融合局部奇异值特性的人脸自动识别

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问题,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,对待识别人脸的特征向量,计算其对各人脸样本的隶属度,最后做出判断。该方法与传统方法在ORL人脸库上进行的对比实验结果,表明了该方法的优越性。

Automatic Human Face Identification by Fusing Local Singular Value

Enough information for human face identification could not be obtained only by applying （Singular Value Decomposition, SVD） on human face, in order to solve the problem, a novel approach, which combined SVD and gray correlation analysis, was proposed. The key of this approach is that SVD could be applied on different parts of human face separately instead of the whole. Abundant information could be obtained and overcome the effect of small sample size. During the identification procedure, before making final decision, the fea- tures vector of input human face was set up, and calculating the relationship between samples respectively. Compared with traditional ap- proaches, the result on ORL face database shows that advantage of the approach.