结合小波变换和稀疏表征的鲁棒人脸识别  

Robust Face Recognition Based on Wavelet Transform and Sparse Representation

作　　者：罗敏[1] 郑明辉[2] 

机构地区：[1]湖北民族学院理学院,湖北恩施445000 [2]湖北民族学院信息工程学院,湖北恩施445000

出　　处：《计算机工程与科学》2012年第12期130-133,共4页Computer Engineering & Science

基　　金：湖北省自然科学基金资助项目(2012FFC023);湖北省教育厅创新团队项目(T201214)

摘　　要：在人脸识别中,如何消除光照、表情、遮挡等不利因素的影响,提高识别的鲁棒性是当前急需解决的热点研究问题。本文提出了一种基于小波变换和稀疏表征的鲁棒人脸识别方法,首先对人脸图像进行小波变换,将变换得到的4个子带LL、LH、HL、HH作为基函数构成字典;然后将测试图像的LL子带在字典上稀疏分解;最后依据重构残差最小原则进行分类识别。在Yale人脸库上的实验结果表明该方法性能优于对比方法。It is currently a hot research topic that how to improve the robustness of the face recognition and reduce the influence of light,expression and occlusion. A robust face recognition method combining sparse representation and wavelet transform is proposed in this paper. Firstly, the original images are decomposed into four sub-band images（LL、LH、HL、HH）by applying wavelet transform,and the four different band images are chosen as basic functions to construct the dictionary.Secondly,the sub-band LL of testing images is decomposed sparsely. Finally, the classifier based minimal reconstruction error is used for identification. The experiment result on Yale face databases shows this method is better than the comparative method.

关 键 词：人脸识别 小波变换 稀疏表征 正交匹配追踪 压缩感知 鲁棒性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]