Computer Engineering and Applications ›› 2011, Vol. 47 ›› Issue (1): 179-181.DOI: 10.3778/j.issn.1002-8331.2011.01.050

• 图形、图像、模式识别 • Previous Articles     Next Articles

Improved partial least squares and feature extraction

YANG Maolong1,2,WANG Yuanfang3,SUN Quansen2,XIA Deshen2   

  1. 1.Nanjing New Town Sci-Tech Park,Nanjing 210019,China
    2.College of Computer Science,Nanjing University of Science & Technology,Nanjing 210094,China
    3.Nanjing Northern Information Industry Corporation,Nanjing 211153,China
  • Received:2009-07-13 Revised:2009-09-13 Online:2011-01-01 Published:2011-01-01
  • Contact: YANG Maolong



  1. 1.南京新城科技园,南京 210019
    2.南京理工大学 计算机学院,南京 210094
    3.南京北方信息产业集团,南京 211153
  • 通讯作者: 杨茂龙

Abstract: Non-iterative PLS based on orthogonal constraints(ONIPLS) can extract PLS features rapidly and effectively,while the features maybe correlative.PLS based on Uncorrelated Score Constraints(UCSNIPLS) can extract uncorrelated features which make image recognition more effectively and steadily.2DPLS can extract features from image matrices,which can solve the small sample problems at the same time.While the classical class label encoding is too simple,fuzzy k-near neighbors’method is employed in order to make use of the sample distributions.Then improved algorithms of PLS and 2DPLS based on sample class label encoding are given.The results of experiments on ORL face database show that the improved algorithms presented are better than classical PLS,and can extract discriminant features efficiently.

摘要: 采用基于正交约束的非迭代PLS可以实现PLS成分的快速有效抽取,但不能保证所抽取的成分之间不相关。而基于统计不相关约束的非迭代PLS建模方法所抽取的成分之间是无关的,从而可以保证图像识别时的有效性和稳定性。基于2DPCA思想的2DPLS特征抽取技术,直接从图像矩阵中抽取特征,能有效地解决小样本问题。但在使用PLS对单特征数据进行维数压缩时,传统的类标编码过于简单,为了充分利用数据分布信息,采用模糊k-近邻法对每个样本赋予一个样本标号,将近邻样本类别信息反映在该样本的类编码中,从而提出了基于样本标号的PLS及2DPLS改进算法。在ORL人脸库上的实验结果表明,该改进算法优于传统的PLS,能够更有效地抽取识别特征,其识别率要高于传统的PLS算法。

CLC Number: