【作者】 王小兰； 刘强； 尹忠科； 王建英；

【Author】 Wang Xiaolan Liu Qiang Yin Zhongke Wang Jianying School of Info Sci.& Tech Southwest Jiaotong University,Chengdu,610031,China

【机构】 西南交通大学信息科学与技术学院；

【摘要】 相似字是造成汉字识别误识率和拒识率高的主要原因之一,而且由于书写习惯会使识别难度更大。本文针对手写相似字的特点,提出基于 Bandelet 的特征选取方法,将传统的结构特征与统计特征有机结合起来。运用 K-L 变换对已提取的特征向量进行降维,最后输入支持向量机进行真伪鉴别。实验结果表明该算法对测试样本具有高识别率。更多还原

【Abstract】 Similar Chinese characters is one of the main reasons to cause the high reject rate and substitution rate.However,it is more difficult to recognize because of person’s handwriting habit.This paper, focus on Similar Chinese characters and present a new feature selection method based on Bandele,which gives full play to the merits of both conventional structure feature and statistical feature.After dimensionality reduction to extracted eigenvector by K-L transform,genuine signatures and forgeries are distinguished through support vector machines (SVM).The result of our experiment has confirmed the effectiveness of the proposed approach.更多还原