Learning optimal classifier through fuzzy recognition rate maximization

The adjustment of the metric (features weighting) and the optimisation of the position of prototypes in the feature space are two of most important problems in minimum distance classifiers. This paper presents a new method to deal with these two problems based on the maximisation of a fuzzy recognition rate functional. The functional is a result of an easy to understand mathematical formulation. Experimental results on the recognition of binary cursive characters and gray-level container code characters follow. A comparison with the standard LVQ method is also made and discussed for the cursive character case.