一种基于KPCA与LDA的人脸识别改进算法  

An improved face recognition algorithm based on KPCA and LDA

作　　者：郝静静[1] 李莉[2] 

机构地区：[1]开封大学信息工程学院,河南开封475004 [2]河南工程学院计算机科学与工程系,河南郑州450007

出　　处：《电子技术应用》2013年第12期132-134,137,共4页Application of Electronic Technique

基　　金：国家自然科学基金项目(51205372)

摘　　要：提出一种核主元分析和线性判别分析相结合的人脸特征识别改进算法。采用核主元分析法对人脸特征信息数据进行主分量提取,以消除数据特征间的相关性和压缩特征向量的维数。通过引入成对加权Fisher准则和正则化规则对线性判别分析法进行改进,进而实现人脸的自动识别。基于ORL人脸库进行的实验表明,此改进算法能够有效识别库中的人脸,识别率达91.7%,与K近邻法和主元分析法相比有较高的识别率。Put forward a kind of the face feature recognition method combining kernel principal component analysis and linear discriminant analysis. According to the characteristics of human face information, first use kernel principal component analysis method to principal component extraction of data, eliminate the correlation between data characteristics and compression feature vec- tor dimensions, and then improve the linear discriminant analysis method by introducing Weighted Pairwise Fisher Criterion（WPFC） and regularization rules, thus realize face automatic identification. The experiment based on ORL face database shows that this method can effectively identify faces in libraries and recognition rate reaches 91.7%. Application of this method has higher recogni- tion rate compared with neighboring method and PCA.

关 键 词：核主元分析 线性判别分析 人脸识别 特征提取 维数灾难 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]