Prototype-based minimum error classifier for handwritten digits recognition

Abstract

This paper describes an application of the Prototype-based Minimum Error Classification (PBMEC) to the offline recognition of handwritten digits. The PBMEC uses a set of prototypes to represent each digit along with an L v-norm of distances as the decoding scheme. Optimization of the system is based on the Minimum Classification Error (MCE) criterion. In this paper, we introduce a new clustering criterion adapted to the PBMEC structure that minimizes an Lv-norm-based distortion measure. The new clustering algorithm can generate a smaller number of prototypes than the standard k-means with no loss in accuracy. It is also shown that the PBMEC trained with the MCE can achieve over 42% improvement from the baseline k-means process and requires only 28Kb storage to match the performance of a 1.46MB-sized k-NN classifier.