Discrimination between printed and handwritten characters for cheque OCR system

Abstract

The identification of printed and handwritten characters is a fundamental and important issue for the cheque OCR system to achieve high-accuracy. In this paper, a novel method is presented to identify the written type based on only 4 or 5 characters in a severely corrupted bank cheque image. We first extract 4 kinds of features, totaling 17 features. Then the most suitable features are selected using the method based on separability measure. Finally, the selected features are used by a naive Bayesian classifier to realize the discrimination. Using 12,158 real checks to test our method, the accuracy is 99.2%.