資源描述：
《Contour Detection and Hierarchical Image Segmentation》由會員上傳分享，免費在線閱讀，更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、898IEEETRANSACTIONSONPATTERNANALYSISANDMACHINEINTELLIGENCE,VOL.33,NO.5,MAY2011ContourDetectionandHierarchicalImageSegmentationPabloArbela′ez,Member,IEEE,MichaelMaire,Member,IEEE,CharlessFowlkes,Member,IEEE,andJitendraMalik,Fellow,IEEEAbstract—Thispap
2、erinvestigatestwofundamentalproblemsincomputervision:contourdetectionandimagesegmentation.Wepresentstate-of-the-artalgorithmsforbothofthesetasks.Ourcontourdetectorcombinesmultiplelocalcuesintoaglobalizationframeworkbasedonspectralclustering.Oursegmen
3、tationalgorithmconsistsofgenericmachineryfortransformingtheoutputofanycontourdetectorintoahierarchicalregiontree.Inthismanner,wereducetheproblemofimagesegmentationtothatofcontourdetection.Extensiveexperimentalevaluationdemonstratesthatbothourcontourd
4、etectionandsegmentationmethodssignificantlyoutperformcompetingalgorithms.Theautomaticallygeneratedhierarchicalsegmentationscanbeinteractivelyrefinedbyuser-specifiedannotations.Computationatmultipleimageresolutionsprovidesameansofcouplingoursystemtore
5、cognitionapplications.IndexTerms—Contourdetection,imagesegmentation,computervision.?1INTRODUCTIONHISpaperpresentsaunifiedapproachtocontour[4],respectively.ThispaperofferscomprehensiveversionsTdetectionandimagesegmentation.Contributionsofthesealgorith
6、ms,motivationbehindtheirdesign,andinclude:additionalexperimentswhichsupportourbasicclaims.Webeginwithareviewoftheextensiveliteratureon.ahigh-performancecontourdetector,combiningcontourdetectionandimagesegmentationinSection2.localandglobalimageinforma
7、tion,Section3coversthedevelopmentofthegPbcontour.amethodtotransformanycontoursignalintoadetector.Wecouplemultiscalelocalbrightness,color,andhierarchyofregionswhilepreservingcontourquality,texturecuestoapowerfulglobalizationframeworkusing.extensivequa
8、ntitativeevaluationandthereleaseofaspectralclustering.Thelocalcues,computedbyapplyingnewannotateddataset.orientedgradientoperatorsateverylocationintheimage,Figs.1and2summarizeourmainresults.Thetwofiguresdefineanaffinitymatrixrepresentingthesimilarity
