一种新的核线性鉴别分析算法及其在人脸识别上的应用

基于核策略的核Fisher鉴别分析（KFD）算法已成为非线性特征抽取的最有效方法之一。但是先前的基于核Fisher鉴别分析算法的特征抽取过程都是基于2值分类问题而言的。如何从重叠（离群）样本中抽取有效的分类特征没有得到有效的解决。本文在结合模糊集理论的基础上，利用模糊隶属度函数的概念，在特征提取过程中融入了样本的分布信息，提出了一种新的核Fisher鉴别分析方法——模糊核鉴别分析算法。在ORL人脸数据库上的实验结果验证了该算法的有效性。

A New Kernel Discriminant Analysis Algorithm and its Application to Face Recognition

Kernel Fisher discriminative analysis(KFD)algorithm based on kernel trick has been one of the effective nonlinear feature extraction methods.But all previous nonlinear feature extraction methods based on KFD algorithm which procedures are based on solving binary classification problem.How to extract effective discriminative information from overlapping(outlier)samples is still open.In this paper,a new KFD algorithm named fuzzy kernel discriminative analysis(FKFD)is proposed.In the proposed algorithm,fuzzy K-nearest neighbor(FKNN)algorithm is incorporated into the process of KFD algorithm and the corresponding fuzzy membership degrees are gained.Therefore,distribution information of samples is embedded in the proposed algorithm through fuzzy membership degrees.Experimental results on ORL face database demonstrate the effectiveness of the proposed algorithm.