紋理特征提取與自動分類算法研究.pdf

1、天津大學碩士學位論文紋理特征提取與自動分類算法研究姓名：龐鵬飛申請學位級別：碩士專業(yè)：信息與通信工程指導教師：何凱201112ABSTRACTAsanimportanttopicinthecomputervisionfield，textureclassificationhasbroughtgreatinfluenceinnumerousfields，andwhetherwecanextractbeRerfeaturestodisting

2、uishdifferenttexturesisthekeyfactorDifferentfromearlymethods，asanewmultiresolutionanalysistool，waveletisappropriatefortextureimageprocessingforitsperfectspatialfrequencylocalizationperformanceanddiversityAsanewdevelopedt

3、extureanalysismethod，localbinarypattern(LBP)caneffectivelyovercometherotationandnonunif01“111illuminationproblemsforitsrotationandgrayscaleinvariantcharacterMoreoVer，asanewmachineleamingtechniquebasedonstatisticallearnin

4、gtheory，supportvectormachine(SVM)iswidelyusedinrecentyearsforitsperfectperformanceinprocessingclassificationproblemsinhighdimensionspaceThisthesisanalyzedseveralbeaertexturefeatureextractionalgorithmsBesidesthat，somerese

5、archwereperformedfortextureimageclassificationwithSVMThemaincontentsandcontributionsareasfollows：(1)Atextureclassificationapproachbasedonthedualtreecomplexwavelettransform(DTCWT)andsupportvectormachine(SVM)wasproposedInt

6、hismethod，weutilizedtheDTCWTwithapproximatetranslationinvarianceandgooddirectionselectivityFirstlythetextureimagesweredecomposedintomultilayersubbandsviaDTCWTThenthefeaturevectorswerecomputedwithmeanandstandarddeviationo

7、feachsubbandimageFinallythesupportvectormachineswereusedforthetextureclassificationTheexperimentresultsshowedthatbetteraccuracyfortextureclassificationcouldbegotcenwiththeproposedmethod(2)Astherotationinvarianttextureima

8、geisconcemed，aclassificationapproachbasedonadvancedLBPandSVMwasrealizedCommonlythelog—polartransformwasusedtoconverttherotationtotranslation，andthentheinfluenceofrotationandscalechangewasreducedDifferentfromlog—polartran

9、sform，wedirectlyextractedgrayscaleandrotationinvariantfeaturesfromtheoriginalimageswiththeadvancedLBRTheexperimentalresultsshowedthatthismethodcouldeffectivelysolvetheproblemofgray—scaleandrotationinvariant(3)Weappliedte