预览加载中，请您耐心等待几秒...

基于区域MRF和贝叶斯置信传播的SAR图像分割.docx

预览
1/2
2/2

在线预览结束，喜欢就下载吧，查找使用更方便

下载提示 文本预览

如果您无法下载资料，请参考说明：

1、部分资料下载需要金币，请确保您的账户上有足够的金币

2、已购买过的文档，再次下载不重复扣费

3、资料包下载后请先用软件解压，在使用对应软件打开

基于区域MRF和贝叶斯置信传播的SAR图像分割 1.Introduction SyntheticApertureRadar(SAR)imagesegmentationhasgainedincreasingattentioninremotesensingapplications.InordertoextractusefulinformationfromSARimagery,accurateandautomaticsegmentationisessential.DuetothecomplexityofspecklenoiseandtexturepatternsinSARimagery,traditionalsegmentationmethodsfailtoachievesatisfactoryresults.Therefore,anewapproach,basedonMarkovRandomFields(MRF)andBayesianbeliefpropagation,isproposedinthispaper. 2.MarkovRandomFields MRFisbasedonasetofrandomvariablesthatarelinkedtogetherbyagraphstructure.Thegraphrepresentsthejointprobabilitydistributionofthevariables.Inimagesegmentation,eachpixelisconsideredarandomvariable,andthegraphstructureisdefinedbythespatialrelationshipsbetweenpixels. InSARimagesegmentation,eachpixelisassociatedwithalabelthatrepresentstheclassmembership.ThegoalistofindtheoptimallabelingconfigurationthatmaximizestheenergyfunctiondefinedontheMRF.Theenergyfunctionconsistsoftwoterms:thedatatermandthesmoothnessterm. Thedatatermmeasuresthelikelihoodofeachpixelbeingassignedtoaspecificclassbasedonitslocalandcontextualinformation.Itisdefinedas: E_data=-log(P(data|label)) wheredatarepresentstheSARimagedata,andlabelrepresentsthelabelassignedtoeachpixel. Thesmoothnesstermmeasuresthecoherenceoflabelsamongneighboringpixels.Itisdefinedas: E_smooth=∑w_ijI(label_i=label_j) wherew_ijisaweightthatrepresentsthestrengthofconnectionbetweenpixelsiandj,andIistheindicatorfunction. Theenergyfunctionisminimizedusingtheiterativegraph-cutalgorithm,whichupdatesthelabelofeachpixelbasedontheneighboringlabelsandtheenergyfunction. 3.BayesianBeliefPropagation Bayesianbeliefpropagationisaprobabilisticinferencealgorithmthatupdatesthebelieforprobabilitydistributionovervariablesbasedontheobserveddataandthepriorknowledge.Itisusefulinimagesegmentationbecauseitenablestheuseofpriorknowledgeaboutthestatisticalpropertiesoftheimage. InSARimagesegmentation,thepriorknowledgeisrepresentedbytheMRF,whichdefinesthestatisticalrelationshipbetweenpixels