On-line Chinese character recognition with tolerance ability of local-stroke variation and structure-size variation

• Main Authors: Jun-wei Liao, 廖君瑋
• Other Authors: Cheng-Huang Tung
• Format: Others
• Language: zh-TW
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/59306563797733788800

碩士 === 國立屏東商業技術學院 === 資訊科技系 === 96 === The thesis proposes an on-line Chinese character recognition with tolerance ability of local-stroke variation and structure-size variation. First, we make the assumption that each on-line handwritten Chinese character has an unique stroke sequence. The recognition rate for the characters with great structure-size variations is poor. Based on the dynamic programming recognition method whose recognition primitives are all line segments, the thesis first proposes a recursive normalization method, which uses the structures of cross-shape and T-shape to describe the features of references characters and recursively constructs the reference characters and, called recursively-normalized dynamic programming (RNDP) method. The proposed RNDP method can successfully resolve the recognition problems of the characters with great structure-size variations. Next, the thesis proposes a modified RNDP method, which can recognize the characters with alternative stroke sequences for each of its internal structure components. Because the modified RNDP method first recursively divides an input handwritten Chinese character into independent internal structure components, its ability of recognizing the characters with great structure-size variations is still outstanding. To recognizing the characters with alternative stroke sequences for each of its internal structure components, the modified RNDP method involves the technique of recognizing off-line handwritten Chinese characters. Thus, the internal structure components of an input handwritten Chinese character can be written in various stroke sequences and can be successfully recognized by the modified RNDP method. Third, the thesis proposes a hybrid system of combining the modified RNDP system and the handwritten Chinese effective phonetic sequence (EPS) recognition system. The advantage of the hybrid system is to tolerate the writing alternative between the Chinese characters themselves and the EPSs of Chinese characters. Since the hybrid system has the ability of determining that the input handwriting is a Chinese character or an EPS precisely, it can provide an user interface in which an user can input the Chinese text by writing the Chinese characters and the EPSs interleavedly.