A Use of Continuity in Character Recognition

Abstract

Within any given recognition class of handprinted characters, many pairs of characters are related to each other by continuous distortion. It has been usual to exploit this fact by using sequential recognition techniques that would not work if the predominant distortions were discontinuous. This paper introduces a new and highly parallel method of exploiting continuity. This method is shown experimentally to yield lower error rates than a comparable method of recognition that does not take account of continuity.