On-line Thai handwritten character recognition using hidden Markov model and fuzzy logic

This paper presents a new on-line recognition of Thai handwri t ten characters. Active researches in Thai handwri t ten character recognition are converged into two distinct methods, H M M a n d Fuzzy logic classifier. T h e former showed poor recognition rate d u e t o Thai fuzzy characters. The shortcoming of t h e la t te r is on difficulties in establishing t h e se t of rules t o cover a whole handwri t ing styles. O u r method is proposed t o exploit t h e best of two wor lds(HMM a n d distinctive feature based Fuzzy classifier). T h e experimental resul ts was shown a n average recognition rate is improved from 89. l%(using H M M ) t o 91.2 using our proposed method. I N T R O D U C T I O N Wide spread acceptance of pen-based computers has attracted effort in finding reliable algorithms for on-line handwriting recognition [5]. One of the most reliable algorithms for many languages such as English, Arabic, Chinese and Japanese characters can be based on either Hidden Markov Model(HA4M) (eg. [4], [2]) or analytically distinctive features using fuzzy set theory [3]. Empirically, we were not successful in using HMM for Thai handwritten characters. This is due to the fact that classification results can be confuscd within pair or among a set of fuzzy characters [l]. In other words, HMM cannot capture some distinctive features (describe later) occurred between confusablc letters. For cxamplc: ‘@I’ and ‘01’ only the notch is diffcrence, In’, 0-7803-8178-5/03/$17.00