一种基于离散小波变换和支持向量机的人脸识别新方法  

Novel face recognition method based on DWT and SVM

作　　者：王孝国[1] 张小宁[2] 杨吉斌[1] 张雄伟[1] 

机构地区：[1]解放军理工大学通信工程学院,江苏南京210007 [2]解放军理工大学科研部,江苏南京210007

出　　处：《解放军理工大学学报（自然科学版）》2006年第6期515-519,共5页Journal of PLA University of Science and Technology(Natural Science Edition)

基　　金：江苏省"图像处理与图像通信"高校重点实验室资助项目(KJS03036)

摘　　要：为提高人脸识别系统的性能,提出了一种基于离散小波变换DW T(d iscrete w avelet transform)特征提取和支持向量机(SVM)分类的人脸识别方法。首先,采用DW T对人脸图像进行降维和去噪,然后,对小波低频子图像进行核辨别分析(KDA)提取人脸特征,最后,结合SVM进行分类识别。基于该方法,对ORL人脸库进行分类识别,采用39个特征识别率达到98.2%。仿真结果表明,该方法明显减少了高频干扰对人脸特征的影响,增强了特征的辨别能力。而且,SVM有效地提高了分类器的分类和推广能力。To improve the performance of face recognition system, a novel face recognition method based on discrete wavelet transform （DWT） and support vector machine was presented. The raw face images were denoised by the DWT at first. Then the kernel discriminant analysis （KDA） was performed on the waveletfaces to enhance discriminant power. Finally, the support vector machine （SVM） was selected to perform face classification. Experimental results on ORL face database show that the proposed method achieves a recognition accuracy of 98.2% using only 39 features.

关 键 词：人脸识别 小波分析 核辨别分析 支持向量机 

分 类 号：TN911.73[电子电信—通信与信息系统]