汉字的线性分类试验-中文信息学报.PDF

中　文 　信 　息 　学 　报 第 14 卷 第 2 期 JOURNAL OF CHINESE INFORMATION PROCESSING Vol . 14 No. 2 汉字的线性分类实验 金奕江 　马少平 智能技术与系统国家重点实验室 　清华大学计算机系 　100084 摘要 　本文通过实验研究了在汉字识别中应用线性分类器的可能性 ,考察了汉字之间的 线性可分性 。实验使用了两种主要的线性分类器 : Fisher 线性判别和感知器 。实验检验每一 对汉字的线性可分性 。实验结果表明 ,汉字之间的线性分类性是相当好的。尤其是 Fisher 线 性判别 ,不能成功线性分类的汉字仅 占百万分之 4 . 25 。这显示了在汉字识别中应用线性分类 器是有着巨大的潜力的。同时 ,线性分类实验结果还可用来检验所选取特征的好坏 ,有利于客 观的评价特征 。 关键词 　汉字识别 　线性分类器 　Fisher 分类器 　感知器 Experiment on L inear Classif ication of Chinese Character J in Yij iang 　Ma Shaoping The St ate Key L aboratory of Intelligence Technology and System Dep art ment of Comp uter Science and Technology Tsinghua U niver sity 　Beij ing 　100084 Email :j yj @s1000e . cs. t singhua . edu . cn Abstract 　This p aper st udies t he linear sep arability of Chinese character s t hrough experiment and discusses t he p robabilit y of applying linear classifier in Chinese character recognition . Two linear classifier s are used in t he experiment . They are Fisher Linear Discriminant ( FLD) and Percep t ron . The experiment examines t he linear sep arabilit y of a p air of Chinese character s. The result is p rett y satisfactory . Especially ,t he result of Fisher Linear Discriminant is very good . There are only 4 . 25 milliont hs p air s t hat cannot be linear sep arated . This show s t hat t he application of linear classifier in Chinese character recognition is potential . On t he ot her hand ,t his met hod can be ap plied to test t he feat ures used in recognition . Key words 　Chinese character recognition 　Linear Classification 　Fisher Linear Discriminant 　 P