基于主動輪廓模型的圖像分割研究.pdf

1、哈爾濱工程大學(xué)碩士學(xué)位論文基于主動輪廓模型的圖像分割研究姓名：張彤申請學(xué)位級別：碩士專業(yè)：信號與信息處理指導(dǎo)教師：楊秀坤201203基于主動輪廓模型的圖像分割研究ABSTRACTImagesegmentationVisionhasbeenanimportantresearchtopicinthefieldofcomputerVision，itaimstoextract矗ommeback掣oundforta瑪etobjectsW1ichw

2、aSneededIntllesemaIlticleVel，tllesegmentationofmeta瑪etcaJlbeamedicalimageiIlo瑪an，tissue；in、，aders仔omasulⅣeillanceVideo；mechaIlicalmoVingpartsinroboticsornallticalandmili伽了goalⅥsibleimagesegmentationaSabaSicdisciplinefiel

3、d，hasplayedaVe巧impon眥troleinmanypracticalapplicationsInactualimages，therearenoise，weakedgesandotherissuesWhjchbriI塢greatobstaclestothe仃aditionalimagesegmentationmethodAndbecausetlleactiVecontollrmodelcangetsub—piXelaccur

4、acya11dclose，smoothcontourcurVe，hasbeen謝delyusedinmefieldofimagesegmentationActiVecontourmodelbasedondi毹rentfIeaturescanbediVidedinto鉚ocategories：actiVecontourmodelbasedonedgeandactiVecontourmodelbaSedonregionThefonnerus

5、esthegradientinfomationofmeimagetocons饑lctedgestoppiIlg脅ction(ESF)，sothatthecun，eeVolutionstopsattheedgesofmeimage，suchaSageodesicactiVecontourmodel(GAC)；thelatteris恤useofimagestatiSticalimnnationtocontrolmeeV01塒onofcurv

6、e，whichhasabettertreatmente丘’ectforthenoiseandweakedgeOurmodelh2Lst11eadVantagesofbothofthem，whjchbasedontheclaSsicGACmodel，andit、vasembedded謝tllthelocalgrayfittinginfonnation，soourmodelfomledanewmemodWhichcanhandlenonun

7、ifonllgraylmageFirstly，becauseoftheintroductionofSignedPressureFunction(SPF)，ourmethodhaS也eVerygoodprocessingefrectforweakandbluⅡededges；secondly，becausewtlichweuseislocalgrayfittingstatisticalinfo吼ation，sothismethodhasa

8、ceIrtainprocessingcapacityforthenonu工lifo吼鏟ay，aJlditalsohasacertainrobustnessfortllenoise，andbecauseweproposedtoestimatetllelocalfittingValuebyGaussmixturemodel(GMM)aJldEMalgoritllm，so也eestimatedresultismoreaccurate；meaI

9、lW【lile，ollrmodelisbaSedonthegeneralizationofmeGACmodel，itir此ritedmeadVarltagesoft11eactiVecontourmodelwmchwasbasedonmeedge，anditcanusethe叩propriateiIlitialconditionstospeci匆theta瑪etwllichwaSneededt0besegmemedCompared晰tl