最大散度差无监督鉴别特征抽取与人脸识别  

Unsupervised discriminant feature extraction with maximum scatter difference and face recognition

作　　者：曹丽[1] 陈才扣[1,2] 

机构地区：[1]扬州大学信息工程学院,江苏扬州225009 [2]南京理工大学计算机科学与工程系,南京210094

出　　处：《计算机应用》2008年第S2期182-184,共3页journal of Computer Applications

基　　金：江苏省教育厅自然科学研究计划项目(07KJB520133)

摘　　要：最大散度差准则是对Fisher准则的改进,消除了小样本问题,但是该方法是基于整体特征的人脸识别方法,没有考虑到样本的局部特性。无监督的鉴别投影(UDP)技术,用于对高维数据进行维数缩减,它同时考虑到样本的局部特征和非局部特征,但是在人脸等高维图像识别的应用中,不可避免地会出现小样本问题。提出一种基于散度差的无监督鉴别特征抽取,避免了局部协方差奇异所产生的问题。在ORL人脸库和AR人脸库上的实验结果验证了该算法的有效性。Maximum scatter different criteria is the improvement of the Fisher criteria,it avoids the small sample size problem.This method is based on the overall characteristics,but it does not take into account local characteristics of the sample.Unsupervised Discriminant Projection(UDP) was developed for dimensionality reduction of high-dimensional data.It takes account both the local characteristics and nonlocal characteristics.The application of UDP to high-dimensional image recognition tasks such as face recognition inevitably suffers from the small sample size problem.Maximum scatter difference unsupervised discriminant feature extraction that adopts the difference of nonlocal scatter and local scatter were proposed.The proposed method was experimented on ORL face database and AR face database.The results show the effectiveness of the proposed method.

关 键 词：局部特征 最大散度差 特征提取 无监督鉴别投影 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]