Learning Handwritten Digit Recognition by the Max-Min Posterior Pseudo-Probabilities Method

Learning is important for classifiers. This paper proposes a new approach to handwritten digit recognition based on the max-min posterior pseudo-probabilities framework for learning pattern classification. Each digit class is modeled as a posterior pseudo-probability function, the parameters in which are trained from positive and negative samples of this digit class using the max-min posterior pseudo-probabilities criterion. In the process of digit classification, an input pattern is classified as one of ten digit classes or refused as being unrecognized according to the posterior pseudo-probabilities. Experiments on NIST database show the effectiveness of the proposed approach in reducing the error rate and making rejection decisions to those input pattern which can not be reliably by even human.