基于MCD稳健估计的PCA人脸识别算法  

PCA face recognition algorithm based on robust MCD estimator

作　　者：张蓓[1] 王顺芳[1] 

机构地区：[1]云南大学信息学院,云南昆明650091

出　　处：《计算机工程与设计》2015年第3期778-782,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(11261068;11171293);昆明市中青年学术和技术带头人后备人选基金项目

摘　　要：针对人脸识别中,识别效果易受人脸修饰、部分遮挡、噪声干扰等不确定因素影响的问题,提出一种MCDPCA人脸识别算法以改进识别效果。基于主成分分析(PCA)进行特征脸提取,结合最小协方差行列式方法 (MCD)进行异常点检测和抗噪。针对人脸图像使用MCD算法,求出稳健的协方差矩阵估计,基于此协方差估计矩阵使用PCA技术提取重要的人脸特征用于识别。实验结果表明,在有遮挡和噪声干扰的情况下,相比传统PCA方法,该方法明显提高了人脸图像识别率。To solve the problems in facial recognition that recognition rates are often affected by face modification,partial occlusion,noise and other uncertain factors,a MCDPCA algorithm was proposed to improve the face recognition effect.The eigenfaces was extracted using the principal component analysis（PCA）method,the minimum covariance determinant（MCD）method was combined for detecting the outliers and removing noise.Specifically,MCD algorithm was used to obtain a robust covariance matrix estimate for face images,then PCA was used on the basis of this covariance matrix estimate to extract important facial features for recognition.Experimental results demonstrate that for face recognition with occlusion or noise,the proposed MCDPCA method performs much better than traditional PCA as far as the recognition rates are concerned.

关 键 词：主成分分析 最小协方差行列式 随机噪声 异常值 人脸识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]