基于多分辨共生矩阵的纹理图像分类

《基于多分辨共生矩阵的纹理图像分类》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、计算机研究与发展ISSN1000—1239／CN1i-1777／TPJournalofComputerResearchandDevelopment48(11)：1991—1999，2011基于多分辨共生矩阵的纹理图像分类钟桦杨晓鸣焦李成(智能感知与图像理解教育部重点实验室(西安电子科技大学)西安710071)(hzhong@mail．xidian．edu．cn)TextureClassificationBasedonMultires0lutiOnCo-occurrenceMatrixZhongHua，YangXiaoming，andJiaoLicheng(KeyLaborator

2、yofIntelligentPerceptionandImageUnderstanding(XidianUniversity)，MinistryofEducation，Xian710071)AbstractGraylevelCO—occurrencematrix(GLCM)iSwidelyusedforthedescriptionofdifferenttexturesbecauseofitsadvantageofrepresentingthetexturestructure．Inordertoholdthemultiresolutionpropertyandspectruminf

3、ormationsimultaneously，GLCMisoftencomputedfromeachwaveletsubhand，which，however，leadstomuchhigherdimensionoffeature．Toovercomethisproblem，anewfeatureextractingalgorithmisproposednamedmultiresolutionCO—occurrencematrix(MCM)，whosefeaturevectorhasacomparativelOWdimension．Forthetaskofdimensionredu

4、ction，thoughtheMC：MiScomputedfromeachsubbandofundecimatedwavelettransformationasGLCMdoes，theparametersfortheMCMarewellchosentakingintoconsiderationthescaleandorientationofwaveletsubbands．Inaddition，thefeaturedimensionoftheMCMcanbefurtherreducedthroughanewfeatureselectionmethodbasedonthecorrel

5、ationanalysisofinter—subbandsandintra—subbands．PerformanceanalysisismadeindetailforthestatisticsoftheMCMandwaveletenergyfeature．TheproposedMCMismeasuredthroughtheclassificationtestontheBrodatzalbum．ExperimentalresultsdemonstratethatMCMstatisticsoutperformsothermethodssuchaswaveletenergy，GLCM，

6、andthecatenationofthesetwofeatures．Extensiveexperimentsonfeatureselectionshowtheeffectivenessoftheproposedmethod，whichcanreducethedimensionsuccessfullywithoutanylossofclassificationperformance．Keywordstextureclassification；multiresolution；CO—occurrencematrix；undecimatedwavelettransformation；f

7、eatureselection摘要共生矩阵是描述纹理特征的一种常用方法．首先提出一种新的特征提取算法——多分辨共生矩阵．多分辨共生矩阵是通过同时在非下采样小波变换的逼近子带和细节子带上提取共生矩阵来实现的，能够有机整合传统小波的多分辨特性和频谱信息，以及空域灰度共生矩阵的纹理结构信息．其次，分析了多分辨共生矩阵、灰度共生矩阵以及小波能量特征的物理意义，并从相关性出发提出了新的特征选择方法，有效地降低了特征维数．对标准纹理库的分类实验结果表明：多分辨共生特征对纹理具有更好的收稿日期