基于环形对称Gabor变换和分块PCA的人脸识别  

Face Recognition Based on Circular Symmetrical Gabor Transformation and Block PCA

作　　者：王甜甜[1] 郭太良[1] 姚剑敏[1] 周龙沙 

机构地区：[1]福州大学物理与信息工程学院,福建福州350000 [2]TCL工业研究院,广东深圳518000

出　　处：《电视技术》2013年第15期173-176,193,共5页Video Engineering

基　　金：国家自然科学基金资助项目(61101169);福建省自然科学基金资助项目(2010J01333);福建省自然科学基金资助项目(2011J01347);福建省教育厅资助重点项目(JA09003);福建省教育厅资助重点项目(JA10033);福建省教育厅资助重点项目(JA11013)

摘　　要：提出了利用一种新方法获取合适的环形对称Gabor核函数窗口提取人脸图像的纹理信息,并结合改进的PCA进行人脸识别的方法。首先将人脸与环形Gabor小波函数卷积得到图像在5个尺度上的变换,该过程采用一种新方法获取合适的Gabor核函数窗口,以保证与人脸图像卷积后得到更为合理的人脸特征,同时利用一种新的分块PCA方法,将环形Gabor滤波后的子图像分组,对分组后的图像平均分块再进行降维,在空间位置上提取出Gabor人脸细节的主要特征,有效降低了人脸特征冗余。通过实验验证该方法在3.5 m内对人脸识别率达到95%,单个人脸识别时间小于0.22 s。The way of extracting facial image texture information is proposed based on using acquire the appropriate circular syrmnetricak Gabor kernel function window combined with an improved PCA algorithm. This way gets the change images during 5 scales through the face and the circular symmetry Gabor transformation. The process obtain the right Gabor kernel function window using a new method, and then convolution with image and cascade up to express face feature, and a new block PCA calculation method is used to divide the sub images of filtered circle Gabor into an average block, then reduce dimension of the block of images,extract the main features of the detail of the Gabor face in the space corresponding to reduce the face feature redundancy. Experimental results verify that the face recognition rate is reached to 95% in the range of 3.5 m. Especially, for the single person, the face recognition time is less than 0.22 s.

关 键 词：人脸识别 环形堆成Gabor变换 分块PCA 

分 类 号：TN919.8[电子电信—通信与信息系统]