资源描述:
《cam12-02Augmented Lagrangian Method for A Mean Curvature.pdf》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、AugmentedLagrangianMethodforAMeanCurvatureBasedImageDenoisingModelWeiZhu∗,Xue-ChengTai†,andTonyChan‡AbstractHighorderderivativeinformationhasbeenwidelyusedindevelopingvariationalmodelsinimageprocessingtoaccomplishmoreadvancedtasks.However,itisanontrivialissuetoconstr
2、uctefficientnumericalalgorithmstodealwiththeminimizationofthesevariationalmodelsduetotheassociatedhighorderEuler-Lagrangeequations.Inthispaper,weproposeanefficientnumericalmethodforameancurvaturebasedimagedenoisingmodelusingtheaugmentedLagrangianmethod.Wedescribehowtocon
3、structanaugmentedLagrangianfunctionalforthemodelanddetailtheproceduresoffindingtherelatedsaddle-pointsofthefunctional.Wepresentnumericalexperimentstoillustratetheeffectivenessandefficiencyoftheproposednumericalmethod,andshowafewimportantfeaturesoftheimagedenoisingmodelsu
4、chaskeepingcornersandimagecontrast.Moreover,acomparisonwiththegradientdescentmethodfurtherdemonstratestheefficiencyoftheproposedaugmentedLagrangianmethod.1IntroductionImagedenoisingprovidesacrucialsteptowardsthecapturingimagesignalsfromnoisyimageswithlotsofapplications
5、inmedicalimagingandvideomonitoringetc.Duringthelastthreedecades,numerousmethodshavebeenappliedtodealwiththisproblem,suchasvariationalmethods,partialdifferentialequations,statisticalmethodsandsoon[3,4,8,23,24,25,27,28,29,33].Amongthesevariationalmethodisoneofthemostpop
6、ularmethods.AclassicalvariationalmodelonimagedenoisingwasdevelopedbyRudin,Osher,andFatemi(ROF)[29],inwhichachievingacleanimagefromanoisyoneamountstotheminimizationofthefollowingfunctional:ZZE(u)=λ
7、∇u
8、+(f−u)2,(1)ΩΩwheref:Ω→RisagivennoisyimagedefinedonΩ(alwaysarectangle
9、inR2)andλ>0isapositiveconstant.TheROFmodelhasprovedtobeveryefficientinthenoiseremoval∗DepartmentofMathematics,UniversityofAlabama,Box870350,Tuscaloosa,AL35487.E-mail:wzhu7@bama.ua.edu.†DepartmentofMathematics,UniversityofBergen,NorwayandDivisionofMathematicalSciences,S
10、choolofPhysicalandMathematicalSciences,NanyangTechnologicalUniversity,637371,Singapore.Email:tai@math.uib.no.‡OfficeofthePresident,Ho