基于加权Fisher准则的线性鉴别分析及人脸识别  

Linear discriminant analysis based on weighted Fisher criteria and face recognition

作　　者：郭娟[1] 林冬[1] 戚文芽[1] 

机构地区：[1]信息工程大学信息工程学院,河南郑州450002

出　　处：《计算机应用》2006年第5期1037-1039,1049,共4页journal of Computer Applications

摘　　要：提出了一种基于加权Fisher准则线性鉴别分析的人脸识别方法。该方法引入了一种新的权函数对Fisher准则加权,以提高样本在低维线性空间中的可分性,然后探讨了高维、奇异情况下如何降低运算量的问题,并给出了一个简单高效的算法。在ORL标准人脸库上进行测试,由该算法抽取的特征在最近邻分类器和最小距离分类器下均达到96%的正确识别率,这一结果优于经典的特征脸和Fisher脸方法在该库上的识别结果。A novel method based on weighted discriminant analysis for face recognition was proposed in this paper. First, the Fisher criterion was redefined by introducing a weighting of the contributions of individual class pairs to the overall criterion. Then, to deal with the high dimensional and singular case in face recognition problems, a simple and efficient algorithm was developed. Finally, the proposed algorithm was tested on ORL face database, and a recognition rate of 96% was achieved by using either a common nearest neighbor classifier or a minimum distance classifier. The experimental results show our method is superior to the classical Eigeafaces and Fisherfaces.

关 键 词：线性鉴别分析 加权Fisher准则 特征抽取 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]