On-Line Natural Writing Recognition Using Contextual;

• Main Authors: Chen Haw Jen, 陳浩楨
• Other Authors: Chen Shu Yuan;Tsay Yin Tay
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/87632399312341870583

碩士 === 元智大學 === 電機與資訊工程研究所 === 82 === An on-line natural writing recognition system is developed in this thesis. In the proposed system, the recognition unit is a word composed of lower case English letters. However, the writing style provided in this system is less constrained; the two styles of writing, print and script, are included, moreover, cursive writing and discrete writing can be mixed in the proposed system. A new set of primitive strokes is proposed to represent twenty-six letters with two styles of writing. Based on the set of primitive strokes, an input word is recognized by the following method. First, the word is broken into segments. Then each segment is classified as one of the predefined primitive strokes using attributed string matching. The results are stroke sequences, Inorder to make recognition work faster, contextual information and dictionary are used to filter out improper stroke sequences. Hence, some admissible letter sequences are generated. Finally, the cost of each letter sequence is computed, and the one with lowest cost is the recognition result. Experimental results with high recognition speed and convincing recognition rate prove the feasibility of the proposed system.