| บทคัดย่อ(English) |
This thesis presents a new on-line recognition method of Thai handwrittencharacters. Nowadays, active researches in Thai handwriting recognition are converged intotwo distinct methods, statistical methods (such as Hidden Markov Model (HMM), ArtificialNeural Networks, etc.) and character-structure-or-rule based methods (such as Fuzzy Logicclassifier). The former, HMM shows poor recognition rate due to Thai fuzzy characters. Theshortcoming of the latter, Fuzzy Logic classifier is on difficulties in establishing setsof rules to cover whole handwriting styles. Our method is proposed to exploit the best oftwo worlds by combining the advantages of each other in order to compensate the HMM's poorrecognition rate of fuzzy characters and the difficulties of constructing the rules. The system was executed on a Pentium III processor at 733 MHz and 256 Mbytes ofRAM. The experimental results showed an average recognition rate 93.2%, in a middle levelcharacter case for independent users, and 91.0% in a upper level character case forindependent users. Furthermore, the average running time of our proposed method was only0.09 seconds/character. |
