On-line handwritten character recognition based on stroke-HMM in non visual-feedback writing condition

摘要

This paper describes a stroke-HMM based on-line handwritten character recognition in wearable computing environments where the writer has no visual feedback of written characters. First, we propose an adaptation technique for correctly recognizing the leaned characters caused by unusual writing positions. Experiments of on-line hand-written Kanji character recognition with a lexicon of 1016 elementary characters revealed that the proposed technique reduced the error rate by about 38% compared with the case of no adaptation Secondly, we propose a method of automatically training a dictionary with a database of on-line handwritten characters in order to tackle the problem of non-unique stroke orders. Experimental results showed that the error rate was reduced by about 28% when 300 new definitions of stroke orders were added to the original handmade dictionary.