摘要
针对仅在整幅人脸图像上进行奇异值分解无法得到人脸识别所需的足够信息的不足,提出了一种利用人脸图像的局部奇异值和灰色关联分析进行人脸识别的方法。该方法的关键是不在整幅人脸图像上进行,而是在人脸的不同区域进行奇异值分解以提取更丰富的信息和克服小样本效应。在识别阶段,对待识别人脸,计算其与各人脸样本的隶属度,最后作出判别。该方法与传统方法在ORL与AR人脸库上进行的对比实验结果表明,该方法不仅提高了识别率,且对人脸姿态变化与部分遮挡也具有一定的鲁棒性。
This paper presented a face recognition method using singular value decomposition (SVD) on human local facial area and gray correlation analysis to solve the problem that it could not provide enough information for face recognition by the method of using singular value decomposition on whole facial image. The key of this approach was that applied SVD to different parts of facial area instead of the whole facial region. So the rich information could be obtained and the problem of small sample size could be solved. In the recognition step, set up the features vector of input facial image, and then computed the membership degrees of these features to each facial sample respectively, and finally obtained the decision. Comparative experimental results on ORL and AR face database show that the performance is better than that of tradition SVD-based method and robustness to the change of impression and occlusion.
出处
《计算机应用研究》
CSCD
北大核心
2010年第5期1952-1954,1958,共4页
Application Research of Computers
基金
国家博士基金资助项目(2007BS053)
关键词
人脸识别
灰色理论
灰色关联分析
奇异值分解
face recognition
gray system
gray correlation analysis
singular value decomposition(SVD)