Learning Probabilistic Models for Contour Completion in Natural Images英文文献资料

《Learning Probabilistic Models for Contour Completion in Natural Images英文文献资料》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、IntJComputVisDOI10.1007/s11263-007-0092-6LearningProbabilisticModelsforContourCompletioninNaturalImagesXiaofengRen·CharlessC.Fowlkes·JitendraMalikReceived:17August2005/Accepted:11September2007©SpringerScience+BusinessMedia,LLC2007AbstractUsingalargesetofhumansegmentednatural1Introductionimages

2、,westudythestatisticsofregionboundaries.WeobserveseveralpowerlawdistributionswhichlikelyariseFindingtheboundariesofobjectsandsurfacesinascenefrombothmulti-scalestructurewithinindividualobjectsisaproblemoffundamentalimportanceforcomputervi-andfromarbitraryviewingdistance.Accordingly,wede-sion.F

3、orexample,thereisalargebodyofworkonob-jectrecognitionwhichreliesonboundarydetectiontopro-velopascale-invariantrepresentationofimagesfromthevideinformationaboutobjectshape(e.g.Borgefors1988;bottomup,usingapiecewiselinearapproximationofcon-Huttenlocheretal.1993;Belongieetal.2002;Felzenszwalbtour

4、sandconstrainedDelaunaytriangulationtocomplete2001).Evenincaseswheresimpleintensityfeaturesaresuf-gaps.Wemodelcurvilineargroupingontopofthisgraph-ficientforobjectdetection,e.g.faces,itisstilldesirableical/geometricstructureusingaconditionalrandomfieldtoincorporateboundarydetectioninordertoprovid

5、epre-tocapturethestatisticsofcontinuityanddifferentjunc-ciseobjectsegmentation(e.g.BorensteinandUllman2002;tiontypes.QuantitativeevaluationsonseverallargedatasetsTuetal.2005;Yuetal.2002).Theavailabilityofhighqual-showthatourcontourgroupingalgorithmconsistentlydom-ityestimatesofboundarylocation

6、willultimatelygoverninatesandsignificantlyimprovesonlocaledgedetection.whetherthesealgorithmsaresuccessfulinreal-worldsceneswhereclutterandtextureabound.TheproblemofboundarydetectionhasbeenattackedatKeywordsGrouping·Naturalimages·Boundaryseveraldifferentlevels:detection·Scaleinvariance·Conditio

7、nalrandomfields·1.Localedgedetection:thisisthetraditionalapproachtoMachinelearningboundarydetectionandhasbeenanareaofcentralre-searchsincetheearlydaysofcomputervision.Alocaledgedetectortypicallyconsidersasmallpatchcenteredateachimageloca