[2011_PAMI]_Single_Image_Haze_Removal_Using_Dark_Channel_Prior

《[2011_PAMI]_Single_Image_Haze_Removal_Using_Dark_Channel_Prior》由會員上傳分享，免費在線閱讀，更多相關內容在學術論文-天天文庫。

1、IEEETRANSACTIONSONPATTERNANALYSISANDMACHINEINTELLIGENCE,VOL.33,NO.12,DECEMBER20112341SingleImageHazeRemovalUsingDarkChannelPriorKaimingHe,JianSun,andXiaoouTang,Fellow,IEEEAbstract—Inthispaper,weproposeasimplebuteffectiveimageprior—darkchannelpriortoremovehazefromasingleinputimage.Thed

2、arkchannelpriorisakindofstatisticsofoutdoorhaze-freeimages.Itisbasedonakeyobservation—mostlocalpatchesinoutdoorhaze-freeimagescontainsomepixelswhoseintensityisverylowinatleastonecolorchannel.Usingthispriorwiththehazeimagingmodel,wecandirectlyestimatethethicknessofthehazeandrecoverahig

3、h-qualityhaze-freeimage.Resultsonavarietyofhazyimagesdemonstratethepoweroftheproposedprior.Moreover,ahigh-qualitydepthmapcanalsobeobtainedasabyproductofhazeremoval.IndexTerms—Dehaze,defog,imagerestoration,depthestimation.?1INTRODUCTIONMAGESofoutdoorscenesareusuallydegradedbytheediting

4、.HazeorfogcanbeausefuldepthclueforsceneIturbidmedium(e.g.,particlesandwaterdroplets)inunderstanding.Abadhazyimagecanbeputtogooduse.theatmosphere.Haze,fog,andsmokearesuchphenomenaHowever,hazeremovalisachallengingproblembecauseduetoatmosphericabsorptionandscattering.Theirradiancethehaze

5、isdependentontheunknowndepth.Theproblemisreceivedbythecamerafromthescenepointisattenuatedunderconstrainediftheinputisonlyasinglehazyimage.alongthelineofsight.Furthermore,theincominglightisTherefore,manymethodshavebeenproposedbyusingblendedwiththeairlight[1]—ambientlightreflectedintoth

6、emultipleimagesoradditionalinformation.Polarization-lineofsightbyatmosphericparticles.Thedegradedimagesbasedmethods[3],[4]removethehazeeffectthroughtwoorlosecontrastandcolorfidelity,asshowninFig.1a.Sincethemoreimagestakenwithdifferentdegreesofpolarization.Inamountofscatteringdependson

7、thedistanceofthescene[5],[6],[7],moreconstraintsareobtainedfrommultiplepointsfromthecamera,thedegradationisspatiallyvariant.imagesofthesamesceneunderdifferentweatherconditions.1Hazeremoval(ordehazing)ishighlydesiredinDepth-basedmethods[8],[9]requiresomedepthinforma-consumer/computatio

8、nalph