基于流形的局部加权协从表示人脸识别  

Face recognition based on locality weighted CRC of manifold

作　　者：潘禹岐 江铭炎[1] 张振月 

机构地区：[1]山东大学信息科学与工程学院,济南250100

出　　处：《计算机应用研究》2016年第7期2206-2209,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61201370);山东省自然科学基金资助项目(ZR2014FM039);山东省自主创新成果转化专项基金资助项目(2012CX30202)

摘　　要：针对稀疏表示分类(sparse representation classification,SRC)运行时间过长和协从表示分类(collaborative representation classification,CRC)仅利用人脸数据的全局特征的不足,提出了一种基于流形的局部加权协从表示方法(locality weighted CRC,LWCRC),并将其应用在人脸识别中。首先将位于高维流形空间中的人脸特征进行局部加权投影到低维空间,然后通过Tikhonov正则化阵用训练样本加权表示测试样本,最后通过最小二乘(residual least square,RLS)分类器进行分类。通过在AR、FERET和Extended-Yale-B人脸数据库上的实验结果表明,提出的方法在性能上优于SRC、CRC、NRS方法,且对光照具有一定的鲁棒性。This paper proposed a face recognition method based on locality weighted CRC of manifold, to make up for highly time-cost of SRC and ignorance of local features in CRC. Firstly, it extracted face features of training sample hidden in the highly-dimension manifold space and projected them onto low dimension space with a local weighted way. Then, it represented test sample by training sample that regularized by Tikhonov matrix. At last, it classified them with a RLS classifier. The resuits from AR, FERET and Extended-Yale-B face databases show that LWCRC outperform SRC,CRC and NRS, and robust to illumination variance.

关 键 词：局部加权 协从表示 流形投影 人脸识别 计算机视觉 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]