Novel framework for selecting the optimal feature vector from large feature spaces

Abstract

There are several feature extracting techniques which can produce a large feature space for a given image. It is clear that only small numbers of these features are appropriate to classify the objects. But selecting an appropriate feature vector from the large feature space is a hard optimization problem. In this paper we address this problem using the well known optimization technique called Simulated Annealing. Also we show that how this framework can be used to design the optimal 2D rectangular filter banks for Printed Persian and English numerals classification, Printed English letters classification, Eye, Lip and Face detection problems. © 2009 Springer Berlin Heidelberg.