数字图像处理-冈萨雷斯-课件(英文)Chapter03 空域图像增强

DigitalImageProcessingChapter3:ImageEnhancementintheSpatialDomain15June2007SpatialDomainWhatisspatialdomainThespacewhereallpixelsformanimageInspatialdomainwecanrepresentanimagebyf(x,y)wherexandyarecoordinatesalongxandyaxiswithrespecttoanoriginThereisdualitybetweenSpatialandFrequencyDomainsImagesinthespatialdomainarepicturesinthexyplanewheretheword“distance”ismeaningful.UsingtheFouriertransform,theword“distance”islostbuttheword“frequency”becomesalive.ImageEnhancementImageEnhancementmeansimprovementofimagestobesuitableforspecificapplications.Example:Note:eachimageenhancementtechniquethatissuitableforoneapplicationmaynotbesuitableforotherapplications.(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.ImageEnhancementExampleOriginalimageEnhancedimageusingGammacorrection(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.=ImageenhancementusingprocessesperformedintheSpatialdomainresultinginimagesintheSpatialdomain.WecanwrittenasImageEnhancementintheSpatialDomain(,)(,)gxyTfxywheref(x,y)isanoriginalimage,g(x,y)isanoutputandT[]isafunctiondefinedintheareaaround(x,y)Note:T[]mayhaveoneinputasapixelvalueat(x,y)onlyormultipleinputsaspixelsinneighborsof(x,y)dependingineachfunction.Ex.Contrastenhancementusesapixelvalueat(x,y)onlyforaninputwhilesmoothingfilteuseseveralpixelsaround(x,y)asinputs.TypesofImageEnhancementintheSpatialDomain-Singlepixelmethods-GrayleveltransformationsExample-Historgramequalization-Contraststretching-Arithmetic/logicoperationsExamples-Imagesubtraction-Imageaveraging-MultiplepixelmethodsExamplesSpatialfiltering-Smoothingfilters-SharpeningfiltersGrayLevelTransformationTransformsintensityofanoriginalimageintointensityofanoutputimageusingafunction:()sTrwherer=inputintensityands=outputintensityExample:Contrastenhancement(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.ImageNegativeWhiteBlackInputintensityOutputintensityOriginaldigitalmammogram1sLrL=thenumberofgraylevels0L-1L-1Negativedigitalmammogram(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.BlackWhiteLogTransformationsFourierspectrumLogTr.ofFourierspectrumlog(1)scrApplication(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.Power-LawTransformationsscr(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.Power-LawTransformations:GammaCorrectionApplicationDesiredimageImagedisplayedatMonitorAfterGammacorrection(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.ImagedisplayedatMonitorPower-LawTransformations:GammaCorrectionApplicationMRIImageafterGammaCorrection(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.Power-LawTransformations:GammaCorrectionApplicationArielimagesafterGammaCorrection(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.ContrastStretchingBeforecontrastenhancementAfterContrastmeansthedifferencebetweenthebrightestanddarkestintensities(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.Howtoknowwherethecontrastisenhanced?NoticetheslopeofT(r)-ifSlope1Contrastincreases-ifSlope1Contrastdecrease-ifSlope=1nochangeDrDsSmallerDryieldswiderDs=increasingContrast(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.GrayLevelSlicing(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.Bit-planeSlicingBit7Bit6Bit2Bit1Bit5Bit3(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.HistogramHistogram=Graphofpopulationfrequencies0246810AB+BC+CD+DFNo.ofStudentsGradesofthecourse178xxxHistogramofanImage()kkhrnจำนวนpixelจำนวนpixel=graphofno.ofpixelsvsintensities(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.BrightimagehashistogramontherightDarkimagehashistogramontheleftHistogramofanImage(cont.)lowcontrastimagehasnarrowhistogram(ImagesfromRafaelC.GonzalezandRichardE.Wood,DigitalImageProcessing,2ndEdition.highcontrastimagehaswidehistogramHistogramProcessing=intensitytransformationbasedonhistograminformationtoyielddesiredhistogram-Histogramequalization-HistogrammatchingTomakehistogramdistributeduniformlyTomakehistogramasthedesireMonotonicallyIncreasingFunction=Functionthatisonlyincreasingorconstant)(rTsPropertiesofHistogramprocessingfunction1.Monotonicallyincreasingfunction2.10for1)(0rrTProbabilityDensityFunctionandrelationbetweensandrisHistogramisanalogoustoProbabilityDensityFunction(PDF)whichrepresentdensityofpopulationLetsandrbeRandomvariableswithPDFps(s)andpr(r)respectively)(rTsWegetdsdrrpsprs)()(rrdwwprTs0)()(HistogramEqualizationLetWeget1)(1)()(1)(1)()()(0rprpdrdwwpdrpdrdsrpdsdrrpsprrrrrrrs!HistogramEqualizationFormulainthepreviousslideisforacontinuousPDFForHistogramofDigitalImage,weusekjjkjjrkkNnrprTs00)()(nj=thenumberofpixelswithintensity=jN=thenumberoftotalpixelsHistogramEqualizationExampleIntensity#pixels0201522531041555610710Tot