A knowledge-based segmentation algorithm for enhanced recognition of handwritten courtesy amounts

Abstract

A knowledge-based segmentation algorithm to enhance recognition of courtesy amounts on bank checks is proposed in this paper. This algorithm uses multiple contextual cues to enhance segmentation and recognition. The system described extracts context from the handwritten numerals and uses a syntax parser based on a deterministic finite automaton to provide adequate feedback to enhance recognition. Further feedback is provided by a simple legal amount decoder that determines word count and recognizes several key words (e.g. thousand and hundred). This provides an additional semantic constraint on the dollar section. The segmentation analysis module presented is capable of handling a number of commonly used styles for courtesy amount representation. Both handwritten and machine written courtesy and legal amounts were utilized to test the efficacy of the preprocessor for the check recognition system described in this paper. The substitution error was reduced by 30-40% depending on the input check mix. © 1999 Pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.