基于KPCA+CCA的人脸性别分类  

Gender Classification of Face Image Based on Kernel Principal Component Analysis plus Canonical Correlation Analysis

作　　者：张天刚[1] 张景安[1] 

机构地区：[1]山西大同大学数学与计算机科学学院,山西大同037009

出　　处：《软件》2011年第7期54-56,60,共4页Software

摘　　要：KPCA+CCA方法在人脸特征的提取、分类和分析中的有效性已受到有关研究人员的重视,用该法可从一张原始灰度人脸图像中直接识别出一个人的性别。将核方法引入到主分量分析中,由于CCA(Canonical Correlation Analysis)用到了KPCA(KernelPrincipal Component Analysis)变换后样本的全部核主分量,在分析中没有丢失任何鉴别信息,因而在不同光照、表情、姿态和脸部细节的原始灰度人脸图像中鲁棒性更高。在ORL人脸数据库中用基于核的最近邻特征分类器进行实验,取得了96%的平均准确率。The method of kernel principal component analysis plus canonical correlation analysis,its efficiency in extraction,classification and analyses of face features has caught attention for relevant researchers,it can identify directly the sex of a person from a original grayscale face image.Kernel method is used within principal component analysis,since CCA uses all the kernel principal component of the converted samples from KPCA,no discrimination is lost in the analysis,so the robust is higher in the original grayscale face image of variations of illumination,expression,pose and face details.The experiments were conducted with nearest feature classifier based on kernel method in the ORL face database,an average accuracy rate of 96% is achieved.

关 键 词：性别分类 核方法 主分量分析 最近邻特征分类器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]