temporal Clustering

with Constraint Propagation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Rodrigo Araujo and Mohamed S. Kamel

Exemplar-Based Human Action Recognition with Template Matching

from a Stream of Motion Capture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
Daniel Leightley, Baihua Li, Jamie S. McPhee, Moi Hoon Yap,
and John Darby

A New Visual Speech Recognition Approach for RGB-D Cameras . . . . . . . 21

Ahmed Rekik, Achraf Ben-Hamadou, and Walid Mahdi

2D Appearance Based Techniques for Tracking the Signer Configuration

in Sign Language Video Recordings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Ville Viitaniemi, Matti Karppa, and Jorma Laaksonen

Computer Aided Hearing Assessment: Detection of Eye Gesture Reactions

as a Response to the Sound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
A. Ferna¤ndez, Marcos Ortega, and Manuel G. Penedo

Multi-sensor Acceleration-Based Action Recognition . . . . . . . . . . . . . . . . . 48

Florian Baumann, Irina Schulz, and Bodo Rosenhahn

Incremental Learning of Hand Gestures Based on Submovement Sharing . . . 58

Ryo Kawahata, Yanrung Wang, Atsushi Shimada, Takayoshi Yamashita,
and Rin-ichiro Taniguchi

Gait Analysis from Video: Camcorders vs. Kinect . . . . . . . . . . . . . . . . . . . 66

Hoang Anh Nguyen and Jean Meunier

Biometrics

Person Re-identification Using Region Covariance in a Multi-feature

Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Volker Eiselein, Gleb Sternharz, Tobias Senst, Ivo Keller,
and Thomas Sikora

Multi-biometric Score-Level Fusion and the Integration of the Neighbors

Distance Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Naser Damer and Alexander Opel
XX Contents – Part II

Adaptive Haar-Like Features for Head Pose Estimation. . . . . . . . . . . . . . . . 94

Nam-Jun Pyun, Halima Sayah, and Nicole Vincent

Face and Palmprint Recognition Using Hierarchical Multiscale Adaptive

LBP with Directional Statistical Features. . . . . . . . . . . . . . . . . . . . . . . . . . 102
Ghada Shams, Mohamed Ismail, Sohier Bassiouny, and Nagia Ghanem

Multispectral Iris Recognition Using Patch Based Game Theory . . . . . . . . . 112

Foysal Ahmad, Kaushik Roy, and Khary Popplewell

Medical Image Processing and Analysis

Periodic Background Pattern Detection and Removal for Cell Tracking . . . . 123
Tiago Esteves, ¢ngela Carvalho, Fernando Jorge Monteiro,
and Pedro Quelhas

Nerve Detection in Ultrasound Images Using Median Gabor Binary Pattern. . . 132
Oussama Hadjerci, Adel Hafiane, Pascal Makris, Donatello Conte,
Pierre Vieyres, and Alain Delbos

Automatic Localization of Skin Layers in Reflectance Confocal Microscopy . . . 141

Eduardo Somoza, Gabriela Oana Cula, Catherine Correa,
and Julie B. Hirsch

Thermal Signature Using Non-redundant Temporal Local

Binary-Based Features. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
Adnan Al Alwani, Youssef Chahir, and Francois Jouen

Image Warping in Dermatological Image Hair Removal . . . . . . . . . . . . . . . 159

Alexandra Nasonova, Andrey Nasonov, Andrey Krylov, Ivan Pechenko,
Alexey Umnov, and Natalia Makhneva

3D Multimodal Visualization of Subdural Electrodes with Cerebellum

Removal to Guide Epilepsy Resective Surgery Procedures. . . . . . . . . . . . . . 167
N˜dia Moreira da Silva, Ricardo Rego, and João Paulo Silva Cunha

Medical Image Segmentation

On the Automatic Normalization of Plaque Regions in Ultrasound Images

of the Carotid. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
Jos¯ Rouco, Jorge Novo, and Aur¯lio Campilho

Automatic Tear Film Segmentation Based on Texture Analysis

and Region Growing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
Beatriz Remeseiro, Katherine M. Oliver, Eilidh Martin, Alan Tomlinson,
Daniel G. Villaverde, and Manuel G. Penedo
 Contents – Part II XXI

An Improved Segmentation Method for Non-melanoma Skin Lesions

Using Active Contour Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
Qaisar Abbas, Irene Fond˘n, Auxiliadora Sarmiento, and M. Emre Celebi

Statistical-Based Segmentation of Bone Structures via Continuous

Max-Flow Optimization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
Jose-Antonio P¯rez-Carrasco, Carmen Serrano-Gotarredona,
Cristina Su˜rez-Mejías, and Begon‹a Acha-Pin‹ero

A Portable Multi-CPU/Multi-GPU Based Vertebra Localization

in Sagittal MR Images. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Mohamed Amine Larhmam, Sidi Ahmed Mahmoudi, Mohammed Benjelloun,
Sa½d Mahmoudi, and Pierre Manneback

An Automated Level-Set Approach for Identification of Aortic Valve Borders

in Short Axis Windows of Transesophageal Echo Sequences (TEE) . . . . . . . 219
C¯sar Veiga, Francisco Calvo, Emilio Paredes-Gal˜n, Pablo Pazos,
Carlos Pen‹a, and Andr¯s Ín‹iguez

Reliable Lung Segmentation Methodology by Including Juxtapleural Nodules . . . 227

J. Novo, J. Rouco, A. Mendonça, and Aur¯lio Campilho

Computer-Aided Diagnosis

Fully Automatic 3D Glioma Extraction in Multi-contrast MRI . . . . . . . . . . . 239

Pavel Dvorak and Karel Bartusek

Grading Cancer from Liver Histology Images Using Inter and Intra Region
Spatial Relations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
Micka˝l Garnier, Maya Alsheh Ali, Johanne Seguin, Nathalie Mignet,
Thomas Hurtut, and Laurent Wendling

eFis: A Fuzzy Inference Method for Predicting Malignancy of Small

Pulmonary Nodules. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
Aydın Kaya and Ahmet Burak Can

Degradation Adaptive Texture Classification: A Case Study in Celiac

Disease Diagnosis Brings New Insight . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
Michael Gadermayr, Andreas Uhl, and Andreas V¯csei

Retinal Image Analysis

Optic Disk Localization for Gray-Scale Retinal Images

Based on Patch Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
F. Sattar, Aur¯lio Campilho, and M. Kamel
XXII Contents – Part II

Automatic Optic Disc Detection in Retinal Fundus Images

Based on Geometric Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
Isabel N. Figueiredo and Sunil Kumar

Optic Nerve Head Detection via Group Correlations in Multi-Orientation

Transforms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293
Erik Bekkers, Remco Duits, and Bart ter Haar Romeny

A Robust Algorithm for Optic Disc Segmentation from Colored

Fundus Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
Anam Usman, Sarmad Abbas Khitran, M. Usman Akram,
and Yasser Nadeem

Coupled Parallel Snakes for Segmenting Healthy and Pathological Retinal

Arteries in Adaptive Optics Images. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
Nicolas Lerm¯, Florence Rossant, Isabelle Bloch, Michel Paques,
and Edouard Koch

Automatic Arteriovenous Nicking Identification by Color Fundus Images

Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
Carla Pereira, Diana Veiga, Luís Gonçalves, and Manuel Ferreira

Detection of Hemorrhages in Colored Fundus Images

Using Non Uniform Illumination Estimation . . . . . . . . . . . . . . . . . . . . . . . 329
M. Usman Akram, Sarmad Abbas Khitran, Anam Usman,
and Ubaid ullah Yasin

Automatic Robust Segmentation of Retinal Layers in OCT Images

with Refinement Stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
Ana Gonz˜lez-L˘pez, Marcos Ortega, Manuel G. Penedo,
and Pablo Charl˘n

3D Imaging

Accurate Multi-View Stereo 3D Reconstruction for Cost-Effective

Plant Phenotyping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
Lu Lou, Yonghuai Liu, Jiwan Han, and John H. Doonan

Truncated Signed Distance Function: Experiments on Voxel Size. . . . . . . . . 357

Diana Werner, Ayoub Al-Hamadi, and Philipp Werner

Human Activity Analysis in a 3D Bird’s-eye View. . . . . . . . . . . . . . . . . . . 365

Gang Hu, Derek Reilly, Ben Swinden, and Qigang Gao

3D Spatial Layout Propagation in a Video Sequence. . . . . . . . . . . . . . . . . . 374

Alejandro Rituerto, Roberto Manduchi, Ana C. Murillo, and J.J. Guerrero
 Contents – Part II XXIII

SASCr3: A Real Time Hardware Coprocessor for Stereo Correspondence . . . 383

Luca Puglia, Mario Vigliar, and Giancarlo Raiconi

Motion Analysis and Tracking

Adaptive Feature Selection for Object Tracking with Particle Filter . . . . . . . 395
Darshan Venkatrayappa, D¯sir¯ Sidib¯, Fabrice Meriaudeau,
and Philippe Montesinos

Exploiting Color Constancy for Robust Tracking Under Non-uniform

Illumination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
Sinan Mutlu, Samuel Rota Bul˚, and Oswald Lanz

Wavelet Subspace Analysis of Intraoperative Thermal Imaging

for Motion Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
Nico Hoffmann, Julia Hollmach, Christian Schnabel, Yordan Radev,
Matthias Kirsch, Uwe Petersohn, Edmund Koch, and Gerald Steiner

A Spatio-temporal Approach for Multiple Object Detection in Videos

Using Graphs and Probability Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
Henrique Morimitsu, Roberto M. Cesar, and Isabelle Bloch

Robot Vision

Adopting Feature-Based Visual Odometry for Resource-Constrained

Mobile Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
Michað Fularz, Michað Nowicki, and Piotr Skrzypczyn¤ski

Strategy for Folding Clothing on the Basis of Deformable Models . . . . . . . . 442

Yasuyo Kita, Fumio Kanehiro, Toshio Ueshiba, and Nobuyuki Kita

Multiple Camera Approach for SLAM Based Ultrasonic Tank Roof Inspection . . . 453
Christian Freye, Christian Bendicks, Erik Lilienblum,
and Ayoub Al-Hamadi

On Tracking and Matching in Vision Based Navigation . . . . . . . . . . . . . . . 461

Adam Schmidt, Marek Kraft, and Michað Fularz

Biologically Inspired Vision for Indoor Robot Navigation . . . . . . . . . . . . . . 469

M. Saleiro, K. Terzic¤, D. Lobato, J.M.F. Rodrigues, and J.M.H. du Buf

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479

Image Representation and Models
Path Descriptors for Geometric Graph Matching
and Registration

Miguel Amável Pinheiro(B) and Jan Kybic

Center for Machine Perception, Department of Cybernetics, Faculty of Electrical

Engineering, Czech Technical University in Prague, Prague, Czech Republic
amavemig@cmp.felk.cvut.cz

Abstract. Graph and tree-like structures such as blood vessels and neu-
ronal networks are abundant in medical imaging. We present a method
to calculate path descriptors in geometrical graphs, so that the similar-
ity between paths in the graphs can be determined efficiently. We show
experimentally that our descriptors are more discriminative than exist-
ing alternatives. We further describe how to match two geometric graphs
using our path descriptors. Our main application is registering images
for which standard techniques are inefficient, because the appearance
of the images is too different, or there is not enough texture and no
uniquely identifiable keypoints to be found. We show that our approach
can register these images with better accuracy than previous methods.

1 Introduction
Blood vessels, nerve fibers or pulmonary airways are examples of biological struc-
tures that can be represented as geometrical graphs with nodes corresponding
to branching points and edges corresponding to curves connecting the branching
points (Fig. 1). We consider the problem of registering two 2D or 3D images
based on a common geometric graph structure both images contain and which
we assume to be already extracted (e.g. [1]). This approach has the potential of
being much faster than standard pixel-based image registration techniques [2]
and tolerate very different image appearances. With respect to key point regis-
tration methods [3], registering geometric graphs provides more clues.
Unlike most existing approaches, our method can in a reasonable time handle
rather general transformations and large displacements, as well as partial over-
laps. The key contribution over our previous work [4,5] is an alternative coarse
alignment step based on finding similarities between path descriptors to restrict
the set of possible correspondences and thus make the matching more efficient.

2 Related Work
Ignoring edges, the geometric graph registration becomes a point cloud matching
problem, which can be solved by RANSAC-like approaches [6,7] These methods

c Springer International Publishing Switzerland 2014
A. Campilho and M. Kamel (Eds.): ICIAR 2014, Part I, LNCS 8814, pp. 3–11, 2014.
DOI: 10.1007/978-3-319-11758-4 1
4 M.A. Pinheiro and J. Kybic

(a) (b) (c) (d)

Fig. 1. Example registration problems with prominent geometric graph structures:

Blood vessels in brain tissue acquired using (a) two photon microscopy and (b) bright-
field optical microscopy; two-photon microscopy images of axons in the brain (c,d)

Fig. 2. Example of the representation and notation for a graph with K = 2

do not need initialization but only work for restricted class of transformations,
such as rigid or affine. On the other hand ICP-like approaches [8,9] can handle
nonlinear transformations but require a good initialization. Finally, the matching
can be viewed as a discrete optimization problem with cost functions assigned
to node or edge pairings [10–12], which is very powerful but computationally
demanding. Pruning the search space is important to increase efficiency [4,5]
and node and edge descriptors [13,14] can be used for that purpose.

3 Problem Definition
Let us have an undirected graph GA = (VA , EA ) with nodes VA = {v1A ,
. . . , v|V
A
A | } and edges E
A
= {eA
1 , . . . , e|EA | }. We represent each edge ek ∈ E
A A A

as a cubic B-spline [15], limiting the generality of the representable shapes in

exchange to better robustness with respect to noise. In doing so, we now have a
continuous representation of each edge, which can be represented by a mapping
ξeA k
: [0, 1] → RD , where if eA
k = (vi , vj ), then ξeA
A A
k
(0) = viA and ξeA k
(1) = vjA .
We define superedges SA of graph GA as paths of at most K consecutive
edges. Similarly to edges, we can now define a mapping ξsA k
: [0, 1] → RD , which
will define the path of each superedge. Superedges are needed to deal with the
case of approximative matching, where some nodes and edges are detected only
in one of the graphs. Our graph can then be represented by G A = (VA , SA ), i.e.
a set of nodes and a set of superedges.
 Path Descriptors for Geometric Graph Matching and Registration 5

Fig. 3. Example of sampling vectors on two curves

Let us now introduce a second graph G B = (VB , SB ), which is related to

G by a matching MA→B = (MV , MS , T ), where MV : V → V is a mapping
A

between nodes and MS : S → S is a mapping between superedges. The transfor-

mation T provides us a mapping of both VA and SA to VB and SB respectively,
in RD . The elements of MA→B are not independent. In fact, if one of them is
given, it is possible to at least approximate the remaining two. The task is there-
fore to find MA→B . We will solve this task by finding similarities between the
superedges SA and SB , using path descriptors to be defined in Section 4.

4 Path Descriptors
The geometrical transformation between images in biomedical applications can
usually be decomposed into a rigid motion plus a mildly non-linear component.
Typically, the scale is known from the acquisition parameters and we can consider
it is equal to zero, without loss of generality. Due to mechanical properties of
the tissue, the nonlinear component is small and we therefore assume that for
all x, y ∈ RD , there is a bound εT on the relative length change, such that
1      
d x, y ≤ d T (x), T (y) ≤ (1 + εT )d x, y , (1)
1 + εT
holds for a transformation T , where d(x, y) is the Euclidean distance and εT is
a deformation parameter, i.e. it is bi-Lipschitz.
We propose to use path descriptors – a vector characterizing each curve asso-
ciated with a geometric graph edge. Unlike previously suggested curve descrip-
tors [13,14,16], we have a good estimate of the change of descriptor values under
our transformation model. Let us have a sampling vector ω = (ω0 , . . . , ωnω +1 ),
such that 0 = ω0 < ω1 < · · · < ωnω < ωnω +1 = 1. Given a geometric path
ξsk : [0, 1] → Rd (Section 3) of a superedge sk , we calculate its descriptor param-
eterized by ω:

nω
  nω
hω (sk ) = d ξsk (ωi ), ξsk (ωi+1 ) = ξsk (ωi+1 ) − ξsk (ωi ). (2)
i=0 i=0

In plain words, we resample the path in nω + 2 points, pass a piecewise linear

approximation through the points and calculate the length of this approximation.
For an allowable transformation (1), we assume that
1  
hω (sk ) ≤ hω T (sk ) ≤ (1 + εh )hω (sk ), (3)
1 + εh
6 M.A. Pinheiro and J. Kybic

for an εh close to εT . This holds as long as the transformation T is not too

far from a rigid body transformation. Given a set of sampling vectors Ω =
(ω 1 , . . . , ω
 |Ω| ), we can calculate a vector of descriptors hΩ (sk ) = hω1 (sk ), . . . ,
hω|Ω| (sk ) . Given a large enough size of Ω, the value of the vector hΩ (sk ) will
describe the geometric disposition of the superedge in RD .
Given two geometric graphs G A = (VA , SA ) and G B = (VB , SB ), we say
that two superedges sA k ∈ S and sl ∈ S
A B B
are compatible with respect to hΩ ,
if (3) holds for all ω ∈ Ω,
1  B
hΩ (sAk ) ≤ hΩ sl ≤ (1 + εh )hΩ (sA
k ), (4)
1 + εh
where the multiplication and comparison is done element by element.

5 Finding a Global Solution

To find a solution based on our descriptors we formalize our problem as an
integer quadratic program (IQP) [10,17]. Given the graphs G A = (VA , SA ) and
G B = (VB , SB ) we define an affinity matrix W with size |VA |·|VB |×|VA |·|VB |
and elements Wik;jl = exp(−||hΩ (sA a ) − hΩ (sb )||/σ ) similarly as in [10,17], if
B 2

sa = (vi , vj ) ∈ S , sb = (vk , vl ) ∈ S and the superedge pair (sA

A A A A B B B B
a , sb ) is
B

compatible, i.e. if eq. (4) holds. Otherwise Wik;jl = 0. If matching node viA with
vkB is consistent with matching node vjA with vlB , Wik;jl is high and vice versa.
The matching is represented by a binary vector x∗ , such that xij = 1 iff
nodes viA and vjB match, maximizing the total affinity
A
||VB |
x∗ = arg max x Wx s.t. x ∈ [0, 1]|V , (5)
x
|U | |V |
 
∀j xij ≤ 1, ∀i xij ≤ 1.
i=1 j=1

Furthermore, there are constraints on node positions:

1
∀xik , xjl = 0, ||vA − vjA || ≤ ||vkB − vlB || ≤ (1 + εT )||viA − vjA ||. (6)
1 + εT i
We use the Reweighted Random Walks method [17] to find an approximate
solution x̃∗ of this NP-hard problem [18]. In order to find the required binary
vector x∗ , we use the Hungarian algorithm [19] to calculate the best assignment
using the weights of x̃∗ . We iteratively select the best individual assignment
which does not contradict the previously selected ones and also (6).
This may give us a partial mapping MV , with only a subset of the nodes
matched. In order to complete and refine MV and also to match the paths
between nodes, we use a fine alignment algorithm described in [5]. This technique
first uses MV to predict the elastic transformation T represented by a Gaussian
process model. Given T , it then uses the Hungarian algorithm to calculate the
optimal matching MV between nodes as well as the matching between other
points on the edges, repeating until convergence.
 Path Descriptors for Geometric Graph Matching and Registration 7

6 Experiments and Results

6.1 Datasets
We test our path descriptors and global matching with datasets from various
applications in medical imaging, featuring graph-like structures to be registered.
In retinal fundus imaging (see Fig. 4(a)), the registration of different frames
taken from various views helps build a single view of the retinal fundus [20].
In neuroscience, to better understand the learning of cognitive functions, images
in vivo of the axons in the brain of a mouse are acquired before and after a learning
task using 2-photon microscopy [21]. Due to the complexity and small (but crucial)
changes in the images (see Fig. 4(b)), the registration procedure helps identifying
the differences between the structures. The registration of images acquired with
different modalities such as electron and light microscopy (see Fig. 4(c)) is helpful
to have a better understanding of the neuronal network [12].
In angiography (see Fig. 4(d)), image registration helps tracking the displace-
ment of the blood vessels in the heart, during heart cycle. The registration of
images of blood vessels in the brain acquired using different imaging techniques,
such as optical and 2-photon microscopy (see Fig. 4(e)) helps find details which
are present in only one of the acquisition techniques [4].

6.2 Path Descriptors

To validate the path descriptors, we take two graphs, G A = (VA , SA ) and G B =
(VB , SB ) and calculate the path descriptors hΩ for all superedges in both graphs.
For each pair of superedges sA k ∈ S
A
and sB l ∈ S
B
we determine if they are
compatible (eq. (4)) with respect to hΩ for different εh .
In Fig. 4, we show the ROC curves for various previously described datasets.
To obtain the ground truth, we assume a superedge sA k is a true match of
sB
l if both their end nodes match. Apart from the proposed path descriptor,
we tested 3-D Curve Matching [16], Determining the Similarity of Deformable
Shapes [13], and Curve Matching using Fast Marching [14]. We have varied εh for
the proposed method, the deformation cost for [13,14] and the residual Euclidean
distance between the matched curves for [16]. Clearly, the proposed descriptor
obtains the best performance in all tested datasets.

6.3 Global Matching

We used the methodology described in Sec. 5 to find the match M between the
given graphs G A and G B . In Fig. 5, we depict the results of applying this final
alignment, based on the proposed path descriptors.
In Table 1, we show the average Euclidean distance between true matches
of the registered graphs and the respective processing times using different
approaches, namely CPD [9], IPFP [10] and ATS [5], with recommended param-
eters. In most cases, our proposed method presents the smallest error of all
tested methods. The average error is close to the performance of ATS, however
accomplished in a much faster time.
Another random document with
no related content on Scribd:
 The man, however, who has crowned with his acumen the
written science of war is Jomini, who first became known as a young
staff-officer of Marshal Ney’s, and died but twenty years ago. Though
he rose to the highest rank in the Russian service, his career was as
military adviser rather than as commander. His chief value to us lies
in his having collated and so plainly set down the lessons taught by
the great captains, particularly Frederick and Napoleon, that all may
now study them, as during the last century they could not be studied,
—were not even understood. He has enabled us to assimilate the
history of war. Other military students have since written with equal
profundity. But our debt to Jomini is not lessened thereby.
Gustavus Adolphus was born in Stockholm, in 1594, the son of
Charles IX. of Sweden, but at a time when his cousin Sigismund III.
occupied the throne. He was a lad of great personal beauty and
strength, and his naturally bright mind profited well by the careful
training he received. His boyhood showed all the traits of strong
earnestness, clean-cut courage, and deep religious feeling which
later characterized the Champion of the Reformation. Of naturally
quick temper, in youth a blow followed a word; in manhood he
acquired exceptional self-control. His education was largely under
the direction of Oxenstiern, who later became his prime-minister,
general, and greatest intimate. He was a constant reader, an
eloquent and persuasive speaker, a poet whose religious verses are
still sung in every household of Sweden. He was famous in athletics,
and was both a noted rider and able swordsman.
The Swedish government was an elective-hereditary monarchy.
Sigismund, a bigoted Catholic, was deposed when Gustavus was
ten years old, and the lad’s father made king. Sigismund retired to
Poland, of which country he was also monarch, and remained
thereafter the sworn enemy of Charles and of Gustavus.
The young prince went through every step of military rank and
training, and at seventeen was declared of age and participated with
distinguished credit, and rare skill and enterprise, in a war with
Denmark. In this same year (A.D. 1611) his father died, and, against
all precedent, Gustavus Adolphus was chosen king. During his reign
of twenty-one years, his people and he were an unit. The world has
never seen a more striking instance of mutual love and confidence,
justly earned, between king and people.
Sweden was at war with Denmark, Russia, and Poland.
Gustavus determined to finish each war, if possible, singly and in
turn. From the very beginning he showed in his military conduct that
his intelligence ranged beyond the conventional method of
conducting war, which he had been taught with so much care. In
1613 he conquered a peace with Denmark.
In 1614 he began war with Russia, making, meanwhile, a two
years’ truce with Poland. In this year, and the next, he drew the
attention of all Europe to his bold invasion of the Russian territory, at
the point where now stands St. Petersburg, and was for the first time
approached by the Protestants of Germany with a request to aid
their cause. In 1617 Gustavus conquered a peace with Russia.
Sigismund would not hear of peace, but under the curious habit
of that day, of conducting war on a sort of picnic system, he did
extend the existing truce for five years. At its expiration, in 1621,
active war began. Gustavus, with twenty-four thousand men, making
Livonia his objective, landed at Riga, took the place, and from thence
as a base, conducted his campaign.
Sigismund represented the Catholic element; Gustavus was the
most prominent Protestant prince, and as such received many
urgent petitions for help from the harassed Protestants of Germany.
The eventual necessity of taking a share in the religious war was
clearly foreseen by Sweden. With the advice and consent of the
ministry and people, Gustavus reorganized the army and created a
distinctly national force of eighty thousand men, and based its
discipline and character on the most intelligent foundation. Sweden
thus acquired the first modern regular military organization. Other
nations, as a rule, whenever a war was imminent, raised troops from
the crowds of soldiers of fortune, with whom all Europe swarmed,
and discharged them after its expiration. The Swedish organization
consisted of one-quarter regular troops for service out of the country,
and three-quarters landwehr for the defence of the Fatherland and
for filling gaps in the regulars. Recruitment was by districts on a well-
settled plan of quotas. The troops in service and the militia were
scrupulously drilled and taught, uniformed, well armed and fed, and
regularly paid.
The Polish war lasted until 1629, the campaigns being annual,
but varying in scope. Gustavus invariably took the offensive, and
was habitually successful. He was always head and front of every
movement, full of intelligence, activity, and courage, ran constantly
great personal danger, and suffered from frequent wounds. No
character of modern history exhibits the qualities of the ancient hero
so distinctly as Gustavus Adolphus. Cautious and intelligent to a
marked degree in his campaigns, he was in battle a very Alexander
for audacity and chivalrous bearing. Always in the thickest of the fray,
he led his men in person, and, despite the protests of his generals
and suite, could never be restrained from exposing himself at the
point of greatest importance. He was unwisely reckless of his own
safety, though never losing for a moment his cool calculation or
power to gauge the situation. His army partook his enthusiasm, as it
shared his earnest religious feeling, and was devotedly attached to
him as man and king.
In 1628, Wallenstein, the distinguished commander of the
Imperial forces, had won great success in northern Germany, and
had laid siege to Stralsund. The German Protestants again turned
with piteous appeals to Gustavus. The king well knew that sooner or
later Protestant Sweden must, in self-defence, enter the lists against
the Catholic Empire, and threw a Swedish garrison into Stralsund,
which, gallantly backed by the citizens, held the place against
Wallenstein’s best efforts.
In the campaign of 1629, the Emperor sent an army to reënforce
the Poles. This the more impelled Gustavus to actively embrace the
Protestant cause. At the end of this campaign, Sigismund, largely
under the influence of Richelieu, was prevailed on to agree to a six
years’ truce. France could not openly join the Protestants in their
struggle against the Catholic Emperor, but was glad to see Gustavus
do so in order to check such success by Ferdinand as might disturb
the balance of power.
 This truce ended the Swedish-Polish wars, which had lasted
eight years (1621–1629). The king had conducted six campaigns
against Poland, and two against Denmark and Russia. These were
to him what the Gallic campaigns were to Cæsar, a practical school
of war, in which both he could learn his trade, and his army be
disciplined and toughened. He had observed the practical working of
his new army organization, and learned the weak points of the
existing system of war. Comparison showed the advantages of his
own conceptions. In the three remaining years of his life he moulded
these into a new art, which pointed the way back to a system full of
intellectual and moral force as well as more consonant with common
Christian charity. The king, during this period, gleaned varied
experience. He learned the habits of different leaders and armies,
and how to adapt his own ways to theirs. His infantry underwent a
good schooling. His cavalry he gradually improved by imitating the
admirable Polish horsemen, and by adding discipline and ensemble
to it. His artillery gave a good account of itself. Under Gustavus’
careful eye, every branch of the service during these campaigns
grew in efficiency. Equipment, arms, rationing, medical attendance,
drill and discipline, field-manœuvres, camp and garrison duty,
reached a high grade of perfection. Each year added to the skill and
self-poise of the Swedish forces. They were distinctly superior to any
European army of the day.
Not only had Gustavus learned to know his generals and men,
but these had gauged their king. There had arisen between them
that mutual confidence, esteem, and affection which only great souls
ever evoke and keep. And as there was no danger or labor of which
Gustavus did not bear with them his equal part, so the Swedish army
saw in its king a harbinger of victory, a sure protection in disaster.
Gustavus’ own character, his bravery, religious ardor, honesty, and
humanity infused itself into every soldier in the Swedish ranks.
Gustavus Adolphus was now in a position to afford efficient aid
to the German Protestants. The efforts of the latter had been noble,
but far from systematic, and they were fast being driven to the wall.
The war had been marked by barbarities characteristic of religious
struggles, and by the adoption of happy-go-lucky plans of campaign.
Armies had moved into a province, not because it was strategically
important, but because it was rich in plunder. Manœuvres were
conducted without reference to base or communications. There was
no aim beyond temporary expediency in any one’s movements. A
fortress would arrest the march of an army, which would sit down
before it without the remotest conception of whether its capture
would have an effect on the general result. Lack of system was
supplemented by religious fanaticism, which made everything
redolent of atrocity. No general but was characterized by some
fearful vice. Gustavus Adolphus was destined to change all this in a
short two years.
As a soldier Gustavus is less noted for his battles than for the
conduct, in 1630, 1631, and 1632, of a campaign on one broad,
intelligent, far-seeing plan, from which he never swerved. This of
itself was an entire novelty in this period of shallow operations. In
lieu of detailing one of his manœuvres, I will give a hasty sketch of
his entire plan of campaign in Germany. This was the first crisply
strategic series of operations since the days of Cæsar.
It was clear that if the Emperor overcame the Protestants of
Germany he would turn on Sweden. To await attack was the
preference of the Swedish ministry. But Gustavus pointed out the
advantages of an immediate offensive war in Germany. The struggle
would be kept from Swedish territory. The Emperor would not gain
so much headway as to lay Sweden open to an exhausting war.
They owed a duty to their oppressed Protestant brethren. He
convinced his people and gained their support. He took with him
fifteen thousand men. This number he expected to, and did in fact,
largely increase in Germany by recruitment and the aid of Protestant
allies.
Gustavus landed in Rügen in June, 1630. He added five
thousand men of the Stralsund garrison to his army, and took
possession of all the islands at the mouths of the Oder. He then
captured Stettin and extended his grasp right and left along the
coast. He proposed to base himself on the Baltic, as Alexander had
done on the Mediterranean. He took and garrisoned many seaboard
towns and others lying not far inland. His army, reënforced by
German allies and recruitment, soon rose to twenty-five thousand
men, and he established a firm footing on the Oder, which river was
an excellent line for operations into the heart of Germany. The
imperial Field Marshal Conti, who had ten thousand men in his front,
was unable to interfere with his operations. Garrisoning Stettin,
Gustavus moved into Mecklenburg to encourage its Protestant
princes, further secure his base, increase his supplies and forces,
and gain active allies. He relied on collecting seventy to eighty
thousand men. Count Tilly had been put in supreme command of the
Imperial forces, in place of Wallenstein, against whom the Catholic
princes had conceived a marked prejudice. This resulted in
disbanding a large part of Wallenstein’s soldiers, who considered
themselves only in his personal service, and left Ferdinand for the
nonce but unimportant armies to oppose to the Swedish advance.
 CAMPAIGN OF GUSTAVUS ADOLPHUS IN
GERMANY, 1630–1–2
Having substantially rescued Mecklenburg from the Imperialists,
Gustavus left a force to operate there and returned to Stettin,
purposing to move with the main army up the Oder (Dec., 1630). The
end of the year was at hand. The Imperial army in his front was in no
condition for a winter campaign, either from habit, discipline, or
equipment. For this very reason Gustavus moved against it, his own
troops being well clad and equipped, and inured to cold. He soon
drove the enemy back to the line of the Warta, and then sat down in
an entrenched camp at Bärwalde till he could recruit his army up to a
standard equal to larger operations. The Protestant Elector of
Brandenburg meanly refused his help to the cause, but Catholic
France subsidized the king, and the Protestants called an assembly
at Leipsic to agree on new measures of defence.
 Tilly now appeared on the scene, thirty-four thousand strong.
The king had but twenty-five thousand men and would not risk a
battle, neither would Tilly assault the Bärwalde camp. But Gustavus
had a better scheme in his head. He planned to draw Tilly into
Mecklenburg, and then quickly return and capture the enemy’s line
on the Warta. He made forced marches into that province, fell on the
Imperialists and again defeated them. Tilly, alarmed, followed with
twenty-four thousand men. Gustavus, by occupying the direct road,
had compelled Tilly to resort to a long circuit. When Tilly was fairly on
the way, Gustavus moved rapidly and secretly back to Stettin,
advanced on Frankfort, took it after a seventeen days’ siege, and
thus broke up the enemy’s line. The Warta fully protected his left
flank in advancing into Germany. Gustavus had completely baffled
his adversary. But Tilly took bitter revenge by the capture of
Magdeburg, which, though it cannot perhaps be charged to Tilly
himself, was given up to sack, and suffered a horrible fate at the
hands of his unbridled soldiery. Gustavus had been unable to cross
neutral Brandenburg to its assistance.
The barbarous treatment of Magdeburg enraged instead of
disheartening the Protestants. Two able allies, Hesse Cassel and
Saxony, joined the king’s train. And by able manœuvring, restless
energy, and clear-headed method he swept Pomerania and
Mecklenburg of Imperial troops.
The pusillanimous conduct of the Elector of Brandenburg, under
the plea of neutrality, finally constrained Gustavus to dictate terms to
him. He marched on Berlin and compelled the Elector to allow free
passage to the Swedes over his territory, as well as to refrain from
damaging the Protestant cause, if he would not help it.
Thus in one year from his landing in Germany, Gustavus had
occupied Pomerania and Mecklenburg, and had neutralized
Brandenburg (June, 1631). By holding the lines of the Havel, the
Spree, and the Oder, he controlled all the territory to the confines of
Poland and Silesia, and with a sufficiency of reënforcements he
could safely advance on central Germany.
 Tilly invaded Hesse Cassel. Gustavus tried a diversion to lure
him away from his new ally. Count Pappenheim opposed him at the
Elbe. Gustavus stole a clever march on him, crossed and went into
an entrenched camp near Werben. These entrenched camps, it will
be perceived, were a feature of this period which Gustavus still
affected. They continued in use until he himself in part, and
Frederick wholly, demonstrated that entrenchments could be taken
by vigorous assault. At this time it was considered the height of
foolhardiness to attack entrenchments.
Tilly vacated Hesse Cassel and moved on the Swedish camp.
Gustavus had but ten thousand men there; Tilly had twenty-seven
thousand; but the king waylaid Tilly’s isolated cavalry, handled it
roughly, and returned safely to camp. Tilly, despite his excess of
force, did not care to risk an assault. Large reënforcements soon
reached both armies. Gustavus’ diversion had accomplished all he
sought. By defending the line of the Elbe and Havel, he prevented
Tilly from making any compromising advance.
Tilly was ordered to Saxony. The cruelties here perpetrated by
his troops made the Elector all the better ally. He offered Gustavus
the support of his army of eighteen thousand men. The king again
crossed the Elbe, at Wittenberg, and joined the Saxons at Düben.
This gave him a force of forty thousand men, of which twelve
thousand were cavalry. Tilly had arrived at Leipsic, and promptly
advanced to meet Gustavus with thirty-two thousand under the
colors. But, at the battle of Breitenfeld, he suffered a stinging defeat,
with the loss of six thousand men.
Tilly’s soldiers were in action much what their commander was,
—a stiff, dense, unwieldy mass, still hide-bound in the Spanish
school, which won its way by mere weight of men in the old
phalangial manner. The Swedes were quite a different body.
Gustavus had reduced the number of their firing-ranks to three,
placed reliance on their individual intelligence, which was marked,
and had drilled his musketeers, as well as his gunners, to fire as
much more rapidly than the enemy, as Frederick’s men with their iron
ramrods, or the Prussians of this generation with their needle-guns.
In this, his first great battle, the result was, despite the ignominious
flight of the Saxons, predetermined by the condition of the respective
armies and their leaders. Here, as on all occasions, the king, in
personal conduct, was an Alexander in audacity; a Cæsar in
intelligence.
Gustavus Adolphus had been only fourteen months in Germany,
but he had by his broad, prescient, cautious, and well-digested
scheme, crowned by the victory of Breitenfeld, completely changed
the prospects of the Protestants. He had got a firm footing in
northern Germany, where he now held most of the strong places. He
had secured his communications with Sweden by the possession of
the sea. He had grown in strength by his treaties with Hesse Cassel
and Saxony, and by accessions of troops from all quarters. He had
gained enormously in moral weight, and his army in aplomb and
confidence. His operations had been slow and cautious,—though
rapid when measured by the times,—but they had been sure, and
were justified by the event. The late victory had placed him on a
totally different footing. The Catholic party no longer looked down on
the “Snow-king,” as Wallenstein had jeeringly called him. The
Imperial army had lost in spirit and organization that which he had
gained. Its present retreat to the Weser opened the heart of the
Emperor’s possessions to the king’s advance. The former’s authority
had received its first severe blow, and the Protestants of north and
west Germany, lately cowed into submission, now rose and joined
Gustavus’ standard. These fourteen months had shifted the moral
superiority from the Catholic to the Protestant cause. But the work
was far from ended. It required the same wise and cautious action,
coupled with vigor and intelligence, to complete what had been so
well begun.
The advisers of Gustavus strongly urged an advance on Austria,
believing that such a course would bring Ferdinand to terms. But so
far Gustavus’ successes had come from a systematic plan of
campaign which embraced the whole of Germany in its scope. He
had secured each step and had risked nothing unnecessarily. He
saw the chances pointed out, but he also saw that if he advanced
south, his right rear would be threatened by Tilly, who had, after his
defeat, retired toward the Rhenish provinces and there made a new
base. The king preferred his own plan of first gaining a firm footing in
western Germany. He held interior lines and saw that he could
operate against his enemies in detail. To complete his plan would
secure him from the lower Elbe to the middle and upper Rhine, and
he could then turn against Bavaria and Austria from the west, as his
advisers would now have him do from the north, and with distinctly
better effect. Meanwhile the Saxons could operate towards Silesia
and Bohemia to secure Gustavus’ left in his advance, and Hesse
Cassel could hold head against Tilly on Gustavus’ right. The scheme
was wise and far-sighted, took into calculation all the political and
military elements of the situation, and was based on broad, sound
judgment. For seventeen hundred years, no one had looked at war
with so large an intelligence.
It may be said that war is a game of risks. But to play a
gambler’s game was not Gustavus’ forte. When the occasion
demanded, he could disregard every danger. What he has taught us
is method, not temerity. His mission was to abolish the Quixotism of
his day.
The Saxon Elector, with a mixed army over twenty thousand
strong, accordingly marched into Bohemia and Silesia (Oct., 1631)
and pushed the Imperialists back from Prague on Tabor. Everything
promised success. But all at once the Elector appeared to lose heart,
arrested his advance, and opened negotiations with the Emperor,
who, seeing that threats had not succeeded, had tried conciliation.
This part of the operation was nullified.
Gustavus moved to Würzburg. Franconia joined the Protestant
cause as Thuringia had already done. Tilly, having recovered from
his late defeat, and his present position being no threat to Gustavus,
marched southerly. With allies he collected over fifty thousand men
and proposed to seek battle. But the Elector of Bavaria, fearful for
his territory, kept Tilly on the defensive.
Gustavus was now firmly established on the Main, and in
Thuringia and Franconia, and he presently moved down the river to
fully secure the Rhineland, leaving a sufficient force opposite Tilly in
Franconia. His men marched along both banks with the baggage on
boats. He crossed the Rhine, took Mainz and transformed it into an
allied fortress.
Germany was metamorphosed. The allies had one hundred and
fifty thousand men in the field. Recruiting was lively. All Protestants
were united in sentiment, purpose, and efforts. France was helpful in
keeping the Catholic princes along the Rhine in a condition of
neutrality, while Gustavus lay in a central position between the
Emperor and these same princes. Bavaria was an uncertain
element. The Emperor had a total of but eighty thousand men, and
of these the bulk were protecting the Danube instead of carrying
desolation into the Protestant territory.
Gustavus now concentrated on the middle Main to the number of
forty-five thousand men and marched on Nürnberg, where he was
received with enthusiasm. Tilly crossed the Danube and took up a
position over against Rain, behind the Lech, with forty thousand
effective. From Nürnberg Gustavus marched to Donauwörth, also
crossed the Danube and sent out a detachment to take Ulm.
The king was daring at the proper time. His whole campaign so
far had been cautious and systematic, neglecting no point in his
general scheme. He was now face to face with the army he had
driven from northern and western Germany, and was ready for
battle. He could not draw Tilly from his entrenched camp; and he
determined to impose on him by boldly crossing the river in his front
and attacking it,—then simply an unheard-of proceeding. He
believed that the moral advantage to be gained by a stroke of
audacity would more than compensate for the danger, and danger
was to Gustavus an incentive. He erected a battery of seventy-two
guns on the left bank of the Lech, opposite Rain, and under cover of
its fire set over a portion of the troops in boats, built in two days a
bridge and a bridge-head, led over the infantry, and sent the cavalry
up stream to ford the river above the enemy’s position (April, 1632).
Tilly and the Elector of Bavaria sought too late to interrupt these
fearless proceedings. They issued from their camp with a select
body of troops and attacked the Swedes, who were backing on the
Lech. But the crossfire of the admirably posted Swedish batteries
was severe; the Swedish infantry held its own, and the cavalry rode
down upon their flank. In this obstinate combat Tilly was mortally
wounded. His second in command suffered a like fate. The Imperial
troops lost heart and took refuge within their breastworks. Oncoming
darkness forestalled pursuit. But Gustavus had gained his object.
The Imperial army had lost morale and organization, and his own
had gained in abundant measure. This is the first instance of forcing
the passage of a wide and rapid river in the teeth of the enemy.
The Elector retired to Ratisbon. The Swedes took possession of
many towns in Bavaria, including Munich. But the country population
was so hostile that a permanent occupation seemed a waste of
energy; Gustavus retired to Ingolstadt.
A disturbing element now arose in a curious suspicion of the
ulterior motives of Gustavus. Both Protestants and Catholics—
Germans alike—began to fear that the king might be tempted by his
successes to make himself autocrat of Germany. This feeling soon
begot a half-heartedness among the king’s supporters. Richelieu
feared that Gustavus, instead of Ferdinand, was reaching a point
which might make him dangerous to France. The Emperor,
meanwhile, went back to Wallenstein, who had been so successful
before his deposition from command. Wallenstein made hard terms,
but he was a power which could no longer be disregarded.
Ferdinand, to gain his aid, gave him uncontrolled authority over the
army he should raise and all its operations.
Wallenstein began recruiting. He soon had forty thousand men.
The Catholics grew braver when the reconciliation of Wallenstein
and the Emperor became known. This, added to the suspicions of
the allies, constrained Gustavus to cease his successful offensive for
a cautious holding of what he already had.
Wallenstein marched into Bohemia, the Saxons offering no
resistance, and took Prague. He then moved to Bavaria and joined
the Elector. Seeing that Wallenstein by this manœuvre had gained a
position from which he might endanger his communications with
northern Germany, Gustavus marched summarily on Nürnberg,
which was the “cross-roads” of that section of the country, to head
Wallenstein off from Saxony, and ordered his outlying detachments
to concentrate there. He had under his immediate command but
one-third of Wallenstein’s total, and could not assume the offensive.
But he would not abandon southern Germany until driven from it. He
entrenched a camp near the town. Despite superior numbers,
Wallenstein did not attack. He could not rise above the prejudice of
the day. He deemed hunger a more efficient ally than assault. He sat
down before Nürnberg (July, 1632). The small-war indulged in
generally ran in favor of the king, who patiently awaited
reënforcements, having provided two months’ provisions for his army
and the town. Oxenstiern meanwhile collected thirty-eight thousand
men and advanced to the aid of his chief. Gustavus marched out to
meet him. Wallenstein did not interfere. The king was prepared for
battle should he do so. It was a grave military error that Wallenstein
took no means to prevent this junction.
Soon after Gustavus had received his reënforcements, he
determined to bring Wallenstein to battle, for famine had begun to
make inroads in Nürnberg and in both camps. He accordingly
marched out and drew up in the enemy’s front, but Wallenstein could
not be induced to leave his entrenchments (Aug., 1632). Failing in
this, the king at last resorted to an assault on the Imperial
fortifications. But after a gallant struggle he was driven back with a
loss of two thousand men. He has been blamed for this assault. He
deserves rather the highest praise for his effort to show the world
that gallantry and enterprise are among the best characteristics of
war. After him, Frederick proved that good troops can more often
take entrenchments than fail. His grenadiers were accustomed to
assault works held by two to one of their own number,—and take
them, too, under the king’s stern eye.
After ten weeks of this futile struggle, and much loss on both
sides, Gustavus, fairly starved out by want of rations and of battle,
determined to regain his communications with northern Germany. He
left five thousand men in Nürnberg, and marching past Wallenstein’s
camp unchallenged, moved to Würzburg. He had but twenty-four
thousand men left. Wallenstein, who again neglected an admirable
chance of falling on Gustavus’ flank, soon after marched to Bamberg
with the relics of his army, reduced to about the same number (Sept.,
1632).
Learning that Wallenstein had left Nürnberg, Gustavus, in the
belief that his opponent would seek repose for a period, marched
back to the Danube to resume the thread of his own work. The
Nürnberg incident had interrupted, not discontinued his general plan.
Wallenstein, as he had anticipated, sat quietly in Bamberg. He had
shown singular disinclination to come to blows with the king, and
exhibited far less activity, though, in truth, Wallenstein was both a
distinguished and able soldier.
On other fields the Swedes and allies were generally successful,
but finally thirty thousand Imperialists concentrated in Saxony, and
Wallenstein joined them and took Leipsic. Gustavus (Sept., 1632)
feared for his Saxon alliance, without which he could scarcely
maintain himself. He again put off the prosecution of his general
scheme, to go where lay the most imminent danger. Oxenstiern
again advised a summary march on Vienna, but Gustavus wisely
rejected the advice. At that day Vienna had not its importance of
1805. The king left a suitable force in Bavaria (Oct., 1632), marched
northward and entrenched a camp at Naumberg. Wallenstein turned
to meet him. His evident duty was to concentrate and attack. But,
according to the idea of that day, he parcelled out his army in
detachments, sending Pappenheim to Halle while he marched to
Merseburg. The Imperial general had blundered into a cardinal
position in the midst of the allies. The Swedes, twenty-seven
thousand strong, were at Naumberg, the Saxons, with eighteen
thousand, at Torgau, and ten thousand allies were marching up the
left bank of the Elbe. Wallenstein’s manifest operation was to fall on
each of these forces singly—on Gustavus first, as the strongest. But
he appeared to lose both head and heart when facing Gustavus. He
grew weaker as Gustavus grew more bold. He made no use of his
advantage, even if he comprehended it.
The king had got possession of the crossing of the Saale, but
Wallenstein stood between him and the Saxons. Gustavus’ generals
advised a manœuvre to join these allies, but the king was instinct
with mettle, and determined upon action.
 The ensuing battle of Lützen has little which is remarkable,
beyond the fiery ardor which ended in the death of Gustavus
Adolphus. It was a battle in simple parallel order, but the better
discipline of the Swedish army and the greater mobility of its
organization showed as marked superiority over Wallenstein’s
masses as the Roman legion, for the same reason, had shown over
the Macedonian phalanx eighteen centuries before. The Swedes
won the victory, but they lost their king, and Germany its protector
and champion.
As is the case with all great captains, Gustavus Adolphus gave
the impulse to every action while on the theatre of operations of the
Thirty Years’ War. For many centuries war had been conducted
without that art and purpose which Alexander, Hannibal, and Cæsar
so markedly exhibited. But in the operations of the Swedish king we
again find the hand of the master. We recognize the same method
which has excited our admiration in the annals of the noted
campaigns of antiquity, and from now on we shall see generals who
intelligently carry forward what Gustavus Adolphus rescued from the
oblivion of the Middle Ages.
The operations of the king, from his appearance in Germany,
showed his exceptional genius for war. He had no military guide,
except his study of the deeds of the ancients, for modern war up to
his day had altogether lacked depth and directness. During the first
fourteen months, he secured his foothold in the northern coast
provinces, in a most clear-witted and orderly manner. Every
circumstance was against him. He had weak forces to oppose to the
Emperor’s might. The half-hearted, fear-ridden Protestants yielded
him little aid and comfort; yet he reached his goal, step by step,
seizing and holding strong places at key points, and accumulating
supplies where he could count on their safety. But once, during his
entire German campaign,—at Nürnberg,—was he out of rations, and
this without ravaging the country. He carefully secured his
communications with the base he had established and with Sweden,
and never manœuvred so as to lose them. He gradually
strengthened himself with allies and recruits. Unlike the armies of the
day, who behaved as if the populations of the countries they
traversed were of less consequence than the beasts of the field,
Gustavus dealt with them in a spirit of kindliness and Christian
charity which won them over to his side. He kept his troops under
strict discipline, and by supplying all their wants and paying them
regularly, could rightfully prohibit marauding and plunder. He
understood how to avoid battle with an enemy too strong to beat,
how to lead him astray on the strategic field, how to manœuvre
energetically against an enemy, his equal or inferior in strength; how
to make the tactical mobility of his troops and his own ardor on the
battle-field tell; how to improve victory; and how to heighten and
maintain the morale of his troops under victory and defeat alike.
When, by his cautious and intelligent plan, the king stood firmly
planted between the sea, the Oder, and the Elbe, with flanks and
rear well guarded, he at once altered his conduct. He crossed the
Elbe and boldly attacked the enemy, adding to his strength by
beating him; and, leaving the allies to protect his flanks and
communications, he advanced with spirit and energy. In thirty days
he had established himself firmly on the Main; in little over four
months more he had moved down the Main, and had possessed
himself of or neutralized the whole middle Rhine; and in twelve
weeks thence had crossed the Danube, beaten the enemy at the
Lech, and occupied almost all Bavaria. Thus in less than nine
months (Sept., 1631 to June, 1632) he had overrun a much larger
territory than he had previously gained in fourteen, and had added
vastly to his standing. He had been bold and decisive, and yet never
lacking in the method and caution which were his guide. He had
established himself as firmly in southern Germany as previously in
northern.
At the height of his reputation and success, he was now ready to
attack Austria from the west. But the policy of France changed, his
allies became suspicious, and Wallenstein moved toward his rear.
The scene changed. Gustavus had no longer the security of whole-
hearted allies to connect him with Sweden, and his policy at once
shifted to the cautious one he had first shown. The thing for him to
consider, if he was to be thrown on his own resources, was first and
foremost his communications. With forces inferior to Wallenstein’s,
he acted on the defensive. With the accessions which made his
army equal to Wallenstein’s, he again went over to an offensive at
that day startling in its audacity. This failing, and provision having
given out, he moved, not to Bavaria, but to the Main, to protect his
line of retreat, which naturally traversed Hesse Cassel. So soon as
Wallenstein retired to Bamberg, Gustavus, leaving a lieutenant to
observe him, felt at liberty to take up his old thread in Bavaria. He
had gauged his opponent aright. When again Wallenstein, by his
Saxon affiliations, threatened, and this time more seriously, the
king’s allies, and remotely the security of his advanced position,
Gustavus again resorted to decisive operations. His march to
Saxony and his attack on the enemy at Lützen were equally bold,
rapid, and skilful.
Herein is a peculiarly intelligent adaptation of work to existing
conditions. From the king’s landing to the passage of the Elbe, while
securing his base, a cautious, but by no means indecisive policy;
from crossing the Elbe to Nürnberg, while moving upon the enemy, a
singular quickness and boldness, but by no means lacking in
intelligent and methodical caution; from Nürnberg to Lützen an
alternation from caution to boldness as circumstances warranted.
After Cæsar’s day, Gustavus was the first who firmly and intelligently
carried through a campaign on one well considered, fully digested,
broad, and intelligent plan, and swerved therefrom only momentarily
and partially to meet exigencies which could not be foreseen. The
advice of his most trusted aides was often opposed to what he did;
but they could not see as far as he saw. Each variation had its
definite object, which attained, the general plan was at once
resumed. There was an entire freedom from blind subservience to
the rules of war as then laid down; an intelligent sequence and inter-
dependence of movement on a plan elastic enough to meet
unexpected obstacles; these produced a perfectly systematic whole,
in which the unity of plan was never disturbed; and with this broad
scheme went hand in hand a careful execution of detail upon which
depended the success of the whole. His occupation remained firm;
his victualling was sufficient to his needs; his movements
accomplished what he sought.
 In pursuance of his cautious plan he neglected no essential
fortress or city; he held the passages of important rivers by erecting
bridge-heads or occupying towns; he kept upon his line of operations
suitable detachments, or met descents upon it by a prompt
movement towards the enemy. He so managed the division of his
forces as not to endanger his strength nor to lose the ability to
concentrate. He used his allies for the work they could best perform.
He kept the main offensive in his own hands, generally so ordering
that his lieutenants should act on the defensive, unless they
outnumbered the enemy, and then made them push with vigor. He
uniformly did the right thing at the right moment.
The secret of Gustavus’ success lay in his breadth of plan, in his
constancy to the work cut out, and in his properly adapting boldness
or caution to the existing circumstances. As with Alexander,
Hannibal, and Cæsar, it was the man himself whose soul illumined
his work; and this man had those transcendent qualities which
produce incomparable results in war, whenever they coexist with
great events. Equal as monarch and soldier, he united in his one
person the art of both. His nation and army were devoted to him soul
and body. His motives were the highest and purest which have ever
inspired a great captain; his pursuit of them was steadfast and noble,
open-handed and above-board, prudent and intrepid. In weighing his
intelligence, sound judgment, strong will, elevated sentiment, energy
and vigilance, he is properly put in the highest rank. But though his
record cannot perhaps vie with the others in the brilliancy of his
tactics, in the splendor of his victories, in extent of conquest, in
immensity of ambition, in the surmounting of all but impossible
natural or artificial barriers, in resisting overwhelming disaster with
heroic constancy,—still, if we look at the man, upon the results of
what he did, at the purposeless and barbarous nature of war as
conducted up to his day; if we weigh the influence of his short
campaigns upon all modern war, and consider how his nobility of
character and his life-work has made toward civilization, we cannot
rate Gustavus Adolphus too high. His pointing out the importance of
key-points in holding a country; the value of feeding an army by
careful accumulation of supplies, instead of by ravaging every
territory it enters; the advantage of a carefully drawn plan extending
over the entire theatre of operations; and the propriety of waging war
in a more Christian and civilized spirit,—marks the first step towards
the modern system. Gustavus Adolphus must be called the father of
the modern art of war; and is acknowledged as the one of all others
who re-created systematic, intellectual war, and stripped war of its
worst horrors.
After his death, his lieutenants tried fruitlessly to carry on his
methods. They retained a part of what he gave them; in many things
they slid back into the old ruts; and war (except with masters like
Turenne, Prince Eugene, and Marlborough) resumed its character of
isolated raids, until Frederick once more elevated it and stamped
upon it a permanence which it cannot now lose.
Among his enemies, during the remainder of the Thirty Years’
War there was nothing but the extremity of barbarous methods, over
which it is well to draw a veil.
Gustavus Adolphus was tall, handsome, and strong. In his later
years he grew so heavy that none but well-bred horses of great bone
and endurance could carry him. But he rode fast and far. His bearing
was noble, full of simple complaisance, and genuine. His quick mind
robbed work of effort; his ideas were clear, and he expressed them
crisply and in happy words; his voice was rich, his manner
convincing. A remarkable memory served to retain the names and
merits of his subordinate officers and numberless worthy men. He
maintained stern discipline in a just and kindly spirit. His religious
fervor was as honest as his courage was high-pitched. The Bible
was his constant companion and guide. He began all his acts with
unaffected prayer, and ended with thanksgiving. The Christian
virtues never resided in a more princely soul. He was sober, of
simple habit, and upright life. Towering over all around him in mind
and heart, and inflexible withal, he was yet modest and ready to
weigh the opinions of others. A tireless worker, he demanded equal
exertion from his officials and aides. But in his intercourse with all
men were kingly condescension and dignity joined. He was more
than monarch,—he was a man.