基于局部保留映射与径向基网络的人脸识别方法  

Face Recognition Based on Locality Preserving Projections and Radial Basis Function Network

作　　者：梅健强[1] 刘正光[1] 

机构地区：[1]天津大学电气与自动化工程学院,天津300072

出　　处：《天津大学学报》2008年第4期419-422,共4页Journal of Tianjin University(Science and Technology)

摘　　要：局部保留映射(locality preserving projections,LPP)选择人脸子空间特征包含非线性信息而不利于最近邻法分类.基于径向基函数(radial basis function,RBF)分类器可以将非线性可分问题转化为线性可分问题的特点,提出了利用LPP子空间和RBF网络相结合进行人脸识别的方法,LPP算法采用监督模式,RBF网络隐层中心采用正交最小二乘(orthogonal least-squares,OLS)法训练.实验结果表明,该方法在Yale-B和Yale-B Extended人脸数据库上的识别率为95.67%,在CMU-PIE人脸数据库上的识别率为98.52%,具有较好的抗噪能力,识别效果优于特征脸、Fisher脸以及拉普拉斯脸法.Face subspace features selected by locality preserving projections （ LPP ） contain non-linear information which leads to the failure of using the nearest neighbor classifier to recognition. Therefore, a face recognition method combining LPP subspace with radial basis function （ RBF ） classifier was proposed according to the advantage of RBF that could convert non-linear separable problem to a linear separability. In this paper, LPP was used in supervised mode and the hidden center of RBF network was trained with orthogonal least-squares （OLS） method.Extensive experimental results show that the recognition rates of the proposed method on Yale-B, Yale-B Extended face databases and CMU-PIE face database are 95.67% and 98.52%, respectively.The improved method has better anti-noise capability and its recognition rate is better than those of Eigenfaces, Fisherfaces and Laplacianfaces.

关 键 词：人脸识别 主成分分析 线性判别分析 局部保留映射 径向基函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]