资源描述：
《motion segmentation with accurate boundaries- a tensor voting approach》由会员上传分享，免费在线阅读，更多相关内容在教育资源-天天文库。
1、MotionSegmentationwithAccurateBoundaries-ATensorVotingApproachMirceaNicolescuandGérardMedioniIntegratedMediaSystemsCenterUniversityofSouthernCaliforniaLosAngeles,CA90089-0273{mnicoles,medioni}@iris.usc.eduAbstracttoillustratetheproblem.FromareaAthatappearsinthefirstimage,onlyha
2、lfisvisibleinthesecondimage,theProducinganaccuratemotionflowfieldisveryotherhalfbeingoccludedbythemovingregion.Atthedifficultatmotionboundaries.Wepresentanovel,non-oppositeside,areaBisstillvisibleinthesecondimage,iterativeapproachforsegmentationfromimagemotion,butisnowsplitinto
3、tworegions,withnew,un-occludedbasedontwovotingprocesses,indifferentdimensionalpixelsinbetween.Evenwherenoocclusiontakesplace,spaces.Byexpressingthemotionlayersassurfacesinasuchasattheupperboundary,areaCisalsosplitinthe4-Dspace,avotingprocessisfirstusedtoenforcethesecondimage,du
4、etothemotionbetweenregions.smoothnessofmotionanddetermineanestimationofConsequently,theapparentmotionaroundboundariespixelvelocities,motionregionsandboundaries.Thecannotbepreciselydeterminedbyusinganysimilarityboundaryestimationisthencombinedwithintensitycriteria,sincetheareasb
5、eingcomparedmusthaveafiniteinformationfromtheoriginalimagesinordertolocallyextent.Moreover,itisnotrealistictoassumethatallthedefineaboundarytensorfield.Thecorrectboundaryiswrongmatchescanbelaterremovedasnoise.Duetotheinferredbya2-Dvotingprocesswithinthisfield,thatsimilarityofpa
6、rtialareas,wrongcorrespondencesareenforcesthesmoothnessofboundaries.Finally,correctoftenassignedinaconsistentmanner,resultinginover-velocitiesarecomputedforthepixelsnearboundaries,asextendedimageregions.theyarereassignedtodifferentregions.WedemonstrateThekeyobservationisthatone
7、shouldnotonlyrelyonourcontributionbyanalyzingseveralimagesequences,motioncuesinordertoperformmotionsegmentation.containingmultipletypesofmotion.Examiningtheoriginalimagesrevealsamultitudeofmonocularcues,suchasintensityedges,thatcanaidinidentifyingthetrueobjectboundaries.Asecond
8、lookat1.IntroductionFigure1willconfirmit.CCThemotivati
