林彰鏗撰

• Main Author: 林彰鏗
• Other Authors: 范子儀
• Format: Others
• Language: zh-TW
• Published: 1993
• Online Access: http://ndltd.ncl.edu.tw/handle/06403523514132219547

博士 === 國立中央大學 === 資訊及電子工程研究所 === 81 === 　　The greatest challenge facing the popularization of the Chinese computer in the age of information is the automation of Chinese information input. On-line Chinese character recognition technique is the fundamental solution to this problem. The proposal of an on-line Chinese character recognition technique able to effectively recognize cursive script, and a practical system is the ideal goal of research and development. The trend of applying the two proposals in question as the Chinese input tools is well in sight. For the demand in the future trend, the on-line square writing Chinese character recognition is selected as our research objectives. 　　For Chinese, due to its huge number of characters, complicated structure and the variations of usage characteristics depending on individual user, it will not be easy to successfully research and develop the cursive script recognition technique. To cope with the characteristics of this problem, we specified a research and development strategy. Based on this strategy, we proposed the idea of adopting the Chinese character heuristic knowledge to represent Chinese feature, and actually put into practice in an experimental system of on-line Chinese square writing character recognition. We hope to prove that this idea is an effective step stone to the development of cursive script recogition, and demonstrate that the importance of the accumulation of research and development experiences in this field. 　　The conventional feature extraction methods mainly rely on extracting the common features shares by characters in a large group of vocabulary. But this is not appropriate for cursive scripts, whose features features represent numerous variations. A better way is to use the concept of heuristic knowledge for extracting individual features from small groups of vocabulary or individual characters. In our experimental system, we proposed two kinds of knowledge and three types of matching methods for the recognition of square writing. The accuracy can reach 98% when the first ten most similar candidates are selected, and over 90% when only the first is selected. Used in the recognition of similar characters, the accuracy can reach 99.9%. The experimental results prove that the idea in this thesis is reasonable, feasible, and sufficient to act the recognition strategy of further research and development.