Radical extraction of off-line chinese characters by stroke clustering method

• Main Authors: Tseng Yao Lung, 曾耀隆
• Other Authors: Fan Kuo Chin
• Format: Others
• Language: en_US
• Published: 1994
• Online Access: http://ndltd.ncl.edu.tw/handle/25847138665265755816

碩士 === 國立中央大學 === 資訊及電子工程研究所 === 82 === In this thesis, a novel radical extraction scheme based on stroke clustering methodology is proposed to identify the type of Chinese characters together with the extraction of the corresponding radicals automatically. The K-means clustering algorithm and rule-based modification method are adopted in our proposed approach to achieve the aforementioned goals. Ten character types are defined for Chinese characters in this thesis. The proposed approach consists of three main modules which are stroke clustering module, rule-based modification module, and character type decision module. In stroke clustering module, K-means clustering algorithm is employed to cluster the strokes of a character into two clusters. Each cluster represents a radical. Since the clustering result may be erroneous. The rule-based modification module is thereby developed to rearrange the mis-clustered strokes. Finally, character type decision module calculating the evaluation distances via dividing paths for each character type is issued to determine which character type the input character is. 2500 most frequently used Chinese characters are tested in our system. Five kinds of fonts are considered in our experiment. The average accurate rate of radical extraction and type decision is 92.57%. The experimental results verify the validity and the feasibility of our proposed approach.