基于核的二维最大散度差投影鉴别分析及人脸识别研究.pdfVIP

ofthe27由ChineseControl Proceedings Conference July16-18，2008，Kunming，Yunnan，China 基于核的二维最大散度差投影鉴别分析及人脸识别木 陈才扣，崔美琳，曹丽，刘永俊 扬州大学信息工程学院，扬州225009 E-mail：yzcck@126．corn 摘要：传统的核方法只能应用于一维数据，本文直接在二维图像上应用核方法，提出一种新的基于核的二维最大散 度差投影鉴别分析方法。该方法不仅抽取了图像中更加有效的非线性鉴别特征，使正确识别率显著提高，而且为二维 非线性鉴别分析提供了一个统一的构架。最后在AR标准人脸库中的实验结果验证了本文算法的有效性． 关键词：散度差，鉴别分析，核方法，人脸识别 - MaximumUmCatter-IDiIIerence KernelbasedTwo．Dimensional1 Scatter DiscriminantandFace ection Proj Analysis Chen Caikou，Cui Li，LiuYongjun Meiling，Cao Information 225009，P．R．China EngineeringCollege，YangzhouUniversity，Yangzhou E—mail：yzcck@126．COlIl Abstract：Traditionalkernelmethodsare toone-dimensionaldata．The a kernel—based onlyapplied paperdevelops thekemelmethodscan be into two．dimensionalmaximumscatterdifference method．where prqjectionanalysis directlyapplied into matriceswhichneednotbetransformedone—dimensionavectors．Itisabletoextractmore originaltwo．dimensionalimage effectivenonlinearfeatureand thecorrect rates．What’smore．italSOoffersaunifiedframeworkforker- ireprove recognition nel．basedtwo—dimensionaldiscriminant onARfacedatabase projection analysis．Finally，extensiveexperimentsperformed theeffectivenessofthe method． verify proposed KeyWords：ScatterDifference，DiscriminantMethods，Facerecognition Analysis，