In this papeer, a classification task on dissimilarity representations is considered. A traditional way to discriminate beiscr objecr recriminat bydissimilaritie isthe neNG# neNG#N meNG# Itsu#e)N howe)N) from anumbe of limitations,i.ei high computational completion a poteNF1x loss of accuracywhe a smallse ofprototype isuse and se1(LO#V1x tonoise TooveFBV( theF shortcomings,we propose touse a normal deal1F)LFN1x classifie constructe onthe same re(N1xFB()B1e We show that such aclassifieB base on aweNVG#1 combination ofdissimilaritiee can significantly improve the neve nevea rule withreh1N) to the reL#(V1xNL accuracy and computational effort.

Recent developments in defining and training statistical classifiers make it possible to build reliable classifiers in very small sample size problems. Using these techniques advanced problems may be tackled, such as pixel based image recognition and dissimilarity based object classification. It will be explained and illustrated how recognition systems based on support vector machines and subspace classifiers circumvent the curse of dimensionality, and even may find nonlinear decision boundaries for small training sets represented in Hilbert space.

Investigating a data set of the critical size makes a classification task difficult. Studying dissimilarity data refers to such a problem, since the number of samples equals their dimensionality. In such a~case, a simple classifier is expected to generalize better than the complex one. Earlier experiments confirm that in fact linear decision rules perform reasonably well on dissimilarity representations. For the Pseudo-Fisher linear discriminant the situation considered is the most inconvenient since the generalization error approaches its maximum when the size of a learning set equals the dimensionality. However, some improvement is still possible. Combined classifiers may handle this problem better when a~more powerful decision rule is found. In this paper, the usefulness of bagging and boosting of the Fisher linear discriminant for dissimilarity data is discussed and a new method based on random subspaces is proposed. This technique yields only a~single linear pattern recognizer in the end and still significantly improves the accuracy.

For learning purposes, representations of real world objects can be built by using the concept of dissimilarity (distance). In such a case, an object is characterized in a relative way, i.e. by its dissimilarities to a set of the selected prototypes. Such dissimilarity representations are found to be more practical for some pattern recognition problems. When experts cannot decide for a single dissimilarity measure, a number of them may be studied in parallel. We investigate two possibilities of combining either dissimilarity representations themselves or classifiers built on each of them separately. Our experiments conducted on a handwritten digit set demonstrate that when the dissimilarity representations are of different nature, a much better performance can be obtained by their combination than on individual representations.

In a dissimilarity (distance) data each pair of objects is characterized by a value which expresses the magnitude of di#erence between them. This type of data can be now classified using various approaches, provided that a new object is represented by its distances to the training samples.

For learning purposes, representations of real world objects can be built by using the concept of dissimilarity. In such a case, an object is characterized in a relative way, i.e. by its dissimilarities to a set of the selected prototypes. Such dissimilarity representations are found to be more practical for some pattern recognition problems.