图像复原—去雾总结报告.pptVIP

Defog/Dehaze----王永虎ThepaperSingleImageHazeRemovalUsingDarkChannelPrior;Visibilityinbadweatherfromasingleimage;Bayesian?Defogging;FastVisibilityRestorationfromaSingleColororGrayLevelImage;SingleImageHazeRemovalUsing

DarkChannelPriorAuthor:KaimingHe;JianSun;XiaoouTangFrom:TheChineseUniversityofHongKongMicrosoftResearchAsiaPublish:IEEEConferenceonComputerVisionandPatternRecognition(CVPR),2009DCP:DarkChannelPrior（暗原色先验）akindofstatisticsofthehaze-freeoutdoorimages.Itisbasedonakeyobservation-mostlocalpatchesinhaze-freeoutdoorimagescontainsomepixelswhichhaveverylowintensitiesinatleastonecolorchannel.大气物理模型利用DCP先验,能够快速估计出大气传输函数t,进而求解清晰图像J。算法非常简单,去雾效果很理想。很多算法都是基于DCP的改进算法。平滑大气传输函数采用softmatting的思路,效果较好但是复杂度过大。对于没有大气光,或者景物在本质上同空气层接近并且没有阴影覆盖其上时,暗原色先验效果不好。处理一幅600*400的图像需要2s（C++）,远远达不到实时的标准。Visibilityinbadweatherfrom

asingleimageAuthor:RobbyT.TanFrom:ImperialCollegeLondonPublish:CVPR2008Thismethodisbasedontwobasicobservations:First,imageswithenhancedvisibility(orclear-dayimages)havemorecontrastthanimagesplaguedbybadweather;Second,airlightwhosevariationmainlydependsonthedistanceofobjectstotheviewer,tendstobesmooth.Tan等人通过统计研究发现,无雾图像相对于有雾图像必定具有较高的对比度,通过最大化局部对比度来消除雾。假设局部区域环境光为常数。在马尔科夫（MRF）模型框架下,构造关于边缘强度的代价函数,使用图分割理论来估计最优光照。该算法旨在增强图像的对比度。尽管明显地改善了图像的可见性,然而由于没有从物理模型上恢复真实场景反照率,所以回复后的图像的颜色显得过于饱和,且在景深突变的边界产生严重的Halo效应。Bayesian?Defogging

Author:KoNishino;LouisKratz;StephenLombardiFrom:DrexelUniversityPublish:IntJComputVis通过利用雾图像潜在的统计结构,利用贝叶斯概率方法联合的从单一图像估计出场景反射率和景深。本方法的关键是讲场景反射率和景深作为阶乘马尔科夫穗机场的两个隐含层。利用期望最大化算法和图分割理论来估计出它们的值。FastVisibilityRestorationfromaSingleColororGrayLevelImageAuthor:Jean-philippeTarel;NicolashautiereFrom:LEPSISFrancePublish:ICCV2009本方法可以用于彩色和灰度有雾图像的能见度提高。最大的优点是速度快。复杂度是图像像素数的线