基于2DLDA与SVM的人脸识别算法  

Face recognition based on 2DLDA and SVM

作　　者：甘俊英[1] 何思斌[1] 

机构地区：[1]五邑大学信息学院,广东江门529020

出　　处：《计算机应用》2009年第7期1927-1929,共3页journal of Computer Applications

基　　金：广东省自然科学基金资助项目(07010869;032356);广东省江门市科技攻关项目(江财企[2006]59号)

摘　　要：二维线性鉴别分析(2DLDA)算法能有效解决线性鉴别分析(LDA)算法的"小样本"效应,支持向量机(SVM)具有结构风险最小化的特点,将两者结合起来用于人脸识别。首先,利用小波变换获取人脸图像的低频分量,忽略高频分量;然后,用2DLDA算法提取人脸图像低频分量的线性鉴别特征,用"一对多"的SVM多类分类算法完成人脸识别。基于ORL人脸数据库和Yale人脸数据库的实验结果验证了2DLDA+SVM算法应用于人脸识别的有效性。＂Small sample size＂ problem of LDA algorithm can be overcome by two-dimensional LDA （2DLDA）, and Support Vector Machine （SVM） has the characteristic of structural risk minimization. In this paper, two methods were combined and used for face recognition. Firstly, the original images were decomposed into high-frequency and low-frequency components by Wavelet Transform （WT）. The high-frequency components were ignored, while the low-frequency components can be obtained. Then, the liner discriminant features were extracted by 2DLDA, and ＂one vs rest＂, strategy of SVMs for muhiclass classification was chosen to perform face recognition. Experimental results based on ORL （Olivetti Research Laboratory） face database and Yale face database show the validity of 2DLDA ＋ SVM algorithm for face recognition.

关 键 词：小波变换 二维线性鉴别分析 支持向量机 人脸识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]