面向无参考图像的清晰度评价方法研究

:2010-09-27;:2010-12-22::(1985-),,,,E-mail:[e-mail]李祚林1,2,李晓辉1,马灵玲1,胡玥3,唐伶俐1(1.中国科学院光电研究院,北京100190;2.中国科学院研究生院,北京100049;3.首都经济贸易大学信息学院,北京100026):图像的清晰度是衡量数字图像质量优劣的重要指标基于面向无参考图像质量评价,重点介绍了目前较为常用的具有代表性的几种清晰度评价算法,并从算法的单峰性无偏性灵敏性,与图像尺度内容的无关性,以及与主观感知的一致性3个方面的性能进行了对比和分析,以求准确客观地评价各清晰度算法,达到在实际应用中合理有效地评价图像清晰度的目的:图像质量;清晰度;性能评价:TP731:A:1004-0323(2011)02-0239-081引言,:[1],;,,,,,,3[1-2]:(Ful-lRef-erenceQA),;(Reduced-ReferenceQA),,;(No-ReferenceQA),,,,,,,,,,,,;,,;,,,,,,,2数字图像清晰度评价方法,,26220114REMOTESENSINGTECHNOLOGYANDAPPLICATIONVol.26No.2Apr.20112.1,,,,,[3]:(1)Brenne[4]Brenne,,,,Brenne:D(I)=xy[I(x+2,y)-I(x,y)]2(1):I(x,y)I(x,y),D(I)()(2)TenengradLaplacianKrotkovTenengrad[5-7],SobelTenengrad:D(I)=xy[S(x,y)]2(S(x,y)T)(2):T,:S(x,y)=I2x(x,y)+I2y(x,y)(3)SchlagSandersonSobel:ix=14-101-202-101,iy=14121000-1-2-1(4)Laplacian[7]Tenengrad,Laplacian,:L=161414-204141(5):[8]Prewitt;[9]Sobel,(3)Jarvis[10],(SMD),:D(I)=1n[|I(x,y)-I(x,y-1)|+|I(x,y)-I(x+1,y)|](6):n[11],[12],,2.2,,,,FFT,:D(I)=1MN(real2+imaginary2)tan-1(imaginary/red)(7):MN,realimaginary,D(I)2.3,,[10,13],,:D(I)=-xyI(x,y)ln[I(x,y)](8):D(I)3,,,,,2.4,,(HVS),,HVS,,240262.4.1基于点锐度的清晰度算法(EAV)[14-15],,:D(I)=ba(dI/dx)2|I(b)-I(a)|(9):dI/dx,I(b)-I(a)[16],8,,,(EAV):EAV=MNi=18a=1|dI/dx|MN(10):EAV,MN,dI,dx,dI/dx8,()1,451351/22.4.2基于二次模糊的清晰度算法(ReBlur)Crete[17],,,;,,,,,,,(Re-Blur),12.4.3基于梯度结构相似度的清晰度算法(NRSS)Wang[18-20](HVS),(SSIM),,[21],,,3:1(ReBlur)Fig.1FlowchartofReBluralgorithm:l(x,y)=2xy+C2x+2y+C1:c(x,y)=2xy+C22x+2y+C2:s(x,y)=xy+C3xy+C3:SSIM(x,y)=[l(x,y)][c(x,y)][s(x,y)](11)===1[22],,(NRSS),:(1)I,Ir=LPF(I),IIr(2)IIr,Sobel,IIrGGr(3)GNGN,,GN,GrN(4)NRSSN,INRSS:NRSS=1-1NNi=1SSIM(xi,yi)(12)[21],,2412:,,[23],3BPBP,3算法比较3[24]:,;;,,,,,,,,3:;;3.11,,(2),,33,3,,(ReBlur);,;,,,,()2Fig.2Imagesofdifferentfocallengths3Fig.3Definitionresultsfromdifferentalgorithms3.22,Lena[25](4),14LenaFig.4DifferentsizesofLenaimages1,Lena,:4,,;ReBlur;NRSS,,,,512512,NRSS242261LenaTable1DefinitionresultsfromdifferentsizesofLenaimages12812825625651251276876810241024Tenengrad2493.2001720.200984.900748.780547.000Laplacian88.22085.70069.49066.22057.000SMD22.11016.55011.3208.8666.935EAV87.21063.52042.96035.07027.420ReBlur0.0400.0400.0370.0430.048NRSS0.6800.6400.5000.4700.450,10241024Lenaabcd4(5),4,25Fig.5Differentregionsofthesameimage2LenaTable2DefinitionresultsfromdifferentcontentsofLenaimageabcdTenengrad389.310560.070713.520440.050Laplacian53.07056.16057.38059.220SMD6.7205.8609.7105.420EAV26.32022.88038.63021.790ReBlur0.0360.0520.0520.051NRSS0.3800.2900.3000.2402,Tenengrad,5,,LaplacianLaplacian,,,4,,0~1,3.33,ReBlurNRSS,,,[26-28]29,JPEGJPEG2000Fastfading(Fast-fading)WhiteNoise()5779,145DMOS(DifferenceMeanOpinionScores),DMOSMOS(MeanOpinionScores)100,DMOS,DMOS,(VQEG)[29]145,,145DMOS;,(VQR),(Logisitic)DMOS2432:VQR[29-31],DMOSp;DMOSpDMOS(CC)(MAE)(RMS)3,:DMOSp=b11+exp(-b2(VQR-b3))(13):b1b2b3Logisitic,29,(SSIM)[20]ReBlurNRSSSSIM36:6Fig.6Fittingcurvesofdifferentevaluationmethods3,380%,,SSIM,SSIM,Re-BlurNRSS,,NRSSReBlur,3Table3PerformanceofdifferentalgorithmsCCMAERMSReBlur0.83067.54508.7852NRSS0.85186.30448.2655SSIM0.87375.76837.67594结语,,,,,,,,ReBlurNRSS,,ReBlurNRSS,NRSSReBlur,,NRSS(References):[1]XiaoWen.SpatialNo-referenceImageQualityAssessment[D].Nanchang:JiangxiUniversityofFinanceandEconomics,2006.[.[D].:,2006.][2]WangZ,SheikhHR,AlanCB.ObjectiveVideoQualityAs-sessment[C]//TheHandbookofVideoDatabases:DesignandApplications.Florida:CRCPress,2003,1041-1078.[3]NgKC,NathanielP,AunN,etal.PracticalIssuesinPixe-lbasedAuto-focusingforMachineVision[C]//Proceedingsofthe2001IEEE.InternationalConferenceonRobotics&Au-tomation,Seoul,KoreaMay21-26,2001:2791-2796.[4]SubbaraoM,TyanJK.SelectiontheOptimalFocusMeasureforAuto-focusingandDepthfromFocus[J].IEEETransac-tionsonPatternAnalysisandMachineIntelligence,1998,2024426(8):864-870.[5]SchlagJF,SandersonAC,Neuman,CP,etal.Implementa-tionofAutomaticFocusingAlgorithmsforaComputerVisionSystemwithCameraControl[R].TechnicalReportCMU-RI-TR-83-14,CarnegieMellonUniversity,1983.[6]TenenbaumJM.AccommodationinComputerVision[D].Ca-lifornia:StanfordUniversity,1970.[7]KrotkovEP.ActiveComputerVisionbyCooperativeFocusandStereo[M].Springer-Verlag,1989.[8]SunYue,LuanXiaoming,ZhaoFang.AnImprovedImage-charityEvaluationFunction[J].AppliedScienceandTechnol-ogy,2009,36(9):52-55.[,,.[J].,2009,36(9):52-55.][9]JiangTing,TanYuegang,LiuQuan.ResearchofaClarity-E-valuationFunctionofImagebasedonSobel[J].Computer&DigitalEngineering,2008,36(8):129-131.[,,.SOBEL[J].,2008,36(8):129-131.][10]JarvisRA.FocusOptimizationCriteriaforComputerImageProcessing[J].Microscope,1976,24(2):163-180.[11]WuLiming,TaoXiaojie.ANewKindofDefinitionEvaluationFunctionofImage[J].InstrumentationCustomer,2008,15(6):84-86.[,.[J].,2008,15(6):84-86.][12]WangYong,TanYihua,TianJinwen.ANewKindofSharp-ness-evaluation-functionofImage[J].JournalofWuhanUn-iversityofTechnology,2007,29(3):124-126.[,,.[J].,2007,29(3):124-126.][13]HornBKP.Focusing[R].TechnicalReportAIM-160,Mas-sachusettsInstituteofTechnology,May,1968.[14]XuGuili,LiuXiaoxia,TianYupeng,etal.ImageCharit