基于最大分离度差的有监督核局部保持投影  

Supervised kernel locality preserving projection algorithm based on maximum separation degree difference

作　　者：邹国锋[1,2] 姜树明[2] 张元元[2] 王科俊[3] 

机构地区：[1]山东理工大学电气与电子工程学院,山东淄博255049 [2]山东省科学院情报研究所,济南250014 [3]哈尔滨工程大学自动化学院,哈尔滨150001

出　　处：《计算机应用研究》2014年第7期2228-2232,共5页Application Research of Computers

基　　金：国家教育部博士点专项科研基金资助项目(20102304110004);中央高校基金资助项目(HEUCF041310)

摘　　要：为有效避免有监督局部保持投影算法在小样本情况下导致类内分离度矩阵奇异的问题,提出一种基于最大分离度差的有监督局部保持投影算法。提出类内分离度和类间分离度的概念,以类内分离度矩阵和类间分离度矩阵为基础构造基于差模式的鉴别准则,并将该算法推广到非线性情况下,提出基于最大分离度差的有监督核局部保持投影算法。Yale、ORL和CAS-PEAL-R1-POSE人脸库上的实验结果表明所提方法是有效的。In order to avoid the singular problem of within-class separation degree matrix caused by small sample problem in supervised kernel locality preserving projection algorithm,this paper proposed a novel supervised locality preserving projection method based on the maximum separation degree difference. It first presented the concept of within-class separation degree and between-class separation degree,then constructed the discriminant criterion of differential mode based on within-class separation degree matrix and between-class separation degree matrix. It extended the algorithm to the nonlinear case,and proposed the supervised kernel locality preserving projection based on maximum separation degree. Finally,experimental results on Yale,ORL and CAS-PEAL-R1-POSE face databases show that the proposed method is effective.

关 键 词：有监督局部保持投影 最大分离度差 小样本问题 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]