随机采样子空间保局投影人脸识别算法  

Random sampling subspace locality preserving projection for face recognition

作　　者：杨利平[1] 龚卫国[1] 李伟红[1] 辜小花[1] 

机构地区：[1]重庆大学光电技术及系统教育部重点实验室,重庆400030

出　　处：《光学精密工程》2008年第8期1465-1470,共6页Optics and Precision Engineering

基　　金：国家863高技术研究发展计划资助项目(No.2007AA01Z423);国防"十一五"基础科研基金资助项目(No.C10020060355);教育部科学技术重点资助项目(No.02057);重庆市自然科学基金重点资助项目(No.CSTC2005BA2002;No.CSTC2007AC2018)

摘　　要：针对单一保局投影(LPP)算法分类识别能力弱的问题,提出了一种随机采样子空间保局投影算法(RSSLPP)。该算法在对训练样本的主元子空间进行随机采样的基础上,利用保局投影得到了多个既有差异且又互补的保局投影子空间,测试样本被分别投影到这些保局投影子空间上,然后利用最近邻分类器进行分类识别。最后,根据多数投票原则融合多个子空间上的分类结果来确定样本所属类别。在FERET人脸图像子库上的实验表明:随机采样子空间保局投影算法的性能明显优于Eigenface、Fisherface、保局投影和鉴别保局投影等算法;和保局投影算法相比,本文所提出的方法人脸识别精度提高了10%以上。结果表明,随机采样子空间保局投影算法有效地融合了各LPP投影空间的互补信息,可以显著地提高人脸识别精度。A Random Sampling Subspace Locality Preserving Projection （RSSLPP） method is proposed to improve the recognition performance of a single Locality Preserving Projection （LPP）. At the training stage, based on random sampling of the principle component subspace of training set, the multiple discrepant and complementary LPP subspaces are generated by LPP method. At the recognition stage, test sample is successively projected into each nearest neighbor classifier is used for classific random sampling ation and recognition. F e component subspace,then the nally, majority voting criterion is used to fuse the recognition result of each LPP subspace. The experimental results on FERET subset illustrate that the performance of RSSI.PP method is superior to those of Eigenface, Fisherface, LPP and discriminant I.PP （DLPP）. The recognition accuracy of RSSLPP is over 10% higher than that of LPP. The RSSLPP method can effectively combine the complementary information of each LPP subspace and can improve face recognition accuracy.

关 键 词：随机采样子空间保局投影 保局投影 子空间 人脸识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]