一种图像感知质量评价模型

1912200712JOURNALOFCOMPUTER2AIDEDDESIGN&COMPUTERGRAPHICSVol119,No112Dec1,2007:2007-03-20;:2007-07-241:(60672072);(Y106505);(2005B100016)1,,1981,,1,,1972,,,(li1junli@vip11631com)1,,1963,,,11)2)2)1)(310027)2)(315211)(icerosew@1631com),,13;3,;,PearsonSpearman,01966019491,1;;Pearson;SpearmanTN919181APerceptionBasedImageQualityAssessmentModelWeiXuehui1)LiJunli2)ChenGang2)1)(DepartmentofMathematics,ZhejiangUniversity,Hangzhou310027)2)(SchoolofInformationScienceandEngineering,NingboUniversity,Ningbo315211)AbstractThreeimportantpiecesofinformationaredrawnoutfromtheprocessofsubjectiveimagequali2tyestimation1Foreachoftheinformation,amodelisestablishedandalsoafunctionisfoundouttosimulateitsworkonthehumanvisualsystem1Thenamulti2linearregressionanalysisisusedtointegratethesefunc2tionsintoanequationandusedasourperceptionbasedimagequalityassessmentmodel1Experimentalresultsshowthatourestimationcloselyapproximateshumansubjectivetestssuchasmeanopinionscore,withhighPearsonandSpearmancorrelationcoefficientsof01966and019491Itisasignificantimprovementoversometypicalobjectiveimagequalityestimationssuchaspeaksignal2to2noiseratio1Keywordsimagequality;humanvisualsystem;Pearsoncorrelationcoefficient;Spearmancorrelationco2efficient,,1()12:,1,(meanopinionscore,MOS)1(MOS),(,),,1,,MSEPSNR11(Coekonowski,QualityIndex[1]PrattEdge),(humanvisualsystem,HVS)1,1HVS,(contrastsensitivityfunction,CSF),1DCTCSFHVSAN[2]HVSL2N[3];HVS[4][5]CSF1[627]1HVS;HVS,PBAM(perceptionbasedimagequalityassessmentmodel);MOSPearsonSpearman,11,,,11HVS,,1,HVS1HVS:HVS;;,1,CSF12,,1,HVS,1,31,3,3,1Q,Q1,Q2Q33,f1,f2f3HVS3;QQ=a+a1f1(Q1)+a2f2(Q2)+a3f3(Q3),a,a1,a2a312211Web-FechnerHVS,f(x,y)S(x,y),S(x,y)=kf(x,y)f(x,y)1S(x,y)=kln(f(x,y))+k0=(klgf(x,y))+k0,,k=kln10,k0,k1,k,19264,k=015;k=11,f1(x,y)f2(x,y)Q=(S1(x,y)-S2(x,y))2MN=k1lg(f1(x,y))-k2(lg(f2(x,y)))2MN,MN1212CSFHVS1CSF,CSF,11,NiganCSFH(r)=(0131+0169r)exp(-0129r),,r=R2f+C2f,RfCff(x,y)1CSF11145112:12,r5,,CSFH(r)h(r):1)1r[0,015),,,1h(r)=1-H(r),A12)1r[015,1),,,1h(r)=H(r),B13)1r[1,6),,1h(r)=1,C14)1r[6,14],,,1h(r)=H(r),D15)1r14,,1h(r)=0,E1f1(x,y)f2(x,y)Q2=(f1(x,y)h1(r)-f2(x,y)h2(r))2MN12131f(x,y)f=GxGy=5fP5x5fP5y,mag(f)=|G2x+G2y|1P2=|(5fP5x)2+(5fP5y)2|1P211,;,1,E1(x,y)E2(x,y)Q3=E1(x,y)-E2(x,y)2MN12143,500(CCIR500)[6],MOS1MOSY:,ikA(i,k),kMOSYk=1nni=1A(i,k),n128,,JPEG,,15[829]124512007Q1Q2Q3,3MOS135,HVSf1,f2f3135[0,1],f1(x)=5104-3152x-11108x2+10149x3,f2(x)=5103-0132x-6153x2+2177x3,f3(x)=3163-11110x-71169x2+78163x315343,,f114,,f2151:HVS,HVS,;,,,13,,Q=-01620+01643f1(Q1)+01485f2(Q2)+0105f3(Q3)1,Q,f1,f2f3,31,,HVS,,1,,1,Q=-01459+01653f1(Q1)+01486f2(Q2)1,MOS1,,,PBAM136,PBAMMOS1MOS,1Pearson;Spearman,MOS345112:1PearsonSpearman1PearsonSpearman[-1,1],21,,1PBAMPearsonSpearman,PSNR621:1)1(meanabsoluteerror,MAE),(meansquareerror,MSE)12)1(SNR),(PSNR)13)1Coekonowski(Co2ekonowskicorrelation,CC),(qualityin2dex,QI)14)1Pratt(Prattedgemeasure,PEM)15)116)HVS1DCTCSFHVS(HVSAN)HVSL2(HVSL2N)16PBAMMOS7,SNRPSNRPearson11,0187001853,,PBAMPearson01966,SNRPSNR1SpearmanHVSHVSL2,7,Spearman0195501957,PBAMSpearman01949,111,#171,PearsonSpearman,PBAM2171PearsonSpearmanPearsonSpearmanRMES-01762-01861QI-01034#01609MAE-01745-01846PEM-01439-01442MIN-01848-01900SME-01752-01753MRE-01631-01916SPE-01456-01578MSE-01597-01861WSE-01790-01760MSS2NR0154701783BSME-01782-01754SNR0187001863BSPE-01428-01573PSNR0185301869BWSE-01737-01709NCC01137#-01141#HVSAN-01644-01955IF-012793#-01060#HVSL2N-01683-01957CC-01698-01704PBAM01966019494,3HVS13,HVS31313,,,1HVS44512007PBAM1PBAM,MOSPearsonSpearman1,PBAMPearsonSpearman,0196601949;21,1[1]WangZhou,BovikAlanC1Auniversalimagequalityindex[J]1IEEESignalProcessingLetters,2002,9(3):77-80[2]EskiciogluAM,FisherPS1Imagequalitymeasuresandtheirperformance[J]1IEEETransactionsonCommunications,1995,43(12):2959-2965[3]AvcibasI,SankurB,SayoodK1Statisticalevaluationofimagequalitymeasures[J]1JournalofElectronicImaging,2002,11(2):206-223[4]WeiChongkui,ChengLizhi1Animagequalitymeasureschemeintheperceptualfieldviamasking[J]1JournalofImageandGraphics:A,2004,9(2):195-200(inChinese)(,1[J]1:A,2004,9(2):195-200)[5]DingXuxing,ZhuRihong,LiJianxin1AcriterionofimagequalityassessmentbasedonpropertyofHVS[J]1JournalofImageandGraphics:A,2004,9(2):190-194(inChinese)(,,1[J]1:A,2004,9(2):190-194)[6]LiJunli,ChenGang,ChiZheru,etal1Imagecodingqualityassessmentusingfuzzyintegralswithathree2componentimagemodel[J]1IEEETransactionsonFuzzySystems,2004,12(1):99-106[7]LiJunli,ChenGang,ManJiaju1Fuzzyintegralbasedobjectiveassessmentofcodingqualityofcolorimage[J]1JournalofCom2puter2AidedDesign&ComputerGraphics,2005,17(8):1823-1827(inChinese)(,,1[J]1,2005,17(8):1823-1827)[8]LiJunli,ChenGang,ChiZheru1Afuzzyimagemetricwithap2plicationtofractalcoding[J]1IEEETransactionsonImageProcessing,2002,11(6):636-643[9]WeiXuehui,LiJunli,ChenGang1AnimagequalityestimationmodelbasedonHVS[C]PPProceedingsofIEEETENCON2006,HongKong,2006:283-286545112: