欢迎来到天天文库
浏览记录
ID:40557811
大小:390.28 KB
页数:4页
时间:2019-08-04
《a novel text detectin and localization method based oncorner response》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、ANOVELTEXTDETECTIONANDLOCALIZATIONMETHODBASEDONCORNERRESPONSELiSun,GuizhongLiu,XuemingQian,DanpingGuoSchoolofElectronicsandInformationEngineering,Xi’anJiaotongUniversity,710049,ChinaABSTRACTmethodholdtheassumptionthatbackgroundsaremuchmoresmooththantextregions.SoitispossibletoclassifytextInfo
2、rmationoftextinvideosandimagesplaysanimpor-regionandnon-textregionaccordingtoedgeortextureinten-tantroleinsemanticanalysis.Inthispaper,weproposeansity.Buthowtoreducenoisescomingfromcomplexback-effectivemethodfortextdetectionandlocalizationinnoisygroundsisstillanopenproblem.Lyuetal.proposeamet
3、hodbackground.Thealgorithmisbasedoncornerresponse.fordetectingmultilingualandmulti-resolutiontext[4].SobelComparedtonon-textregions,thereoftenexistdenseedgesedgemapsareusedasfeaturesandaspeciallocalthresholdandcornersintextregions.Sowecangetrelativelystrongisadoptedtotolocatethecandidateregio
4、noftext.Lietal.responsesfromtextregionsandlowresponsesfromnon-textproposeamethodbasedoncentralmomentsofimageblockregions.Theseresponsesprovideususefulcuesfortextde-[1].Theyprovethatdifferentscaletextcanbedetectedusingtectionandlocalizationinimages.Thenusingasimpleblockthisfeature;3)Machinelea
5、rningbasedmethodusefeaturesbasedthresholdscheme,wegetcandidateregionsfortext.extractedfromtextregionandnon-textregiontotrainsupportTheseregionsarefurtherverifiedbycombiningotherfeaturesvectormachineorneutralnetworkandthentextdetectionbe-suchascolorandsizerangeofconnectedcomponent.Finally,comes
6、asupervisedclassificationproblem.In[5],aSVMTextlineislocatedaccuratelybytheprojectionofcornerre-basedalgorithmusingresponsefromstrokefilterisproposed.sponse.TheexperimentalresultsshowtheeffectivenessofHuetal.proposeanadaptiveSVMbasedparadigm,basedourmethods.onmaximumgradientdifferencesandotherc
7、onnectedcom-IndexTerms—Textdetection,textlocalization,cornerponentfeatures,whichcanobtainrelativelylowfalserate[6].responseTheshortcomingofmachinelearningbasedmethodisthatitneedsalargenumberoftrainingsamplesofdifferentkind.Inthispaper,anewtex
此文档下载收益归作者所有