資源描述：
《two-stage image segmentation based on edge and region information》由會員上傳分享，免費在線閱讀，更多相關(guān)內(nèi)容在教育資源-天天文庫。
1、Two-stageimagesegmentationbasedonedgeandregioninformationJournalofHarbinInstituteofTechnolo(NewSeries),Vo1.14,No.4,2007Two-stageimagesegmentationbasedonedgeandregioninformationRANXin.QIFei—hu冉鑫,戚飛虎(1.MerchantMarineCollege,ShanghaiMaritimeUniversity,Shanghai20
2、0135,China,E—mail:simonran@sjtu.edu.cn2.Dept.ofComputerScienceandEngineering,ShanghaiJiaotongUniversity,Shahghai200030,China)Abstract:Atwo—stagemethodforimagesegmentationbasedonedgeandregioninformationisproposed.Differ-entdeformationschemesareusedattwostagesf
3、orsegmentingtheobjeetcorrectlyinimageplane.Atthefirststage,thecontourofthemodelisdividedintoseveralsegmentshierarchicallythatdeformrespectivelyusingaf-finetransformation.Afterthecontourisdeformedtotheapproximateboundaryofobject,afinematchmecha—nismusingstatis
4、ticalinformationoflocalregiontoredefinetheexternalenergyofthemodelisusedtomakethecontourfittheobject~boundaryexactly.Thealgorithmiseffective,asthehierarchicalsegmentaldeformationmakesuseoftheglobeandlocalinformationoftheimage,theaffinetransformationkeepstheco
5、nsistencyofthemodel,andthereformativeapproachesofcomputingtheinternalenergyandexternalenergyareproposedtore—ducethealgorithmcomplexity.Theadaptivemethodofdefiningthesearchareaatthesecondstagemakesthemodelconvergequickly.Theexperimentalresultsindicatethatthepr
6、oposedmodeliseffectiveandrobusttolo—calminimaandabletosearchforconcaveobjects.Keywords:activecontourmodel;regiongrowing;imagesegmentation;affinetransformationCLCnumber:TP391Documentcode:AArticleID:1005—9113(2007)04-0533-08Activecontourmodels,alsoknownassnakes
7、…,areenergy—mi'ni'mi'zi'ngsplinesthatcandeformtofitim—agefeaturesundertheinternalforcescomingfromwith.inthecontouritselfandexternalforcescomputedfromtheimagedata.Asactivecontourmodelsutilizetheglobalconstraintsandkeepsmoothandcontinuousdur—ingdeformation,they
8、canextracttheobject~boundarycorrectlyevenwhenlocalboundaryisweakordiscon—nected.Now,activecontourmodelshavebeenusedextensivelyinimageanalysis,videosegmentationandtracking['.andbiomedicala
