关键词: 核函数, 条件正定核, 人脸识别, 支持向量机, 分类器

Abstract: In order to improve the performance of classifier for the face recognition，this paper introduces an improved kernel function named conditionally positive definite kernel.This kernel does not satisfy the well known Mercer condition，but it focuses on computing the distance between samples in the kernel space and finding the feature dissimilarity.Extensive face recognition experiments are based on three standard human face databases，namely ORL，YALE and ESSEX.The experimental results show that the proposed technique has higher recognition precision than other kernel function without more time on training.Furthermore，this method performs better robustness when the number of classes is adding.

Key words: kernel function, conditionally positive definite kernel, face recognition, Support Vector Machine（SVM）, classifier