基于核最小距离鉴别投影的人脸识别算法  

Face Recognition Algorithm Based on Kernel Minimum-Distance Discriminate Projection

作　　者：刘君[1] 黄燕琪[1] 汪传忠[1] 

机构地区：[1]南昌航空大学信息工程学院,江西南昌330063

出　　处：《测试技术学报》2015年第5期369-373,共5页Journal of Test and Measurement Technology

基　　金：国家自然科学基金资助项目NSFC(61163047);江西省教育厅基金资助项目(GJJ14503)

摘　　要：针对人脸图像的非线性分布特征,本文在最小距离鉴别投影(MDP)算法的基础上,提出了一种新的核最小距离鉴别投影算法.该方法首先通过核方法将样本映射到高维非线性空间,继而在得到的高维非线性空间上进行最小距离鉴别投影并计算其相应的投影矩阵,使得样本与类内、类间中心距离达到最优.最后采用最近邻分类方法对样本进行分类识别.通过在ORL和YALE人脸库上进行的实验,验证了该算法取得的识别效果优于其他方法.For the nonlinear distribution characteristic of face images,the paper propose a new nuclear minimum distance differential projection algorithm,which is based on the minimum distance differential projection（MDP）algorithm.Firstly,this method made samples mapped to high-dimensional nonlinear space by the method of nuclear。Then,the minimum distance differential projection was done and the corresponding projection matrix was calculated on the high-dimensional space,which made the points to within-class center distances and between-class center distances optimal。Finally,this paper adopt the nearest neighbor classification method to classify samples.Experiments on ORL and YALE face database shows that this algorithm has better recognition performance than other methods.

关 键 词：最小距离 最近邻 鉴别投影 人脸识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]