The Dissimilarity Representation for Pattern Recognition:
Foundations and Applications
by E. Pekalska & R.P.W. Duin
World Scientific, 2007
Reviewed by: Arjan Kuijper
Click above to go to the publisher’s web page where there is a description of the book and links to the Table of Contents and sample pages.
Book Reviews Published in
the IAPR Newsletter
Practical Algorithms for Image Analysis, 2 ed.
by O’Gorman, Sammon and Seul
Handbook of Biometrics
by Jain, Flynn, and Ross (Editors)
Advances in Biometrics –
Sensors, Algorithms, and Systems
by Ratha and Govindaraju, (Editors)
Dynamic Vision for Perception and Control of Motion
by Polanski and Kimmel
Introduction to clustering large and high-dimensional data
The Text Mining Handbook
by Feldman and Sanger
Information Theory, Inference,
and Learning Algorithms
“Foundations and Trends in Computer Graphics and Vision”
Curless, Van Gool, and Szeliski., Editors
Applied Combinatorics on Words
by M. Lothaire
Human Identification Based on Gait
by Nixon, Tan and Chellappar
Mathematics of Digital Images
by Stuart Hogan
Advances in Image and Video Segmentation
Graph-Theoretic Techniques for Web Content Mining
by Schenker, Bunke, Last and Kandel
Handbook of Mathematical Models in Computer Vision
by Paragios, Chen, and Faugeras (Editors)
The Geometry of Information Retrieval
by van Rijsbergen
Biometric Inverse Problems
by Yanushkevich, Stoica, Shmerko and Popel
Correlation Pattern Recognition
by Kumar, Mahalanobis, and Juday
Pattern Recognition 3rd Edition
by Theodoridis and Koutroumbas
Dictionary of Computer Vision and
by R.B. Fisher, et. Al
Kernel Methods for Pattern Analysis
by Shawe-Taylor and Cristianini
Machine Vision Books
CVonline: an overview
The Guide to Biometrics by Bolle, et al
Pattern Recognition Books
Jul. ‘04 [pdf]
IAPR die-hards know that the IAPR has 20 Technical Committees. The first and second are called “TC1: Statistical Pattern Recognition Techniques” and “TC2: Structural & Syntactical Pattern Recognition”. These TCs have been organising joint biannual “S+SSPR” workshops in conjunction with the biannual ICPR conferences. Often these workshops are held in a relatively close neighbourhood of the ICPR location (Florida 2008, Hong Kong 2006), although the “dissimilarity representation” applies recently where the neighbourhoods were further apart (Cambridge – Lisbon 2004….). The acronym S+SSPR stands for Statistical + Syntactical and Structural Pattern Recognition, showing the relation to the TCs. The authors of this book, as well as the group from which they come at Delft University of Technology, the Netherlands, have been active in these IAPR sub-communities for quite some time.
The TCs deal with two different approaches in pattern recognition. The Statistical, quantitative, approach uses for instance well-developed mathematical theories of vector spaces and feature vector-based methods of learning. On the other hand, the Syntactical and Structural, qualitative, approach focuses on human cognition and perception inspired methods, and uses graphs, trees, etc. with appropriate grammars. One of the intensions for organising the combined workshops is to facilitate a possibility for researchers in one area to learn from those in the other one, vice versa. This book can be seen as an attempt to record some of the progress that has been made in getting the best of both worlds.
The book consists of two main parts, both covering roughly 250 pages - this is quite a comprehensive book, I must admit! In the preceding introductory chapter, the motivation is given for this work. The main observation is that there is no such thing as “general object similarity”. A comparison always takes place with respect to a reference frame. One can think of two brothers who may look very similar when viewed within a large group, but who appear to be rather dissimilar when they are seen without other persons. Similarly, throwing features in a vector (or a weighted graph) and applying sophisticated mathematical tools on it often give results that heavily depend on the underlying structure (e.g. function space) that is chosen. Consequently, the outcome may be mathematically optimal, but practically unsatisfactory. The authors apply the dissimilarity concept to both approaches. Although this book serves as only a starting point for further research, the results shown in this book are already promising.
Before getting to the practical results, the first part of the book gives relevant theory of dissimilarity representations and learning paradigms. It consists of four chapters dealing with i) spaces, ii) characterisation of dissimilarity matrices, iii) learning aspects, and iv) dissimilarity measures. The underlying principles and mathematical framework are described well. A very nice overview of existing (similarity-based) methods is given, together with their relation to the dissimilarity approach.
The second part contains applications in unsupervised learning (visualisation and data exploration) and supervised learning (domain descriptors, classification, and combining), followed by perspectives (representation review and conclusions / open problems). The authors show many examples and results. Some of them are already published – the book is actually an (extremely!) extended version of the PhD thesis of the first author, but this is never disturbing. Only every now and then one can detect a trace of the merging of the original journal papers. It is definitely much better than a lot of other PhD theses that end up as books; I enjoyed reading this. The less-mathematically interested reader will probably get a little bit scared by the eight pages “notation and basic terminology” and the most mathematical chapter on spaces, but even excluding these, readers will find much interesting and accessible material.
All in all, I think that this is a very useful book for researchers working in the IAPR TC1 &2 areas. It is useful to have an overview of methods and mathematics used in these fields – to peek in the kitchen of neighbouring colleagues and understand that together a very promising meal can be prepared.