基于单幅图像的雾天降质图像去雾方法分析-analysis of defogging method for fog - degraded images based on single image.docxVIP

AbstractTitle:RESEARCHONDEHAZEMETHODFORFOG-DEGRADEDIMAGESBASEDONSINGLEIMAGEMajor:ComputerApplicationTechnology.Name: HuandiYangSignature:问阳Ji(().-Supervisor: Associateprof.ZhenghaoShiAbstractSignature:Inthehazy weather，theimagetaken by outdoorvisual systems isdegraded duetoatmosphericscattering.Theseimagesarefeaturedbylowcontrast，colordistortionanddetailsblurring，etc.Thesecharactersleadtosomeworksisdifficulttodo，suchasfeatureextraction，targetidentification，targettracking，etc.Therefore，theresearchonhazeremovalhasimportantpractical significance.Thispaperfocusesonthedehazemethod withasingleimage.Twonewmethodsareproposed，andexperimentalresultsverifiedthevalidi句rofproposedmethods.Main contributionsofthepaperaresummarizedupasfollows:Firstly，based onthedegradation process andcharacteristic ofhaze images，thehazeremovalmethodusingdarkchannelpriorisdiscussed，experimentalresultsshowthatthealgorithmcanefficientlyrecoverthesceneoftheimage，butitdoesnotalwaysproducegoodresults，especiaIIywhenthesceneobjectisinherentlysimilartotheair lightoveralargelocalregion.Aimatsuchdrawbacks，thispaperproposed animprovedmethod，whichtermed astwo-stagedehazingmethod.Theproposedmethodrevisethecontrastofdehazeresultsofdarkchannelpriorbylocalhistogramequalization.Experimentalresultsdemonstratethattheproposedmethodcaneffectivelyovercometheshortcomingsofdarkchannelprior.Besides，theproposedmethodcanimprovethecontrastandsharpnessofhazeimages，andobtainagoodvisualeffectSecondly，based onakey observation -mostlocal patches inhaze-freeoutdoorHSIimagescontainsomepixelswhichhaveverylowintensitiesinitsluminancecomponent，anewimagepriortermedas thedarkpriorisfirstlypresented.Usingthispriorwiththetraditionalhazeimagingmodel，wecandirectlyestimatethethicknessofthehazefromtheluminancecomponentonly.Togetatransmissionmapwithac1earedgeoutlineanddepthlayerofthesceneobjects，bilateralfilterisemployed.ExperimentalresuIts demonstratethattheproposedmethodcanunveilthedetailsandrecovervividcolorseveninheavilyhazyregions,and pro