On-line recognition of cursive Hangul characters by modeling extended graphemes

We propose an effective method for online cursive Hangul recognition. Extended graphemes are modeled separately to recognize cursive characters, and rule processing is combined with elastic matching to discriminate similar characters. The extended graphemes consist of basic graphemes and connected graphemes of two or three basic graphemes which are frequently found in cursive Hangul characters. The rule based processing catches the specific features of graphemes whereas the elastic matching catches the general features of graphemes so the integration of two methods could complement the deficiencies of each other. In terms of integrating rules with elastic matching we could reduce 40.35% of error rates of grapheme recognition. The experiments produce 94.1% recognition rate for 479,326 Hangul characters (2350 different characters).