基于LDA变换验证码识别技术初探.pdf

18 73 2009 1 V ol. 18 Sum N o. 73 JOURNAL O F GUANGDONG RAD OI TV UN VI ERS TI Y N o. 1. 2009 LDA The R ecogn ition o V alidation Code B ased on M esh ing Feature and LDA 司徒毅 SITU Y i (广东轻工职业技术学院, 广东广州, 5 10300) (Guangdong Industry Technical Colledge, Guangzhou, Guangdong, China, 510300) 本文采用基于递归算法的去 离散点法消 孤立噪声, 选用扫描边界的方法分割字符, 来研究验证码自动 识别技术, 选择和提取稳定而又便于表示的特征向量是本系统的核心之 一本文提出了简单的字符特征提取方法: 采用 网格灰度特征并对该特征进行线性鉴别分析( LDA, L inear discr mi inant analysis) 变换, 结合最小距离分类器完成字符识别 过程, 通过提高训练样本数, 有效解决了形近字符识别率低的问题, 取得了很好的识别效果 验证码识别; 线性鉴别分析; 网格特征 T P3930 8 A 100 8- 9764 ( 2009) 01- 0 102- 07 A bstrac t: Th is thesis researches on techn ique o va lidation code recognition. It adopts a deno ising approach based on recu r sive a lgo rithm and em ploy s the segm entation m ethod through scann ing borde rline betw een characters. T o select and ex tract stab le and s mi p le eature is key to the va lida tion code recogn ition system. T hu s w e propose a smi ple charac ter eature extraction ap proach. T h is approach ca lcu la tes only one g rid ea ture and trans orm s the orig inal eature by LDA a lgor ithm. Then the ea ture set is put into the m in mi um - distance class i ier or recogn ition. In order to mi prove the low recognition ra te or smi ilar charac ters, the number o training set is increased. It is proved that better recogn ition per o rmance is ach ieved. K ey w ord s: va lida tion code recognition; linear discr mi inant ana lysis; g rid eature 1 12 11 , ,