图像增强算法综述

10420178ChineseOpticsVol．10No．4Aug．20172017-03-072017-04-14No．201204515No．20150101017JCNo．2016201SupportedbyNationalForestryPublicWelfareFoundationofChinaNo．201204515NaturalScienceFounda-tionofJinlinProvinceofChinaNo．20150101017JCYouthInnovationPromotionAssociationCASNo．20162012095-1531201704-0438-1112*121231．1300332．1300333．150040。4、、Retinex。、、。9。。RetinexTP394．1Adoi10．3788/CO．20171004．0438ReviewofimageenhancementalgorithmsWANGHao12*ZHANGYe12SHENHong-hai12ZHANGJing-zhong31．KeyLaboratoryofAirborneOpticalImagingandMeasurementChineseAcademyofSciencesChangchun130033China2．ChangchunInstituteofOpticsFineMechanicsandPhysicsChineseAcademyofSciencesChangchun130033China3．HeilongjiangForestProtectionInstituteHarbin150040China*[e-mail]entalgorithmscanenhancecontrastbetweenthewholeandpartialimagesandhighlightthedetailsofimages．Italsocanmaketheenhancedimagesmoreinlinewiththevisualcharacteris-ticsofthehumaneyesanditappliestomachineidentificationwhichhasawiderangeofapplicationsinmili-taryandcivilianfields．Basedontheprincipleofimageenhancementalgorithmfourtypesofimageenhance-mentalgorithmsandtheirimprovedalgorithmsaresummarizedinthispaper．Thesealgorithmsincludehisto-gramequalizationimageenhancementalgorithmwavelettransformimageenhancementalgorithmpartialdif-ferentialequationimageenhancementalgorithmandRetineximageenhancementalgotithm．Theseimprovedal-gorithmswhichcombinethehumanvisualcharacteristicsnoisesuppressionbrightnesspreservingandinfor-mationentropymaximizationcanfurtherimprovethequalityofimagesinadditiontoenhancingthecontrast．Inthispaperninetypicalimageenhancementalgorithmsareimplementedandtheirenhancementeffectsarecomparedwithsubjectiveandobjectiveevaluationmethods．Theadvantagesanddisadvantagesoftheseen-hancementalgorithmsareanalyzedandthecalculationtimeofthealgorithmsaregiven．Thestudyontheseal-gorithmscanpromotetheimageenhancementtechnologytoahigherlevelsoastomaketheimageenhance-menttechnologyplayanimportantroleinmanyfields．KeywordsimageenhancementhistogramequalizationwavelettransformpartialdifferentialequationRet-inextheory1。。、、1。。1。。2。。3。。4。。。30。HE、、Retinex。HE。HEPDF2-4。HEBBHE5DSIHESEHEHE6-9MMBEBHE7LMHE8WT10-17。kneegamma18。19-20。21。PDE22-27。TotalVaria-tionModelTVPDE934428。2629。Retinex30-35。MRF“”36。ADOFFTRetinex37。38。4。22．12。HECDF3。2．1．1I∈IijLIijijIij∈0L－1Ipk=nkNk=01…L－11Nnkk。Ick=∑ki=0pik=01…L－1．2fk=L－1×ck．3。4。2．1．2BBHE5。BBHEI2ILIUI=IL∪IUIL∩IU=。DSIHE26。DSIHE。MMBE-BHE7。LMHE8Wever-Fech-ner。flgk=lgk×λ－1+1lgL－1×λ－1+14k。4pk。pk。SEHE9。SEHEM×NhkM×N04410Sk=－∑Mm=1∑Nn=1hkmnlog2hkmn5fk=Sk/∑L－1l=0l≠kSl．6fkSEHEFk=∑kl=0fl7。SEHE。2．2WavelettransformWTFourier。。ψxφx10fx=∑kcj0kφj0kx+∑!j=j0∑kdjkψjkx8j0cj0kdjk。2．2．1fxyDWT11。WT13、12-13。。。14-15。1。2。3Wo=Wi+G×T－1Wi＞TG×Wi|Wi|≤TWi－G×T－1Wi＜－T{9GTWiWo。4。。16-17。2．2．2。kneegammaKGWTkneegamma18。KGWT。kneegammapx=xx＜tax3+bx2+cx+dγx≥t{10abcdtγgamma。19。H=－∑PlgP11P。20。144421Curvelettrans-form。2．3PartialdifferentialequationPDE。fxyxy∈Ω=0≤x≤N－10≤y≤M－1{}Ixygx=fxyx=fx+1y－fxy12gy=fxyy=fxy+1－fxy．13p∈ΩVIp=gxgyT。。22-23。24-25。2．3．1VI'pVIpI'IVI'pVIpI'II'I。IPDEI'VI'p=k·VIpp∈Ω14VI'pkk＞1k。14II'p=k·Ip+p∈Ω15。I'pIpk。0～255。15I'p。I'p。15fpfp∈0255p∈Ω16∫∫Ωfp－VI'2dΩ．1626。27。2．3．2TotalvariationmodelTVPDE28TVPDE。E=∑x∈fEd+Eg17EdEg。Ed=λdf－ν218Eg=λgfx－Gx2+fy－Gy2219λdλgffxxfyyνGxGy。TVPDEfp181917E。fp∈0255。。2914S=1－cos‖u‖－ba－b·π()[]·a2·u‖u‖20uab。ba0a。26244100+!|VI'p|=cπ/2·arctank·|VIp|21ck。2．4RetinexRetinexRetinaCor-tex。RetinexLand3031。2．4．1RetinexRetinexRijLij32-33Iij=Rij×Lij22ijIijRijLij。Retinex。34。22RetinexSSRlgRij=lgIij－lgLij．23LijLij=Iij*Fij24*。Fij。Fij=Ke－r2/c225Krc。RetinexRetinexMSR35。MSR。MSRlgRij=∑Kk=1WklgIij－lgLij26WkK。。2．4．2RetinexRetinex。36MRFRetinex。“”。37ADOFFTRetinex。38。33．1HE4、BBHE5、LMHE8、WT11、KGWT18、PDE26、TVPDE28、SSR33MSR359。、。12。HE。BBHE。LMHE344411Fig．1Enhancedimagesobtainedbydifferentalgorithmsandtheircorrespondinghistograms122Fig．2Enhancedimagesobtainedbydifferentalgorithmsandtheircorrespondinghistograms244410。WT。KGWT。PDETVPDE。TVPDE。SSRMSR。MSR。3．2、。Cij=1N∑n=kn=－k∑m=km=－k‖Iij－Ii+mj+n‖27IijN。。R=Gt－Gbσ28GtGbσ。H=－∑Mi=0pklog2pk29pkkM。2、1。1212。1Tab．1ObjectiveevaluationresultsofimagesqualityobtainedbydifferentalgorithmsEvaluationresultOriginalHEBBHELMHEWTKGWTPDETVPDESSRMSR12121212121212121212Contrast8．718．331．724．925．326．828．530．529．033．924．936．133．542．338．128．423．738．027．430．6Signaltonoiseratio7．59．410．312．59．714．113．913．611．410．312．413．923．013．728．316．330．621．532．924．9Informationentropy1．63．03．73．43．43．73．94．03．84．33．64．64．24．74．14．54．85．15．15．01。。、。Reit-nex。3．3VS2010IntelCorei5-3337U1．8GHzCPU4G932。2msTab．1ComputationtimeofdifferentalgorithmsmsResolution/pixel×pixelHEBBHELMHEWTKGWTPDETVPDESSRMSR256×256357182217281330640×5121115349510890116661251024×1024364510930638127731620834354442Retinex。4。、、39-43。、。。。GPUGraphicsProcessingUnit。GPU。GPU。4。。1．J．201694423-431．HAOZCWUCYANGHetal．．Imagedetailenhancementmethodbasedonmulti-scalebilateraltexturefilterJ．Chi-neseOptics201694423-431．inChinese2ZIMMERMANJBPIZERSMSTAABEVetal．．Anevaluationoftheeffectivenessofadaptivehistog