On-line overlapped handwriting recognition based on substroke HMM

摘要

This paper describes substroke HMM-based on-line text handwriting recognition which can decode free-direction handwriting sequences without using any character segmentation technique in advance. Moreover, we propose the overlapped handwriting recognition assuming the application to small equipments with a narrow input area, such as a Personal Digital Assistant. Experimental evaluations were carried out by employing the character bigram probabilities as a language model, which consists of educational 1,016 Kanji and 71 Hiragana. We achieved correct rate of 87.99% on the recognition of 578 kinds of collected handwriting sequences and achieved 93.84% correct rate on the recognition of 1,714 characters in those sequences.