Pattern classification using a generalised Hamming distance metric

In this paper we present a new class of minimum distance binary pattern classifiers based on a generalized Hamming distance metric applied to binary patterns. While classical minimum distance classifiers and especially the ones using Hamming-distance consider pattern features as having the same significance for the classification task, the proposed new distance metric based classifiers assign weights to the features according to their distinguishing abilities. Concerning neural network implementation of such weighted Hamming distance based classifiers, it is demonstrated that calculation of their weights is very simple. Finally we evaluate their distinguishing properties and we find that their performance is much better than the one of traditional Hamming distance classifiers.