局部和全局加权的二维统计不相关鉴别分析  

Local and Global Weighted Uncorrelated Two-dimensional Discriminant Analysis

作　　者：黄明晓 荆晓远[1] 李力[1] 姚永芳[1] 

机构地区：[1]南京邮电大学自动化学院,江苏南京210003

出　　处：《计算机技术与发展》2014年第6期114-117,共4页Computer Technology and Development

基　　金：国家自然科学基金资助项目(61073113);教育部博士点博导类基金(20093223110001);教育部新世纪人才项目(NCET-09-0162);江苏省普通高校研究生科研创新计划(CXLX13_465)

摘　　要：传统的统计不相关鉴别分析方法使用样本的均值来估计期望,计算出总体散度矩阵。这些方法在数据不满足高斯分布的情况下会出现大的偏差,影响最优鉴别特征的提取。为了解决该问题,文中结合二维鉴别分析的思想,分别提出了基于局部的二维统计不相关鉴别变换(L2DUDT)方法和基于全局加权的二维统计不相关鉴别变换(WG2DUDT)方法。L2DUDT通过用样本的近邻中心来定义每个样本的期望,而WG2DUDT用样本间的欧几里得距离加权来定义期望。基于AR和FERET人脸数据库的实验表明,文中提出的方法与一些相关方法相比,有效地提高了识别性能。The traditional uncorrelated discriminant analysis methods employ the mean of sample-set to estimate the expectation for all samples,thus computing the total scatter matrix.However,when the data are not Gaussian distributions,these methods may not extract optimal discriminant features.In order to address this problem,propose two approaches named Local Two-Dimensional Uncorrelated Discriminant Transform（ L2DUDT） and Weighted Global Two-Dimensional Uncorrelated Discriminant Transform（ WG2DUDT） on the basis of two-dimensional discriminant analysis respectively.L2DUDT redefines the expectation for each sample using the sample＇s neighbor center,while WG2DUDT uses Euclidean distance between samples as weighted value to construct the expectation.The experimental results on AR and FERET databases demonstrate that the proposed approaches can effectively improve the recognition performance,as compared with some related methods.

关 键 词：统计不相关鉴别分析 鉴别特征 二维鉴别分析 二维统计不相关鉴别变换 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]