基于暗通道先验和Retinex理论的快速单幅图像去雾方法-刘海波-(1)

417Vol.41,No.720157ACTAAUTOMATICASINICAJuly,2015Retinex1;21131,Retinex.,,,,;,Retinex,,;,,,;,.,,,.,,,Retinex,,,,,.Retinex.,2015,41(7):1264¡1273DOI10.16383/j.aas.2015.c140748AFastSingleImageDehazingMethodBasedonDarkChannelPriorandRetinexTheoryLIUHai-Bo1;2YANGJie1WUZheng-Ping1ZHANGQing-Nian3DENGYong1AbstractImagingintheatmospherepresentsthephenomenonsoflowcontrast,lowsaturationandhueo®setduetoatmosphericparticlessuchashazeandfog.Inthispaper,afastmethodisproposedtoremovehazefromasingleimagebasedondarkchannelpriorandRetinextheory.Basedontheatmosphericscatteringmodel,darkchannelpriorandgray-scaleopeningoperationareusedtoestimatethevalueofglobalatmosphericlightbyaninterval.Meanwhile,theinitialestimationofmediumtransmissionisobtained.Andthewhitebalanceisperformedtosimplifytheatmosphericscatteringmodel.Then,asimpleGaussian¯lterisadoptedtogetthecoarseestimationofmediumtransmissionbasedonRetinextheory,whosegrayvalueisalteredbylinearmapping.Afterthepixelsfusionbetweentheinitialestimationandcoarseestimationofmediumtransmission,afastjointbilateral¯lteringisusedtore¯needge,andadynamicparameterstrategyisadoptedtoimprovethemediumtransmissionoflargeskyregioninfogimage.Finally,thesimpli¯edatmosphericscatteringmodelandtonemappingareusedtogettherestoredimage.Comparedtosomestate-of-the-artmethods,theproposedmethodcanachieveafasterprocessingspeed,e®ectivelyimprovethevisibilityandcontrastoftherestoredimage,andobtaingoodcolore®ect.KeywordsImagedehazing,atmosphericscatteringmodel,darkchannelprior,Retinextheory,imagefusionCitationLiuHai-Bo,YangJie,WuZheng-Ping,ZhangQing-Nian,DengYong.AfastsingleimagedehazingmethodbasedondarkchannelpriorandRetinextheory.ActaAutomaticaSinica,2015,41(7):1264¡12732014-10-282015-03-20ManuscriptreceivedOctober28,2014;acceptedMarch20,2015(51479159),(2013-322-811-470)SupportedbyNationalNaturalScienceFoundationofChina(51479159)andSoftScienceProjectofChina0sMinistryofTransport(2013-322-811-470)RecommendedbyAssociateEditorJIAYun-De1.4300702.4210023.4300701.KeyLaboratoryofFiberOpticSensingTechnologyandIn-formationProcessing,MinistryofEducation,WuhanUniver-sityofTechnology,Wuhan4300702.SchoolofElectricalandInformationEngineering,HunanInstituteofTechnology,Hengyang4210023.SchoolofTransportation,WuhanUni-versityofTechnology,Wuhan430070,.,[1¡2].,[3].,,,:.,,,.Retinex,[4].7:Retinex1265,,,.,,.,[5¡6].Tan[7],.,,,Halo.Fattal[8],.,,.He[9],,,.0,,.Tarel[10],,.,,,Halo.,He[9]Retinex,,Retinex.,,,,,;,Retinex,,;,,,;,,.,.1,[7¡8]I(x)=J(x)t(x)+A(1¡t(x))=A½(x)t(x)+A(1¡t(x))(1),x,I(x),J(x),A,½(x),t(x).,t(x)t(x)=e¡¯d(x);0·t(x)·1(2),¯,d(x).(1)J(x)t(x)[7¡9],,;A(1¡t(x))[10],,.At(x)I(x)J(x),t(x).,,.2Retinex1.,,1Fig.1Algorithm°owchartofimagedehazing126641;,,,Retinex,;,,,.At(x),(1),:Idark=minC2fR;G;BgIC(x)=minC2fR;G;BgA½C(x)t(x)+A(1¡t(x))=½dark(x)At(x)+A(1¡t(x))(3),IC(x)½C(x)I(x)½(x),minC2fR;G;Bg(R;G;B),Idark½darkI(x)½(x).2.12.1.1[11],.[9]0:1%,.,,.,,,.,Idark,(3),:^Idark=^½darkAt0(x)+A(1¡t0(x))(4),^Idark,^½darkt0(x)Idark,½darkt(x).,Idark,^½dark0[9].,(4)^Idark=A(1¡t0(x))(5)(5)t0(x)t0(x)=1¡^IdarkA(6),t0(x)t(x).t0(x)t(x),^Idark=A[0,1].,A^Idark,AA¸max(^Idark)(7),,A.,AmaxµmaxC2fR;G;BgIC(x)¶¸A¸max(^Idark)(8)(8)AA=(1¡®)max(^Idark)+®maxµmaxC2fR;G;BgIC(x)¶;0·®·1(9),®,®=mean(^Idark)(10),mean(¢).®^Idark,A.,,(8),(9).2.1.2WP(Whitepoint),Max-RGB,R,G,B[12].(9),.WPA[13].,(1)I(x)A=½(x)t(x)+1¡t(x)(11),I(x)=A1.,I(x)=A[0,1],I0(x)I0(x)=minµI(x)A;1¶(12),I0(x)=½(x)t(x)+1¡t(x)(13)I0(x),A(1,1,1)T.7:Retinex12672.22.2.1,(13),:I00(x)=minC2fR;G;BgI0C(x)=minC2fR;G;Bg½C(x)t(x)+1¡t(x)=½dark(x)t(x)+1¡t(x)(14),I00(x)I0(x)(R;G;B).(14),:1¡I00(x)=(1¡½dark(x))t(x)(15)Land,(15)1¡I00(x):t(x)1¡½dark(x).t(x),1¡½dark(x).Retinext(x),1¡½dark(x),,[14].,t(x)t00(x)t00(x)=(1¡I00(x))­F(x)(16),­,F(x).,F(x),0.5.hI00(x)m,h=°oor³m50´(17),°oor(¢).t00(x)[0,1],(16)t00(x)=min((1¡I00(x))­F(x);1)(18),t00(x).2.2.2,t00(x),t0(x),t(x).,t0(x),t00(x),t00(x)=imadjust(t00(x);[min(t00(x));max(t00(x))];[td¢min(t0(x));min(tu¢max(t0(x));1)])(19),imadjust(¢)t00(x),tdtu.,tdtu0.951.05.(6)(19),t0(x),t00(x)Retinex,,.,t0(x)t00(x).[15].(),t000(x)=!1¢t0(x)+!2¢t00(x)(20),!1!2,!1+!2=1.Burt[16],.,0.50.5;,,01.t0(x)t00(x),.,t0(x)t00(x),,!1!20.550.45.2.2.3t0(x),t00(x),,t000(x).(3)(6)(20),t000(x),1¡^Idark=A1¡Idark=A,1¡Idark=A.,Paris[17],1¡Idark=At000(x).t000(x)1¡Idark=A,tjb(x)=1WjbxXy2SG¾°µkt000(x)¡µ1¡Idark(y)A¶k¶£G¾s(kx¡yk)µ1¡Idark(y)A¶(21),G¾sG¾°,¾s,¾°.Wjbx,Wjbx=Xy2SG¾°µkt000(x)¡µ1¡Idark(y)A¶k¶£126841G¾s(kx¡yk)(22),tjb(x).2.2.4,^½dark0,t0(x)1¡^Idark=A.,tjb(x),.[1]AIdark,,.,.[18],t0=min³max³nN;0:15´;0:3´(23),Ntjb(x),ntjb(x)0.15,0.150.3t0.t0,t0t0.(2),,t0,t0,t(x)t(x)=min¡max¡tjb(x);t0¢;1¢(24)2.3(13)(24),V(x)V(x)=1¡t(x)=1¡min(max(tjb(x);t0);1)(25),¯,(25)V(x)=1¡¯¢min(max(tjb(x);t0);1)(26),¯0.95.(13),½(x)½(x)=I0(x)¡V(x)t(x)(27),[0,1],O(x),O(x)=min(max(½(x);0);1)(28)2.4,,O(x).[19¡20],.Drago[21].O(x)OC(x),Op(x)=Odmax£0:01lg(OCmax(x)+1)£ln(OC(x)+1)lnÃ2+óOC(x)OCmax(x)´ln(b)ln(0:5)!£8!(29),Op(x)OC(x),Odmax,100[21].b,,0.85[21].OCmax(x)OC(x).2.2(b),,,.,2(c).2(d)2(e),.33.1sx£sy,(21).[17],O(sxsy),.,O(sxsy),,.,Tarel[10]O(sxsys2vlnsv),,sv,,