Topic-12_Color_Image_Processing彩色图像处理

EECE-5626:ImageProcessing&PatternRecognitionColorImageProcessingColorImageProcessingTechniquesinPseudo-ColorImageprocessingIntensitySlicingColorLUTDesignsNoiseinColorImagesNoiseinRGBNoiseinHSVSmoothingofColorImagesSharpeningofColorImagesOutlineEECE-5626:ColorImageProcessing1/24Pseudo-ColorImageProcessingAblack-and-whiteimageistransformedintoacolorimageusingpixel-pointprocessing(colorLUTs).Themainapplicationisforhumanvisualizationandinterpretationofgray-leveldetails.Theprincipalchallengeistoselectapaththroughthecolorspacethatcan“amplify”thelowcontrastdetailsintheimage.Techniques:IntensitySlicingColorLUTdesign(Gray-leveltoColorTransformations)EECE-5626:ColorImageProcessing2/24Ifanimageisviewedasa2-Dintensityfunction,thenthisintensitycanbeslicedbyaplaneparalleltothecoordinateplane.Ifweassigndifferentcolorsoneachsideoftheplane,anypixelwithgraylevelabovetheplaneiscodedwithonecolorandanypixelbelowtheplaneiscodedwithanothercolor.Theresultisatwocolorimageasshown.Thisapproachcanbeextendedtomorethantwoplanes.IntensitySlicingEECE-5626:ColorImageProcessing3/24IntensitySlicing(2)ExampleTwo-levelslicing:Inthisexample,thegray-level255isassignedacoloryellowhiletherestofthegray-levels,[0–254],areassignedcolorblue.Thegray-level255signifiesfailureinaweld.EECE-5626:ColorImageProcessing4/24IntensitySlicing(3)Example:Eight-levelslicingEECE-5626:ColorImageProcessing5/24ColorLUTDesignsTheseLook-UpTablesdefineamappingfromthe1-Dgray-levelspacetothe3-DRGBcolorspace.ThismappingisgivenbythreePVMs,oneforeachprimarycolor.Almostallmodernimageprocessingboardscontain(programmable)hardwaretablesbetweenaframebufferandamonitor.EECE-5626:ColorImageProcessing6/24ColorLUTDesigns(2)Thesegray-leveltransformationsessentiallyareuniquepathsfromtheblackintensity(0)tothewhiteintensity(1)insuchawaythattheassignedcolorscanaidinvisualizingandidentifyingimagefeatures.EECE-5626:ColorImageProcessing7/24ColorLUTDesigns(3)Designofthesetransformationsrequiresagoodknowledgeofthecolortheory.Specifically,ifweassigncomplementarycolorstotheadjacentgray-levelsthenwecanvisualizethesegray-levelsbetter.EECE-5626:ColorImageProcessing8/24ColorLUTDesigns(4)Sometypicaltransformationsare:SpectrumorRainbowEECE-5626:ColorImageProcessing9/24ColorLUTDesigns(5)TypicalTransformations(continued)SpectrumorRainbow:ExampleEECE-5626:ColorImageProcessing10/24ColorLUTDesigns(6)TypicalTransformations(continued)SoftColors:Hereeachtransformationissinefunctionofthesamefrequencybutdifferentphase.Thisproducessoftcolorsandisusefulinenhancingbusydetails.EECE-5626:ColorImageProcessing11/24ColorLUTDesigns(7)TypicalTransformations(continued)SoftColors:ExampleEECE-5626:ColorImageProcessing12/24ColorLUTDesigns(8)TypicalTransformations(continued)SoftColors:ExampleEECE-5626:ColorImageProcessing13/24ColorLUTDesigns(9)TypicalTransformations(continued)Bitcolor:Hereeachbitplaneisassignedadifferentcolor.Forexample,ifwerepresentapixelinbinaryasb7b6b5b4b3b2b1b0thenonepossibleschemeis:RRRGGGBBEECE-5626:ColorImageProcessing14/24ColorLUTDesigns(10)TypicalTransformations(continued)Random:Herecolorschemesareassignedinarandomfashion.Usefulforimagesthathaveverysmoothappearance.EECE-5626:ColorImageProcessing15/24ColorLUTDesigns(11)PredefinedColormaps:EECE-5626:ColorImageProcessing16/24NoiseinColorImagesColorimagesareacquired(orformed)viacolorcamerasthatpredominantlyusetheRGBcolormodelandemployCCDarraysensorsforeachR,G,andBcolor.Henceitisreasonabletomodela“noisy”colorimageasbeingformedbythecorresponding“noisy”R,G,andBcomponentimages.ThenoisyimagemodeltheniswhereηisanequivalentRGBcolornoiseimagewhileeachnoisevariable,R,G,orBisanindependentnoisefield.xnoisy=x+hxnoisy,Rxnoisy,Gxnoisy,Béëêêêêùûúúúú=xRxGxBéëêêêêùûúúúú+hRhGhBéëêêêêùûúúúúEECE-5626:ColorImageProcessing17/24NoiseinColorImages(2)Example:NoiseinRGBspaceversusHSVspacef=imread([imagedir,'Fig0604(a)(iris).tif']);%Loadimagefigure;imshow(f);%Showimageg=imnoise(f,'gaussian',0,0.1);%AddNoiseinRGBspacefigure;imshow(g);%Showimagew=rgb2hsv(f);%RGBtoHSVConversionw=imnoise(w,'gaussian',0,0.1);%AddNoiseinHSVspacew=hsv2rgb(w);%HSVtoRGBconversionfigure;imshow(w);%ShowimageOriginalImageNoiseinRGBNoiseinHSVEECE-5626:ColorImageProcessing18/24ColorImageSmoothingSpatialaveragingorsmoothingofmonochromeimagesisaccomplishedbyconvolvingwithamaskwhichissymmetric(toavoidphasedistortionproblems).Theprocessofsmoothingacolorimageisformulatedinasimilarfashion,exceptthatinsteadofsinglepixelswenowhavethreepixelsateachspatiallocationasshownbelow:EECE-5626:ColorImageProcessing19/24ColorImageSmoothing(2)LetS[m,n]denoteaneighborhoodcenteredat(m,n)inacolorimage.ThentheaverageofanRGBimageisgivenbywhereKisthenumberofpixelsintheneighborhood.Thustheaveragingoveraneighborhoodcanbecarriedouteitheronthecolorvectorspaceorontheindividualcomponentbasis.y(n1,n2)=x(k1,k2)K(k1,k2)ÎSn1,n2éëùûå=1KxR(k1,k2)(k1,k2)ÎSn1,n2éëùûåxG(k1,k2)(k1,k2)ÎSn1,n2éëùûåxB(k1,k2)(k1,k2)ÎSn1,n2éëùûåéëêêêêêêêêêêùûúúúúúúúúúúEECE-5626:ColorImageProcessing20/24ColorImageSmoothing(3)Smoot