计算机工程与应用 ›› 2010, Vol. 46 ›› Issue (4): 1-3.DOI: 10.3778/j.issn.1002-8331.2010.04.001

• 博士论坛 • 上一篇    下一篇

结合小波变换和图像主元分析的人脸识别

杨 军1,2,3,袁红照1,2   

  1. 1.四川大学 图形图像研究所,成都 610064
    2.四川大学 视觉合成图形图像技术国防重点实验室,成都 610064
    3.四川师范大学 计算机科学学院,成都 610066
  • 收稿日期:2009-08-12 修回日期:2009-11-18 出版日期:2010-02-01 发布日期:2010-02-01
  • 通讯作者: 杨 军

Combining wavelet transform with image principal component analysis for face recognition

YANG Jun1,2,3,YUAN Hong-zhao1,2   

  1. 1.Institute of Image & Graphic,Sichuan University,Chengdu 610064,China
    2.Key Laboratory of Fundamental Synthetic Vision Graphics and Image for National Defense,Sichuan University,Chengdu 610064, China
    3.College of Computer Science,Sichuan Normal University,Chengdu 610066,China
  • Received:2009-08-12 Revised:2009-11-18 Online:2010-02-01 Published:2010-02-01
  • Contact: YANG Jun

摘要: 提出了一种基于小波变换和图像主元分析(IMPCA)相结合的人脸识别方法。小波变换具有保留主要信息,去除噪声的作用,对人脸图像进行小波变换,对变换后的近似图像采用IMPCA方法进行识别。IMPCA是一种快速有效的直接通过图像抽取特征的方法,从图像重构的角度分析了实现IMPCA的两种模式,两种模式分别增强了图像的行特征和列特征,将它们的识别结果进行决策融合可以获得更好的识别效果。基于ORL人脸数据库的实验表明,所提出的方法在识别率上优于单独的IMPCA方法。

Abstract: A face recognition method based on wavelet and Image Principal Component Analysis(IMPCA) is presented.Approximate coefficients of an image can be gotten and its noise is weakened by transforming it with wavelet.The proposed method firstly transforms face image with wavelet to get approximate image,then recognizes with the approximate image based on IMPCA.IMPCA is a rapid feature extract method from matrix itself,no needing regard an image as a vector.This paper presents the basic theory of IMPCA from the view of minimizing the mean reconstruction error and shows two different modules of feature extract based on IMPCA.It analyzes the feature generated from two modules and finds they respectively enhance row characters and column characters.It fuses the recognition results with two features to achieve better accuracy rate.The experiment result on ORL face database shows the proposed method is efficient and the recognition accuracy rate is better than IMPCA only.

中图分类号: