Fast feature selection in an HMM-based multiple classifier system for handwriting recognition

Abstract

A novel, fast feature selection method for hidden Markov model (HMM) based classifiers is introduced in this paper. It is also shown how this method can be used to create ensembles of classifiers. The proposed methods are tested in the context of a handwritten text recognition task. © Springer-Verlag Berlin Heidelberg 2003.