首页 | 本学科首页   官方微博 | 高级检索  
     

2DPCA在图像特征提取中优于PCA的判定条件
引用本文:程正东,章毓晋,樊祥.2DPCA在图像特征提取中优于PCA的判定条件[J].工程数学学报,2009,26(6).
作者姓名:程正东  章毓晋  樊祥
作者单位:程正东(清华大学电子工程系,北京100084;合肥电子工程学院,合肥230037);章毓晋(清华大学电子工程系,北京,100084);樊祥(合肥电子工程学院,合肥230037;中国科技大学六系,合肥230027) 
基金项目:国家自然科学基金(60872084):高等学校博士学科点专项科研基金 
摘    要:主元分析(PCA)与二维主元分析(2DPCA)是两种典型的图像特征提取方法,它们所提取的图像特征的优劣可由重建误差来衡量.通过对PCA和2DPCA的重建误差分析发现,二者的重建误差在理论上相同,但在实际应用中取决于它们的样本协方差阵的估计准确度.本文以均方误差为度量给出了PCA与2DPCA样本协方差阵的估计准确度表达式,并由此得到2DPCA图像特征优于PCA的判定条件是2DPCA协方差阵的特征值平方和大于PCA.本文还指出行2DPCA与列2DPCA所提取的图像特征孰优孰劣也取决于它们各自协方差阵的特征值平方和的大小.在人脸图像库与人脸表情图像库上的实验验证了上述判定条件的正确性.

关 键 词:图像特征  协方差阵  重建误差  均方误差

Criteria for 2DPCA Superior to PCA in Image Feature Extraction
CHENG Zheng-dong,ZHANG Yu-jin,FAN Xiang.Criteria for 2DPCA Superior to PCA in Image Feature Extraction[J].Chinese Journal of Engineering Mathematics,2009,26(6).
Authors:CHENG Zheng-dong  ZHANG Yu-jin  FAN Xiang
Affiliation:CHENG Zheng-dong1,2,ZHANG Yu-jin1,FAN Xiang2,3(1-Department of Electronic Engineering,Tsinghua University,Beijing 100084,2-Hefei Electronic Engineering Institute,Hefei 230037,3-Six Department,Science and Technology University of China,Hefei 230027)
Abstract:PCA and 2DPCA are two typical image feature extraction methods.The wellness of image features extracted by PCA and 2DPCA can be judged by their reconstruction errors.Our analysis shows that the reconstruction errors of PCA and 2DPCA are equal in theory,but depend on the estimation accuracy of their covariance matrices in practice.This paper gives the accuracy expressions for estimations of covariance matrices for PCA and 2DPCA,respectively,based on the criterion of mean square error.Further,it is shown that...
Keywords:PCA  2DPCA  PcA  2DPCA  image feature  covariance matrix  reconstruction error  mean square error
本文献已被 CNKI 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号