基于广义典型相关分析的仿射不变特征提取方法

张洁玉; 陈强; 白小晶; 孙权森; 夏德深

doi:10.3724/SP.J.1146.2008.01344

基于广义典型相关分析的仿射不变特征提取方法

doi: 10.3724/SP.J.1146.2008.01344

基金项目:

国家自然科学基金(60773172)和江苏省博士后基金(AD41158)资助课题

计量
- 文章访问数: 3882
- HTML全文浏览量: 90
- PDF下载量: 1006
- 被引次数: 0
出版历程
- 收稿日期: 2008-10-14
- 修回日期: 2009-03-23
- 刊出日期: 2009-10-19

Affine Invariant Feature Extraction Algorithm Based on Generalized Canonical Correlation Analysis

摘要

摘要: 该文结合广义典型相关分析(GCCA)理论，提出了一种新的图像仿射不变特征提取方法。首先，基于多尺度自卷积变换(MSA)构造了一组新的变换量多尺度自卷积熵(MSAE)。然后证明了该熵具有仿射不变性；再利用GCCA将MSA和MSAE变换值作为两种特征进行融合，得到具有更丰富图像信息的组合特征。最后利用MSA，MSAE和组合特征，结合最近距离分类器分别对视点变换图像以及加噪声、加部分遮挡视点变换图像进行分类识别实验。结果表明，组合特征得到了最高的正确识别率，MSAE次之，MSA最低。
- 图像识别; 多尺度自卷积; 多尺度自卷积熵; 特征融合; 仿射不变性
Abstract: A novel method of extracting affine invariant feature is proposed using the theory of Generalized Canonical Correlation Analysis(GCCA). First, a new kind of transformation named MSAE is constructed based on MSA. Second, MSAE is proved to be affine invariant. Then MSA is combined with MSAE using GCCA to obtain a new feature with more information. Finally, the coil-100 image database viewed from different angles in the case of Gaussian noise or occlusion is put into recognition experiments using minimum distance classifier. The comparing results among MSA, MSAE and combined feature indicate that the combined feature can obtain highest recognition accuracy followed by MSAE and MSA in turn.

HTML全文

参考文献(1)

MingKuei H. Visual pattern recognition by momentinvariants [J]. IEEE Transactions on Information Theory,1962, 8(2): 179-187.[2]Zahn C T and Roskies R Z. Fourier descriptors for planeclosed curves [J].IRE Transactions on Computer.1972, 21(3):269-281[3]Teague M. Image analysis via the general theory of moments[J].Journal of the Optical Society of America.1980, 70(8):920-930[4]Resnick J. The radon transforms and some of its applications[J].IEEE Transactions on Acoustics, Speech and SignalProcessing.1985, 33(1):338-339[5]Xiong H, Zhang T, and Moon Y S. A translation- and scaleinvariantadaptive wavelet transform [J].IEEE Transactionson Image Processing.2000, 9(12):2100-2108[6]夏永泉, 刘正东, 杨静宇. 不变矩方法在区域匹配中的应用[J].计算机辅助设计与图形学学报, 2005, 17(10): 2152-2156.Xia Yong-quan, Liu Zheng-dong, and Yang Jing-yu.Application of moment invariant approach in region matching[J]. Journal of Computer-aided Design Computer Graphics,2005, 17(10): 2152-2156.[7]Rahtu E, Salo M, and Heikkil. J. Affine invariant patternrecognition using multiscale autoconvolution[J].IEEETransactions on Pattern Analysis and Machine Intelligence.2005, 27(6):908-918[8]Petrou M and Kadyrov A. Affine invariant features from thetrace transform [J].IEEE Transactions on Pattern Analysisand Machine Intelligence.2004, 26(1):30-44[9]蔡红苹, 雷琳, 陈涛等. 一种通用的仿射不变特征区域提取方法[J]. 电子学报, 2008, 36(4): 672-678.Cai Hong-ping, Lei Lin, and Chen Tao, et al.. A generalapproach for extracting affine invariant regions[J]. ActaElectronica Sinica, 2008, 36(4): 672-678.[10]刘小军, 杨杰, 刘惠等. 基于主成分分析的仿射不变特征图像匹配方法[J]. 系统仿真学报, 2008, 20(4): 977-980.Liu Xiao-jun, Yang Jie, and Liu Hui, et al.. Affine invariantfeatures image matching approach based on principalcomponents analysis [J]. Journal of System Simulation, 2008,20(4): 977-980.[11]Kannala J, Rahtu E, and Heikkil. J. Affine registration withmulti-scale autoconvolution[C]. Proc. InternationalConference on Image Processing, Genoa, 2005, 3: 1064-1067.[12]Rahtu E, Salo M, and Heikkil. J. Multiscale autoconvolutionhistograms for affine invariant pattern recognition[C]. Proc.16th British Machine Vision Conference, Edinburgh, 2006, 3:1039-1048.[13]唐涛, 粟毅, 陈涛等. 一种新的图像局部仿射不变特征提取方法[J]. 计算机仿真, 2007, 7(24): 229-234.Tang Tao, Su Yi, and Chen Tao, et al.. A novel method forlocal affine invariant feature extraction [J]. ComputerSimulation, 2007, 7(24): 229-234.[14]徐学强, 汪渤, 于家城等. 一种新型不变矩在图像识别中的应用[J]. 光学技术, 2007, 33(4): 580-583.Xu Xue-qiang, Wang Bo, and Yu Jia-cheng, et al..Application of a new invariant moments on image recognition[J]. Optical Technique, 2007, 33(4): 580-583.[15]Kim T-K, Kittler J, and Cipolla R. Discriminative learningand recognition of image set classes using canonicalcorrelations[J]. IEEE Transactions on PAMI, 2007, 29(6):1005-1018.[16]洪泉, 陈松灿, 倪雪蕾. 子模式典型相关分析及其在人脸识别中的应用[J]. 自动化学报, 2008, 34(1): 21-30.Hong Quan, Chen Song-can, and Ni Xue-lei. Sub-patterncanonical correlation analysis with application in facerecognition[J]. Acta Automatica Sinica, 2008, 34(1): 21-30.[17]Sun Q S, Zeng S G, Liu Y, Heng P A, and Xia D S. A newmethod of feature fusion and its application in imagerecognition[J].Pattern Recognition.2005, 38(12):2437-2448[18]孙权森, 曾生根, 王平安等. 典型相关分析的理论及其在特征融合中的应用[J]. 计算机学报, 2005, 28(9): 1524-1533.Sun Quan-sen, Zeng Sheng-gen, and Wang Ping-an, et al..The theory of canonical correlation analysis and itsapplication to feature fusion[J]. Chinese Journal of Computer,2005, 28(9): 1524-1533.[19]Sun Q S, Heng P A, Jin Z, and Xia D S. Face recognitionbased on generalized canonical correlation analysis[C].International Conference on Intelligent Computing(Hefei,China), Lecture Notes in Computer Science, Springer-Verlag,Heidelberg, Berlin, 2005, 3645: 958-967.[20]雷琳, 蔡红苹, 唐涛等. 基于MSA特征的遥感图像多目标关联算法[J]. 遥感学报, 2008, 12(4): 586-592.Lei Lin, Cai Hong-ping, and Tang Tao, et al.. A MSAfeature-based multiple targets association algorithm inremote sensing images[J]. Journal of Remote Sensing, 2008,12(4): 586-592.[21]Heikkila J. Pattern matching with affine momentdescriptors[J].Pattern Recognition.2004, 37(9):1825-1834[22]Yang Z and Cohen F. Cross-weighted moments and affineinvariants for image registration and matching [J].IEEETransactions on Pattern Analysis and Machine Intelligence.1999, 21(8):804-814[23] Columbia University, Coil-100 image database. Http://www1.cs.columbia.edu/CAVE/ software/softlib/coil-100.php,2008, 8.

施引文献

资源附件(0)

访问统计