资源描述：
《Chapter 5 - Image Restoration》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、§5-ImageRestorationDr.DirkSchniedersDepartmentofComputerScienceTheUniversityofHongKongIntroduction★Asinimageenhancement,theultimategoalofimagerestorationistoimprovetheimageinsomepredefinedsense★Restorationattemptstorestoreorrecoveranimagethathasbeendegraded,byusingaprioriknowledgeofthedegrad
2、ationphenomenon★Approaches★Modelthedegradationandapplytheinverseprocessinordertorecovertheoriginalimage★Spatialdomaintechniquesandfrequencydomaintechniques2013-2014COMP75022ImageEnhancementvs.ImageRestoration★ImageEnhancement★Largelyasubjectiveprocess★Heuristicproceduresdesignedtomanipulatea
3、nimageinordertotakeadvantageofthepsychophysicalaspectsofhumanvisualsysteme.g.,contraststretching★ImageRestoration★Formostpartanobjectiveprocess★Involvesformulatingacriterionofgoodnessthatwillyieldanoptimalestimateofthedesiredresulte.g.,removalofimageblurbyablurringfunction2013-2014COMP75023I
4、mageDegradationandRestorationg(x,y)f(x,y)f’(x,y)n(x,y)★ThedegradationprocessismodeledasadegradationfunctionHthat,togetherwithanadditivenoisetermn(x,y),operatesonaninputimagef(x,y)toproduceadegradedimageg(x,y)★Giveng(x,y),someknowledgeaboutHandn(x,y),theobjectiveofrestorationistoobtainanestim
5、atef’(x,y)oftheoriginalimage2013-2014COMP75024ImageDegradationandRestoration★Itisdesirabletohavetheestimatef’(x,y)asclosetotheoriginalinputimageaspossible★Ingeneral,themoreknowledgeaboutHandn(x,y)thatisavailable,thecloserf’(x,y)willbetof(x,y)★IfHisalinear,position-invariantprocess,thenthedeg
6、radedimageisgiveninthespatialdomainbyandinthefrequencydomainby2013-2014COMP75025NoiseModels★Theprinciplesourcesofnoiseindigitalimagesariseduringimageacquisition(performanceofimagingsensors)andtransmission(interferenceinchannel)★Tosimplifytheanalysis,assumethenoiseis★Independentofspatialcoord
7、inates★Uncorrelatedwithrespecttotheimagei.e.,nocorrelationbetweenthepixelvaluesandthevaluesofthenoisecomponents★Commonlyusednoisemodels★Gaussian,Rayleigh,Erlang,exponential,uniform,impulse2013-2014COMP75026GaussianNoise★Gaussiannoiseisfrequentlyuse
