An Analytic Word Recognition Algorithm Using a Posteriori Probability

Abstract

Word recognition algorithms are classified into two major groups. One is an "analytic" approach of recognizing individual characters, while the other is a "holistic" approach dealing with an entire word image. In the former approach, matching scores used to be calculated using heuristic functions, such as an average of confidence values on character recognition. In some non-heuristic studies, a stochastic evaluation function is employed, which is a ratio between an "a posteriori" probability and an "a priori" probability ("a posteriori" probability ratio). In this research, a new evaluation function is proposed, which is an improvement of "a posteriori" probability ratio. A result of an experiment using real images shows 9.1% improvement on handwritten word recognition.