基于迭代特征选择的快速多角度人脸识别研究.pdfVIP

11-1777／TP 计算机研究与发展 ISSN1000—1239／CN of Researchand Journal 42(增刊)：325～329，2005 Computer Development 基于迭代特征选择的快速多角度人脸识别 范志刚 吕宝粮 (上海交通大学计算机科学与工程系 上海200030) (zgfan@sjtu．edu．Cn) Fast ofMulti-ViewFacesBasedonIterativeFeatureSelection Recognition Fan andLU Zhigang Baoliang Scienceand (DepartmentofComputer JiaotongUniversity，Shanghai200030) Engineering，Shanghai Abstract vector iterativefeatureselectionmethodis formulti— Byutilizingsupportmachines，an proposed viewface tothe betweenthetwo statistical tasks， recognition．vectormachines．Accordingrelationship featureselectionandmulti—class twotasksare intoa consistent classification，the integratedsingle frameworkandthe offeatureselectionisrealized classificationcanbemade goal effctively．The process morefastwithout the thisiterativefeatureselectionmethod． degradinggeneralizationperformancethrough The conductedontheUMISTmulti—viewfacedatabaseshowthatthisiterativefeature experiments selectionmethodcan themulti—viewface without the speedup recognitionprocess degradinggeneralization performance． wordsmulti—viewface classification；featureselection Key recognition；fast 摘要针对快速多角度人脸识别问题，提出了一种利用支持向量机实现迭代特征选择的方法．根据特 征选择和多类分类任务之间的统计关系，将这两个任务集成到一个统一的框架中，从而更有效地完成特 征选择．这种迭代特征选择方法的特点是能在不降低泛化能力的情况下大大加快分类速度．在UMIST 多角度人脸数据库上的实验结果验证了所提方法的有效性． 关键词 多角度人脸识别；快速分类；特征选择 中图法分类号TP391