计算机工程与应用 ›› 2008, Vol. 44 ›› Issue (29): 36-39.DOI: 10.3778/j.issn.1002-8331.2008.29.010

• 博士论坛 • 上一篇    下一篇

二维共轭正交偏最小二乘分析及图像识别应用

杨茂龙1,2,孙权森2,夏德深2   

  1. 1.南京国际关系学院,南京 210031
    2.南京理工大学 计算机学院,南京 210094
  • 收稿日期:2008-05-27 修回日期:2008-06-23 出版日期:2008-10-11 发布日期:2008-10-11
  • 通讯作者: 杨茂龙

Conjugate orthonormalized partial least squares regression and its application in image recognition

YANG Mao-long1,2,SUN Quan-sen2,XIA De-shen2   

  1. 1.International Studies University,Nanjing 210031,China
    2.Department of Computer Science,Nanjing University of Science & Technology,Nanjing 210094,China
  • Received:2008-05-27 Revised:2008-06-23 Online:2008-10-11 Published:2008-10-11
  • Contact: YANG Mao-long

摘要: 偏最小二乘(PLS)是一种有效的图像特征抽取方法。不同于其他的多元数据分析方法,PLS综合了PCA与CCA的优点,抽取对样本具有最佳解释能力的成分。讨论了偏最小二乘法建模思想及非迭代算法、共轭正交算法和基于2D特征抽取时的算法原理和特点,以及PLS用于图像识别时类隶属矩阵的构造。在ORL与Yale人脸库上的实验结果表明用2DCOPLS抽取的特征进行图像识别的效果更好,更稳定。

关键词: 偏最小二乘, 非迭代偏最小二乘, 共轭正交, 二维特征提取, 图像识别

Abstract: It is an effective approach for image feature extraction by Partial Least Squares(PLS) regression method.The good qualities of PCA and CCA in feature extraction are combined in PLS,which can extract the components interpreting the samples optimally.The theories of feature extracting by PLS,non-iterative PLS and orthonormalized PLS are discussed,and the class membership matrix of 1D and 2D is constructed in image recognition,too.The experiment results on ORL and Yale face image database have shown that the extracted feature by 2DCOPLS can achieve good performance in image recognition.

Key words: Partial Least Squares(PLS), Non-iterative PLS(NIPLS), Conjugate Orthonormalized PLS(COPLS), 2D Feature extraction, image recognition