Handwritten legal amounts segmentation for cheque reader based on simple Bayesian classifier

Two kinds of information are used by character segmentation: character structural information and contextual information. Both of them are important However, since current structural-based segmentation methods are more or less dependent on human´s analysis to acquire knowledge about the character´s structure and seldom acquire knowledge by statistical methods, they cannot fully use the character structural information. In this paper, a structural segmentation algorithm based on the simple Bayesian classifier is presented, which can use the structural information more effectively. This approach is tested on real check images and the correct segmentation rate is 90.3%, which has increased by 5.1% from the previous approach.