基于差空间和最大散度差鉴别分析的人脸识别方法  

Face recognition based on residual space and maximum scatter difference

作　　者：李晓东[1,2] 费树岷[1] 张涛[1] 

机构地区：[1]东南大学复杂工程系统测量与控制教育部重点实验室,南京210096 [2]临沂师范学院信息学院,临沂276005

出　　处：《东南大学学报（自然科学版）》2009年第6期1130-1134,共5页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(60574006)

摘　　要：为了提高最大散度差鉴别分析方法在人脸识别中的识别率,提出了一种改进的基于差空间的最大散度差鉴别分析人脸识别算法.该方法把类内平均脸方法应用到2DPCA算法中,并基于改进的2DPCA方法分别建立训练样本和测试样本的差空间,然后用类内中间值代替类内均值修改了最大散度差鉴别算法中类内散布矩阵的定义.用改进后的最大散度差鉴别法对得到的差空间进行鉴别分析,分别提取训练样本和测试样本的鉴别特征,用最近邻分类器分类.在ORL人脸数据库上的实验结果表明,该方法可以有效地改善识别率.To improve the recognition rate of maximum scatter difference（MSD）,A modified method of discriminate feature extraction based on maximum scatter difference criterion in residual space is proposed.Firstly,within-class average face is combined with two dimension principal component analysis（2DPCA）.The improved 2DPCA is used to construct residual spaces of training samples and testing samples.At the same time,the definition of within-class matrix which is in the definition of MSD is modified by replacing within-class mean vector with within-class median vector.Then improved maximum scatter difference discriminate analysis is performed on the residual space to extract discriminate features of training samples and testing samples.Finally,nearest distance classifier is conducted for classifying.A lot of experiments results based on ORL（Olivetti research laboratory） face database show that the proposed algorithm can improve the recognition rate.

关 键 词：二维主成份分析 类内平均脸 差空间 最大散度差鉴别分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]