基于小波分解系数的贝叶斯人脸识别方法  

BAYESIAN FACE RECOGNITION BASED ON THE COEFFICIENTS OF WAVELET DECOMPOSITION

作　　者：彭进业[1] 王大凯[2] 俞卞章[1] 李楠[2] 

机构地区：[1]西北工业大学电子工程系,西安710068 [2]西北大学电子科学系,西安710069

出　　处：《光子学报》2001年第10期1263-1269,共7页Acta Photonica Sinica

基　　金：陕西省科技研究发展计划 (8k0 7-G4);西安市科技攻关计划 (GG990 7)资助项目

摘　　要：本文给出了贝叶斯人脸识别方法中匹配准则的多个近似表达式及一种实用的快速计算方法 .在此基础上 ,利用反对称双正交小波变换的微分算子功能 ,提出了一种利用两幅人脸图象的小波变换系数差作为模式矢量的贝叶斯人脸识别方法 ,并利用 AR人脸图象库进行了实验 .实验结果表明本文方法与基于图象灰度的类似方法相比 ,识别率提高 8%左右 .In this paper,a number of approximate expressions for matching score in the Bayesian face recognition method and their calculations are presented.And then a new pattern vector that is formed by using coefficients of anti symmetrical biorthogonal wavelet decomposition is proposed for Bayesian face recognition.Experiments with AR facial database show that the performance of our method is 8% better than that of similar method based on the pattern vector formed by grey differences.In addition,our method provides a way to recognize facial image in compressed data domain without need of reconstructing the image.

关 键 词：人脸识别 小波分解系数 反对称双正交小波 贝叶斯理论 匹配准则 图像处理 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] Q983.1[自动化与计算机技术—计算机科学与技术]