一种新颖的基于NMF的人脸识别方法  

Novel NMF-based Face Recognition Method

作　　者：宿韬[1] 张强[1] 魏小鹏[1,2] 周昌军[2] 

机构地区：[1]大连大学辽宁省智能信息处理重点实验室,大连116622 [2]大连理工大学机械工程学院,大连116024

出　　处：《系统仿真学报》2009年第12期3614-3616,3621,共4页Journal of System Simulation

基　　金：教育部新世纪优秀人才支持计划(NCET-06-0298);辽宁省高校创新团队支持计划(2008T004);辽宁省高等学校优秀人才支持计划(RC-05-07,2006R06);辽宁省教育厅科学研究计划(05L020);大连市科学技术计划(005A10GX106)

摘　　要：提出了一种新颖的基于非负矩阵分解(NMF)进行人脸识别的方法,该方法首先对人脸图像作NMF得到由人脸基图像组成的子空间矩阵及其对应的系数矩阵,然后将其重构获得NMF重构人脸图像,进而从原始图像中减去重构图像以获取残差图像。最后对残差图像作局部非负矩阵分解(LNMF)以提取其非负子空间及其系数矩阵,并将两次得到的系数矩阵混合,根据最近邻原则进行识别。在ORL标准人脸数据库上的实验表明,该方法具有更好的识别率。A novel NMF-based method for face recognition was put forward. First, NMF was applied to the face images for extracting non-negative subspaee and the corresponding coeffeient matrices, then they were reconstructed to obtain the NMF reconstructed images. Moreover, the residual images were computed by subtracting reconstructed images from original face images. Finally, the residual images were applied by LNMF to extract non-negative subspace and the corresponding coefficient matrices, the two coefficent matrices were combined for face recognition based on the nearest neighbor principle. The simulation experiments illustrate that this method has better recognition rate on the ORL face database.

关 键 词：非负矩阵分解 人脸识别 重构图像 残差图像 LNMF 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]