资源描述：
《Efficient Algorithms for Image Template and Dictionary Matching》由会员上传分享，免费在线阅读，更多相关内容在应用文档-天天文库。
1、JournalofMathematicalImagingandVision12,81–90(2000)°c2000KluwerAcademicPublishers.ManufacturedinTheNetherlands.EfficientAlgorithmsforImageTemplateandDictionaryMatchingSUNG-HYUKCHADepartmentofComputerScienceandEngineering,StateUniversityofNewYorkatBuffaloscha@cse.buffalo.eduAbstract.Givenalargetextim
2、ageandasmalltemplateimage,theTemplateMatchingProblemisthatoffindingeverylocationwithinthetextwhichlookslikethepattern.Thisproblem,whichhasreceivedattentionforlow-levelimageprocessing,hasbeenformalizedbydefiningadistancemetricbetweenarraysofpixelsandfindingallsubarraysofthelargeimagewhicharewithinsomet
3、hresholddistanceofthetemplate.Theseso-calledmetricmethodstendstobetooslowformanyapplications,sinceevaluatingthedistancefunctioncantaketoomuchtime.Wepresentamethodforquicklyeliminatingmostpositionsofthetextfromconsiderationaspossiblematches.Theremainingcandidatepositionsarethenevaluatedonebyoneagain
4、stthetemplateforamatch.Wearestillguaranteedtofindallmatchingpositions,andourmethodgivessignificantspeed-ups.Finally,weconsidertheproblemofmatchingadictionaryoftemplatesagainstatext.Wepresentmethodswhicharemuchfasterthanmatchingthetemplatesindividuallyagainsttheinputimage.Keywords:templatematching,met
5、ricssimilarity,filtration,dictionarymatchingI.IntroductionPtheM-differencebetweenSi;jandTasM.T;Si;j/Dm;mjI.iCk;jCl/¡T.k;l/j,thatis,asthesumofkD0;lD0Recenttechniqueshavefacilitatedthesolutionofprevi-theabsolutevaluesofdifferencesincorrespondinggreyouslyinfeasiblequeriessuchas,“Findallimageswhichlevel
6、s.ThesmallertheM-difference,themoresimilarlooklikethispicture,”or“Findallimageswhichcon-Si;jistoT.WedefinetheImageTemplateMatchingtainatumorlikethis”[5].Althoughthereisnouniver-Problem(ITM)assaldefinitionfortheterm“lookslike”,theideathatoneimagelookslikeanotherisusuallymodeledintermsofInput:AnimageI,
7、atemplateT,andathresholdasimilaritymeasures.Inthissetting,animagelooksvalued.likeanotherifthedistancebetweenthemislessthanOutput:All.i;j/’ssuchthatM.Si;j;T/·d.somethresholdvalue.Manysimilaritymeasureshavebe
