基于复合核函数KPCA的红外人脸识别  

Infrared Face Recognition Based on Composite Kernel Function KPCA

作　　者：张朝福[1] 马力妮[1] 荆禄娜[1] 

机构地区：[1]北京信息科技大学计算机学院,北京100192

出　　处：《计算机仿真》2013年第2期369-372,共4页Computer Simulation

基　　金：北京市属市管高等学校人才强教计划资助项目(PHR201007131)

摘　　要：研究人脸优化识别问题,提出一种复合核函数KPCA的红外人脸特征提取法。利用最优或者接近最优的复合核函数主元分析KPCA方法对训练样本核映射到高维空间进行特征提取预处理,并结合最近邻法分类器分类进行红外人脸识别。该方法不仅有效的提取了训练样本的非线性信息,而且有效的改进了识别效果。多次实验结果表明了,基于复合核函数KPCA的红外人脸识别率优于传统的核主元分析法(KPCA)和主元分析法(PCA)。结果表明,改进方法可减少识别时间,并保证了识别率一直稳定在比较高的水平。In this paper, the extraction method of infrared face feature, which is a kind of KPCA based on composite kernel function, was proposed. This method makes use of compound kernel function method which is optimal or close to the best to map the training sample to a high dimensional space for feature extraction pretreatment, and combines with nearest neighbor classifier classification method to process infrared face recognition. It not only effec- tively extracts nonlinear information of the training sample, but also improves the result of the recognition. The experiments results show that, the infrared face recognition rate of KPCA based on composite kernel function is better than traditional KPCA and PCA. On the basis of reducing the recognition time as much as possible, the method ensures that the recognition rate steadily lies in the higher level.

关 键 词：核函数主元分析 红外人脸识别 主元分析 核函数 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]