Shape-Oriented Chinese Character Clustering And Classification Using An SOFM Neural Network

• Main Authors: Chen, Chih-Wei, 陳志偉
• Other Authors: Tai-Wen Yue
• Format: Others
• Language: zh-TW
• Published: 1996
• Online Access: http://ndltd.ncl.edu.tw/handle/49377359593911859669

碩士 === 大同工學院 === 資訊工程學系 === 84 === In this thsis, we propose a clustering method based on Self- OrganizingFeature Map (SOFM) model for Chinese character classification. Since the total number of Chinese characters is extremely large, we can not directly train a neural network (NN) using all the Chinese characters in a way thatused for alphanumeric character recognition. Therefore, we propose a two- stage recognition scheme for Chinese character recognition. In the first stage, we incorporate the so-called GVF feature extraction mechanism to the SOFM so as to elicit the shape related feature for characters. BecauseGVF is insensitive to shape distortion, the characters with similar shapes will be categorized into the same class. By that, the number ofcharacters in each class will be dramatically reduced. Therefore, we canuse certain well-known NN models to recognize the characters of a particular class. This implies that the next stage of our OCCR system consists of several NN's, each of which is effectively in classificationa small set of characters. In this thesis, the training set of 5401 commonly used Chinese charactersare obtained from Eten. This training set is categorized into 100 classesvia a 10 X 10 SOMF. A number of experiments are conducted to verify theeffectiveness of the proposed work.