基于主分量特征与独立分量特征的人脸识别实验

PCA是一种基于二阶统计的最小均方误差意义上的最优维数压缩技术，PCA方法所抽取特征的各分量之间是统计不相关的。ICA方法使用数据的二阶和高阶信息抽取数据的独立分量特征。在人脸图象识别的实际应用中，PCA与ICA方法各有胜负。PCA方法抽取出的主分量特征与ICA方法抽取出的独立分量特征是对原数据的两类不同描述，并设计出一个基于这两类特征的分类器组合方案；联合使用这两类特征，实验得出的人脸识别结果显示，基于分类器组合方案的识别结果优于单独使用PCA特征或ICA特征的单分类器方法。

Face recognition experiment based on principal components and independent components

PCA (principal component analysis) is the optimal dimension compression technique based on second-order information, in the sense of mean-square error. Features extracted by PCA are statistically uncorrelated to each other. ICA (independent component analysis) extracts features for data using their second-order and higher-order information. In the applications on face image recognition, it is hard to say that PCA is superior to ICA or ICA is superior to PCA. The two kinds of features extracted by PCA and ICA represent data are from different points of view. A hybrid classifier is proposed. The hybrid classifier based on PCA feature and ICA feature of face images have achieved good classification result, and the hybrid classifier outperforms the nearest neighbor classifier and the cosine classifier only using PCA feature or ICA feature.