正交约束的无监督统计不相关最佳鉴别平面  

The Unsupervised and Uncorrelated Optimal Discriminant Plane Based on Orthogonal Constraints

作　　者：曹苏群[1,2] 王骏[1,3] 王士同[1] 

机构地区：[1]江南大学信息学院,江苏无锡214122 [2]淮阴工学院机械工程学院,江苏淮安223003 [3]南京理工大学计算机科学与技术学院,江苏南京210093

出　　处：《计算机工程与科学》2010年第5期34-36,153,共4页Computer Engineering & Science

基　　金：国家863计划资助项目(2007AA1Z158);国家自然科学基金重点项目(60773206;60704047);江苏省高校自然科学重大基础研究资助项目(09KJA460001);江苏省高校自然科学重大基础研究资助项目(09KJA460001);淮安市国际科技合作项目(HG004);淮阴工学院青年科技基金资助项目(HGQN0701)

摘　　要：赵海涛等提出的改进的最佳鉴别平面(IODP)只能用于有监督模式,基于此,本文提出将IODP扩展到无监督模式下的方法。在优化模糊Fisher准则求取第一条最佳鉴别矢量的基础上,求取同时满足正交约束与模糊总体散布矩阵共轭正交约束的第二条最佳鉴别矢量,构成正交约束的无监督统计不相关最佳鉴别平面(OUUODP),进而获得一种新的无监督特征抽取方法。对CMU-PIE人脸数据库进行实验,结果表明,当类别差异较大时,该方法能够抽取有利于分类的特征,获得了优于主成分分析与独立成分分析方法的性能。An improved optimal discriminant plane（IODP） proposed by Zhao can only be used in the supervised pattern. Based on this point, a novel method is presented to extend IODP to the unsupervised pattern. On the basis of optimizing the fuzzy Fisher criterion to calculate the first optimal discriminant vector, the second optimal discriminant vector with the or- thogonal constraint and the conjugated orthogonal constraint of the fuzzy total-class scatter matrix can be figured out. These two vectors constitute the orthogonal-constraint-based unsupervised and uncorrelated optimal discriminant plane（OUUO- DP）. With these, a novel unsupervised feature extraction method is obtained. The experimental results for the CMU-PIE face database demonstrate that this method can extract the features which are conducive to classification and is superior to principal component analysis and independent component analysis when the between-class difference is big.

关 键 词：无监督模式 特征降维 最佳鉴别平面 人脸识别 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]