Character segmentation heuristics for check amount verification

Automatic check amount verification involves three basic steps: 1) finding the courtesy amount (the numerical amount) on the check, 2) segmenting that amount into its individual characters, and 3)Using a recognition engine to recognize the individual characters and, thus, recognizing the check's amount. This thesis will focus on improving the segmentation process. In particular, it will focus on the classical approach of segmentation based on charcter features alone (i.e. not aided by recognition). A segmentation workbench is developed for the purpose of studying current and potential techniques for numerical digit segmentation. The problem of segmenting overlapping or touching characters is studied in detail. Also, a hybrid heuristic involving a combination and variations of the "drop-falling" algorithm is developed which has promising results. This algorithm is then incorporated into a complete optical character recognition system. Thesis Supervisor: Amar Gupta Title: Co-Director, Productivity from Information Technology (PROFIT) Initiative, Sloan School of Management