A FUZZY SUBSTROKE EXTRACTOR FOR HANDWRITTEN CHINESE CHARACTERS

This paper incorporates fuzzy technique into handwritten Chinese chara cter substroke extraction. By computing a fuzzy score for every possib le substroke, feature loss is reduced without imposing unnecessary res trictions on the writing style. Our extractor is as good as existing c risp substroke extractors in unambiguous situations, and it outperform s them in ambiguous cases in terms of the number of desirable substrok es extracted. Comparing with the ''explore-every-possibility'' approac h based on crisp decision, we arrive at the correct feature set in a s maller number of iterations. The extractor's performance could also be tuned via a number of system parameters. Copyright (C) 1996 Pattern R ecognition Society.