Corner detection and shape classification of on-line handprinted kanji strokes

摘要

All on-line handprinted Kanji characters consist of strokes which are the loci from pen-down to pen-up positions. Such strokes can be classified into subpatterns to facilitate machine recognition of the characters. This paper describes a stroke analysis procedure which identifies corners within a given stroke and then classifies the stroke into a certain shape category according to the locus direction from a current corner to the next one and the global convex/concave property of those corners. Thus, it can provide high level information about the constituent strokes of a Kanji character.