资源描述：
《A Text Detection, Localization and Segmentation System for OCR in Images 》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、ATextDetection,LocalizationandSegmentationSystemforOCRinImages111,2JulindaGllavata,RalphEwerthandBerndFreisleben1SFB/FK615,UniversityofSiegen,D-57068Siegen,Germany2Dept.ofMath.andComputerScience,UniversityofMarburg,D-35032Marburg,Germany{gllavata,ewerth}
2、@fk615.uni-siegen.de;freisleb@informatik.uni-marburg.deAbstractretrievalofimages(orvideos).Atfirst,extractionoftextmayseemtobeatrivialapplicationforexistingOnewaytoincludesemanticknowledgeintotheopticalcharacterrecognition(OCR)tools.However,processofinde
3、xingdatabasesofdigitalimagesistocomparedtoOCRfordocumentimages,extractingusecaptiontext,sinceitprovidesimportanttextfromrealimagesfacesnumerouschallengesdueinformationabouttheimagecontentandisaverytolowerresolution,unknowntextcolor,sizeandgoodentityforqu
4、eriesbasedonkeywords.Inthisposition,orcomplexbackgrounds.paper,weproposeanapproachtoautomaticallyTextextractionandrecognition,whichincludestextlocalize,segmentandbinarizetextappearingindetection,localization,binarizationandrecognition,iscompleximages.Fir
5、st,anunsupervisedmethodbasedausefulprocessregardingtext-basedimageindexing.onawavelettransformisusedtoefficientlydetecttextThefirstthreeprocessingstagesareimportanttoregions.Second,connectedcomponentsaregenerated,achievehigh-qualitytextrecognitionresults
6、whenandtheexacttextpositionsarefoundviaarefinementapplyinganOCRsystem.Therearetwoclassesofalgorithm.Third,anunsupervisedlearningmethodformethodsfortextlocalization:connectedcomponenttextsegmentationandbinarizationisappliedusinga(CC)-basedandtexture-based
7、methods.Thefirstclasscolorquantizerandawavelettransform.Comparativeofmethods[1,2,3]employsconnectedcomponentexperimentalresultsdemonstratetheperformanceofanalysis,whichisbasedontheanalysisoftheourapproachforthemainprocessingsteps:textgeometricalarrangeme
8、ntofedgesorhomogeneouslocalizationandsegmentation,andinparticulartheircolorandgrayscalecomponentsthatbelongtocombination.characters.Theyaresimpletoimplement,buttheyarenotveryrobustfortextlocalizationinimageswithcomplexback
