Special Issue on Similarity-Based Pattern Recognition

[Andrea Torsello]

                         CALL FOR PAPERS
                        PATTERN RECOGNITION

                          Special Issue on
                Similarity-Based Pattern Recognition

               Submission deadline: December 15, 2004

Guest Editors:
Manuele Bicego, University of Sassari, Italy (bicego at uniss.it) 
Vittorio Murino, University of Verona, Italy (vittorio.murino at univr.it) 
Marcello Pelillo, University of Venice, Italy (pelillo at dsi.unive.it) 
Andrea Torsello, University of Venice, Italy (atorsell at dsi.unive.it) 


Traditional Pattern Recognition techniques are centered around the notion of 
"feature". According to this view, the objects to be clustered or classified 
are represented in terms of properties that are intrinsic to the object 
itself. Hence, a typical pattern recognition system makes its "decisions" by 
simply looking at one or more feature vectors fed as input. The strength of 
this approach is that it can leverage a wide range of mathematical tools 
ranging from statistics, to geometry, to optimization techniques. However, in 
many real-world applications the objects are not naturally representable in 
terms of a vector of features. For example, graph representations lack a 
canonical order or correspondence between nodes. Furthermore, even if a 
vector mapping can be established, the vectors will be of variable length, 
hence would not belong to a single vector-space. On the other hand, quite 
often it is possible to obtain a measure of the similarity/dissimilarity of 
the objects to be classified. It is therefore tempting to design a pattern 
recognizer which, unlike traditional systems, accepts as input a matrix 
containing the similarities between objects and produces class labels as 
output. 

Indeed, recently, there has been a renewed interest in similarity-based 
techniques, both in developing and studying new effective distances between 
non vectorial entities, like graphs, sequences, structures, and in proposing 
alternative distance-based paradigms. These methods typically keep the 
algorithm generic and independent from the actual data representation, 
allowing the use of non-metric similarities (thereby violating the triangular 
inequality). Further, they make the approaches applicable to problems that do 
not have a natural embedding to a uniform feature space, such as the 
clustering of structural or graph-based representations or the analysis of 
sequences. Finally, these representations are well suited to both supervised 
and unsupervised classification. The literature of pairwise algorithms 
includes, among others, kernel methods, spectral clustering techniques/graph 
partitioning, quadratic optimization, self-organizing maps, etc. These 
techniques are successfully applied to very diverse problems like object 
classification, image retrieval by content, color quantization, image 
segmentation, perceptual grouping, and bioinformatics (gene alignment, gene 
classification or phylogenetic analysis). 

The goal of this special issue is to solicit and publish high-quality papers 
that bring a clear picture of the state of the art in this area. We aim to 
appeal to researchers in Pattern Recognition and Computer Vision who are 
using or developing similarity-based techniques. Papers are solicited that 
address theoretical as well as practical issues related to the Special 
Issue's theme. Topics of interest include (but are not limited to): 
	- Similarity-based classification 
	- Similarity-based clustering 
	- Embeddings 
	- Kernel methods 
	- Spectral techniques 
	- Definition and analysis of distances between sequences, 
	  structures, and images 
	- Applications 


SUBMISSION PROCEDURE

Only electronic submissions will be accepted. Manuscripts must be submitted 
via the special issue's web page http://www.dsi.unive.it/Similarity-PR/

The manuscripts must be submitted by December 15, 2004, and should conform to 
the standard guidelines of the Pattern Recognition journal. All submitted 
papers will be reviewed by at least three independent reviewers.

-- 
Andrea Torsello PhD <torsello at dsi.unive.it>
Dipartimento di Informatica, Universita' Ca' Foscari di Venezia
via Torino 155, 30172 Venezia Mestre, Italy	
Tel: +39 0412348468	Fax: +39 0412348419
http://www.dsi.unive.it/~atorsell