偏最小二乘回归分析与特征融合在图像识别中的应用  

Image recognition based on partial least squares regression and feature fusion

作　　者：杨茂龙[1,2] 孙权森[2] 夏德深[2] 

机构地区：[1]南京国际关系学院六系,南京210031 [2]南京理工大学计算机学院,南京210094

出　　处：《计算机工程与应用》2009年第26期187-189,215,共4页Computer Engineering and Applications

基　　金：国家自然科学基金No.60773172~~

摘　　要：为了更有效地进行图像识别,对同一模式的不同特征进行融合是有效途径。讨论了偏最小二乘法及其改进算法、特征融合方法在图像识别中的应用。首先讨论了偏最小二乘法的基本原理和非迭代偏最小二乘法、基于共轭正交的偏最小二乘法用于特征抽取的原理和特点,给出了三种特征融合方法,在ORL与Yale人脸库上的实验结果表明进行对用PLS抽取的特征融合后可以有效地进行图像识别。It is an effective approach for image recognition more efficiently by fusing different features of a pattern.The Partial Least Squares regression（PLS） method and its improved algorithms are discussed,as well as feature fusion methods and their application.The theories of feature extracting by PLS,non-iterative PLS and orthonormalized PLS are discussed,and the three feature fusion methods are given,too.The experiment results on ORL and Yale face image database have shown that the fused feature can achieve good performance in image recognition.

关 键 词：偏最小二乘 特征融合 图像识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]