HANDWRITTEN WORD RECOGNITION WITH CHARACTER AND INTER-CHARACTER NEURAL NETWORKS

An off-line handwritten word recognition system is described. Images o f handwritten words are matched to lexicons of candidate strings. A wo rd image is segmented into primitives. The best match between sequence s of unions of primitives and a lexicon string is found using dynamic programming. Neural networks assign match scores between characters an d segments. Two particularly unique features are that neural networks assign confidence that pairs of segments are compatible with character confidence assignments and that this confidence is integrated into th e dynamic programming. Experimental results are provided on data from the U.S. Postal Service.