Crawly Crypt Collection 1

home *** CD-ROM | disk | FTP | other *** search

/ Crawly Crypt Collection 1 / crawlyvol1.bin / graphics / aim_xtra / aimtutor.doc < prev next >

Wrap

Text File | 1988-04-07 | 46KB | 957 lines

06600020202800 1AIM TUTORIALPage # ofFile: AIMTUTOR.DOC 2 F0110030 9[...................................................]011 ü ëAIMTUTORIAL 9[...................................................]011 Ç by TomHoeksmaandAdHerweijer DelftUniversityofTechnology FacultyofAppliedPhysics PatternRecognitionGroup Lorentzweg1 2628CJDELFT theNetherlands ëIntroductionê Ç TheimageprocessingsoftwarepackageAIM(AtariImageManager) hasbeendevelopedatDelftUniversityofTechnologybyFrans Groen(the'imageprocessor')andwasadaptedfortheAtariST computerincooperationwithRobertdeVries(theGEMprogram⑨ mer).BothweremembersoftheSignal-/System-techniquesGroupin theFacultyofAppliedPhysics,Fransasaseniorstaffmember andRobertasagraduatestudent.Recently,Franschangedjobsto becomeprofessorofAppliedInformaticsintheUniversityof Amsterdam. AIMisdesignedtobeanentrylevel'appetizer'fortheartof imageprocessing.Itgivestheinterestedlaypersonan opportunitytoacquainthimselfwiththetechniquesand terminologyusedinthisdynamicfield.Theexperiments describedmayberegardedasalowerboundofwhatcanbe accomplishedwithAIM,theupperboundbeinglimitedbythe user'simaginationonly. Thistutorialisnotafullcourseondigitalimageprocessing. SuchcoursesarefrequentlyorganizedbythePatternRecognition Group.Forcourseinformation,pleasecontacttheSecretaryof thePatternRecognitionGroup,FacultyofAppliedPhysics, Lorentzweg1,2628CJDelft,theNetherlands,Phone015-781416. Notethattheoperationsusedintheexamplesanddemoscovera minorpartofAIM'scapabilitiesonly.Acomprehensivedescrip⑨ tionofalltheoperationsavailableisgiveninafilecalled AIMANUAL.DOC.Likethistutorialitiscontainedinacompressed file(DOCS_IMS.ARC),whichcanbefoundinthefolderCOMPRESSon thedistributiondisk(togetherwithadecompressingprogram). Imageprocessingisnotparticularlydifficult;manypeoplewill havehadtheirfirstimageprocessingexperienceinthe photographicdarkroom.Note,however,thatAIM,beingdedicated todigitalimageprocessing,requiressomebasicknowledgeof digitalsystems.Sothoseofyouwhountilnowhavesuccessfully avoidedlearningaboutbits,bytesandbinarynumbers,be preparedtofacetheunavoidable!The15minutesorsospenton readingabouttheseconceptswillberichlyrewardedbyabetter understandingofAIM'soperations,ifnotbythedeepadmiration ofyourenvironmentforthenewlyacquiredvocabulary. AIMisaresultofmanyyearsofresearchanddevelopmentinthe PatternRecognitionGroup.Manyoftheimageprocessingroutines werecontributedby(a.o.)GertvanAntwerpen,FransGroen,Piet Verbeek,BenVerwer,LucasvanVlietandIgorWeber.This introductionwaswritten(ifnotcompiledfromexistingtexts)by TomHoeksmaandAdHerweijer. PreviousversionsofAIMhavealreadyextensivelypenetratedthe AtariSTcommunity.However,asthedocumentationsuppliedwith thosefirstpublicdomainversionswasratherlimited,the explorationofthesoftwarehasbeenamere'imageprocessing adventure'tomostofthe'earlybirds'.Theauthorshopethat thisintroduction,togetherwiththerelatedexamplesanddemos includedonthedisk,willenableallprospectiveusers(early birdsandnewcomersalike)tobetterunderstandwhatwas,resp. isgoingon. Ç Feedbackplaysanimportantroleintheeffectiveuseofimage processingaswellasinsoftwaredevelopment.Therefore,we inviteyoutosendyoureventualcommentsandsuggestionsto: Prof.dr.ir.F.C.A.Groen c/oDelftUniversityofTechnology FacultyofAppliedPhysics PatternRecognitionGroup Lorentzweg1,2628CJDELFT theNetherlands ë ëWHATCANWEEXPECTOFIMAGEPROCESSING?Ç 9[...................................................]011 MaybeyouhavealreadyhadsomeexperiencewithAIMorseen somebodyelseworkwithit.Youmayhavewondered,then,how imageprocessingcouldmakefeaturesappearinthedisplaythat weren'ttherebeforetheoperation.Textthatatfirstwas unreadable,suddenlycouldbereadeasily;frecklesinagirl's face,thatdidn'tshowintheoriginaldisplay,could(toher distress!)becountedaftersomeappropriatekeystrokes.Youwere perhapsconvincedtohaveacquiredalittlemagician,hiddenin thattinyfloppydisk. Bewarned,however!Imageprocessingcanneverreproducefeatures oftheoriginalimage,thatsomehowhavebeenlost.So,be preparedforsurprises,butdon'texpectmiracles!Image processingcanenhanceparticulardetailsinanimage,but usuallyotherdetailswillhavetobesacrificed. If'Apictureisworthathousandwords',imageprocessingallows youtoarrangethosewordstomakethepicturerevealits (secret)'stories'.Butremember:onepicture'sthousandwords aredifferentfromanother's.Everystorywillinheritthe characteristicsofthepictureitwasextractedfrom.Theartof imageprocessingis:tomakethepicturetelltheonestory that'susefultoyou. ë STAGESINANIMAGEPROCESSINGAPPLICATIONÇ Inalmosteveryimageprocessingapplicationwecanrecognize somestages,thateventuallyleadtothefinalresult:ürecordingÇ, üpreprocessingÇ,üsegmentationÇ,üpostprocessingÇ,üanalysisÇandüinter⑨Ç üpretationÇ. Probablythemostimportantstageistherecordingofthe originalpicture.Asstatedabove,informationthatgetslostin therecordingprocesscanneverberecovered.Fortherecordinga video-cameraisrequired,capableofconvertingthelight, reflectedfromtheobject,intoanelectricalsignal.Asthe videosignalcannotbereaddirectlybythecomputer,thissignal mustbesampledanddigitizedandthedigitalinformationmustbe transferredinanorderlymannertothecomputer'smemory.This isdonebyaspecialpieceofhardware,aso-called'üframe grabberÇ'.VariousframegrabbersareavailablefortheAtariST. InthePatternRecognitionGroupmonochromeandcolorframe grabbershavebeendevelopedthateventuallywillbemade commerciallyavailable. Aftertheimageisstoredinthecomputeritmustbepre⑨ processed.Heretheusertriestoenhanceorfilterout particularwantedresp.unwantedfeatures,tosatisfy requirementsoffurtheranalysis.Examplesofpreprocessingare thecorrectionforchanginggreyvaluesinthebackground (shading),suppressingnoise,determinationofedgesinthe image,etc. ÇInordertobeabletodeterminepropertiesofsingleobjectsin theimage,theseobjectsmustfirstbemadedetectable.Ifagrey valueimagecontainsdarkobjectsonalightbackground, segmentationcan,forinstance,resultinabinaryimagewith whiteobjectsinablackbackground. Postprocessingcanimprovetheresultofthesegmentation. Examplesarethemorfological(binary)operations.Holesin objectscanbeclosed,edgespeeledoff,etc. Intheanalysisstage,measurementscanbeperformedonthe segmentedobjects.Examples:area,perimeter,curvature. Intheinterpretationstagetheresultsofthemeasurementsare evaluated. An'intelligentrobot'inafactory,thathastomakeadecision onwhetherhe'sees'anutorabolt,mustbeprogrammedtogo throughallstagesmentionedhere.Programmingsucharobot requiresmuchintuitiononhowcomputers'think'.Animportant requirementfortheprogrammeristobeabletoforgetthewayhe woulddothejobhimself!Hemustbelongtothenewbreedof 'computerpsychologists'.ü ë SOMETERMINOLOGYUSEDINIMAGEPROCESSING 9[...................................................]011 Ç Imagesanddisplaysê ÇInimageprocessingwemustdistinguishbetween'images'and their'displays'.InthiscontextwedefineanüimageÇasthe informationthathasbeenstored(as1'sand0's)witha particularfilenameinthecomputer'smemoryoronamagnetic disk.AüdisplayÇistherepresentationofsuchanimageonavideo screenor-asa'hardcopy'-onapieceofpaper. AnimportantpropertyofanimageisitsüinformationÇcontent.The informationinanimagedependsonthepresenceofdetails (lines,edges,intensities,colors,etc.).Moreover,the informationcontentdeterminestheamountofcomputermemory requiredtostoretheimage. Notethatit'sonethingtostoreanimageaccurately,butthat it'sanothertodisplayitwithfullconservationofinformation! Thedisplayhardware(videocontroller,monitor,printer,etc.) determinesthedegreetowhichtheinformationinanimagecanbe observedexactly.Inimageprocessing,thereisatendencyto maketheinformationcontentofstoredimageshigherthancan actuallybedisplayed.Thisreducesthechancethattheeffects ofroundingerrorsinintermediateresultswilldeterioratethe finaldisplay.(Moreover,RAM-chipsandfloppydisksaremuch cheaperthandisplayhardware!).AIM,forexample,uses64 kilobytes(kB)ofRAMforthestorageofimages,whereasthe Atarihardwarecanonlyhandle32kBforthedisplayofone screen(eithermonochromeorcolor). ÇThebasicelementofanimageisaüpixelÇ(pictureelement).Tobe observableonadisplay,apixelmustbeattributedaügreyvalue ÇorücolorÇinformation.  Asanexample,considertheAtariST.Inthemonochrome(high resolution)modetheAtariSTsupportsthedisplayof2grey valuesonly.Eachpixel(theAtariscreenhas640*400=256,000 ofthem!)hasanumber('0'or'1')associatedwithitthat determineswhetheritistobeblackorwhite:thereisnothing inbetween.Wewillcallthisa'übinarydisplayÇ'. Notethattermslike'black','white'and'greyvalue'inthe contextofmonochromedisplaysactuallyrefertothecolor emittedbythephosphorusedintheCathodeRayTube(üCRTÇ)ofthe displaydevice(amber,green,white,etc.). Colorimagesgenerallyconsistofthreeüsub-imagesÇ,displayed simultaneously.Thesub-imagesrepresentthedecompositionofthe colorimageintothreemonochromeimages(red,greenandblue, respectively).The(visual)additionofthesub-imagesrestores theoriginalcolorsfortheobserver.Sub-imagesmaybebinary, butusuallytheyhavemorethantwogreyvalues.AnAIMcolor image,forexample,has256possiblegreyvaluesforeachsub- image. Inthecolormodes(medium-,resp.low-resolution),theSTis capableofdisplaying8greyvaluesforallthreeprimarycolors. Consequently,thereare8*8*8=512differentcombinations,each combinationrepresentinganothercolor.Suchagroupofavailable colorsiscalledaüpaletteÇ.Thepaletterepresentsallpossible statesthatanindividualpixelmayhave. Eachcombinationofred,greenandblueintensitiesisperceived withaparticularbrightness,theüluminanceÇ.Luminance informationcanbeusedforthedisplayofcolorimagesona monochromemonitor;AIMallocatesanextrasub-imageforit. Ifapixelhasequalintensitiesforred,greenandblue,it willdisplaythecolor'grey'.Thus,inthepaletteofthemed- resandlo-rescolormodesthereare8levelsofgrey,extending fromblack(R=G=B=min=0)towhite(R=G=B=max=7).Veryoftenthe propertyofacolormonitortodisplaygreyvalues(asopposedto justblackorwhite)ismoreimportanttotheuserthanits capabilityofdisplayingcolors! Aswehaveseenabove,inbinaryimageseachpixelcanhave2 greyvaluesonly.Wesaythateachpixelhas1'übitÇ'(binary digit)ofinformationassociatedwithit.Equivalently,incolor imagestheinformationperpixelishigher;AIMuses32bitsof colorandgreyvalueinformation.Wewillreturntothisshortly. Pixelsandneighbors.ê ÇAnAIMimagecanbevisualizedasa'checkerboard'with256*256 or128*128squarefields(formonochromeandcolorimages, respectively).Eachfieldcorrespondstoapixelcarrying8bits (=1üByteÇ)ofgreyvalueinformationor32bits(= 3+1Bytes)of colorandluminanceinformation. ÇPixels,beingneatlyarrangedinasquareürasterÇofrowsand columns,haveüneighborsÇ.Thenumberofneighborsofaparticular pixelisreferredtoasitsüconnectivityÇ.Onacheckerboard neighborscanbedefinedashavingeitherjustsides,orsidesas wellascorners,incommonwiththereferencepixel.Dependingon thesedefinitionstheconnectivityofpixelsonasquareraster is4or8.Pixelssometimesarearrangedhexagonally.Insucha 'honeycombconfiguration'theconnectivityis6. Imageprocessingroutinescanbedividedintothreegroups: üpointÇ,ülocalÇandüglobaloperationsÇ. Pointoperationsusetheinformationofonesinglepixelinthe inputimagetocalculatetheresultforacorrespondingpixel (notnecessarilythesame)intheoutputimage.Anexampleisthe INVERToperation,wheretheintensitybitforeverypixelinthe (binary)imageisreversed. Inalocaloperationtheresultforonesinglepixelofthe outputimageisdeterminedbytheoriginalpixelandits üneighborhoodÇ.AnexampleofsuchanoperationisLmin('Local minimum'),wheretheoutputpixelgetstheminimumgreyvalueof itsneighborhood(includingitself)intheinputimage.This operationextendsdarkpartsintheimage.Theneighborhoodof interestcanoftenbespecified(i.e.1*3,3*3,5*5,etc.).In AIMtheneighborhoodcanonlybeanoddnumber. Inglobaloperationstheresultforonesingleoutputpixelis derivedfromtheinformationofallpixelsintheinputimage. OneexampleofaglobaloperationistheFouriertransform,used todetermineperiodicfeaturesintheimage.Fouriertransforms arenotsupportedinthisversionofAIM. ëPROPERTIESOFTHEATARISTVIDEOHARDWARE ÇAlmostallofAIM'simageswillprobablyhavetobedisplayedon avideomonitor(eithermonochromeorcolor).Howeverperfectan imagemaybe,itwilleventuallyhavetopassthebottleneckof thecomputer'svideodisplayhardware.Muchoftheeffortof programmingAIMhasbeendevotedtoovercominginevitable limitationsofacomputer'sdisplayhardware.Foracorrect interpretationoftheresults,wewillextendbrieflyintosome detailsofthehardwareusedbytheAtariSTforthecontrolof displays:üvideodisplaymemoryÇ,üvideocontrollerÇandümonitorsÇ. Thevideodisplaymemory. Thecentralpartofthedisplayhardwareisthevideodisplay memory.Independentoftheresolutionchosen(lo-,med-,hi-res), theAtariSTallocates32kB(=256,000bits)ofRAMforstorage oftheinformationofonescreen. Inthehighresolutionmode640*400=256,000pixelscanbe displayed,leavingjustonebit/pixelforgreyvaluedefinition. Asaresultonlybinaryimagescanbedisplayedonamonochrome monitor. ÇAsexplainedabove,acolorimagerequiresmorethan1bitof memoryperpixel.YoumaywonderhowtheSTmanagestoattribute morethan1bittoapixelifinthehi-resmode(1bit/pixel) alltheavailable32kBisusedupalready! Theanswertothisquestioncanbeintuitivelyfoundfromthe terms'medium-resolution'and'low-resolution'forthecolor modes.Inthesemodes,thespatialresolutionofthehi-resmode, 640*400=256,000pixels,issacrificedinordertogivethe pixelsmoreluminance(monochromedisplaysoncolormonitors)or colorinformation.Byreducingtheresolutionto640*200= 128,000(med-res)or320*200=64,000pixels(lo-res),thepixels canbeattributed2or4bitsforluminanceorcolordefinition. Thisallowspalettesof4or16colorstobedefinedinthemed- resorlo-resmode,respectively. Thus,thevideodisplaymemorymaybethoughtofasrepresenting: 1binaryimage of640*400pixels(hi-res) 2binaryimagesof640*200pixels(med-res) 4binaryimagesof320*200pixels(lo-res). Thecolorofaparticularpixelisdeterminedbythe(weighted) bit-valuesofthe2/4correspondingpixelsinthebinaryimages. Thevideocontroller. Thenextstoponourjourneyfromthevideodisplaymemorytothe monitorscreenisthevideocontroller.Thischiptakescareof allthehousekeeping,requiredforconversionofthebytesinthe videodisplaymemoryto'realworld'voltages,compatiblewith theintensityinput(s)ofthemonitors. Sinceinthehi-resmodethepixelscontainjust1bitofgrey valueinformation,thevideocontrollerincorporatesavery simple1bit'A/Dconverter'togeneratetheintensitysignalfor themonochromemonitor. Wehaveseenthat,asaconsequenceofthelimited32kBofvideo displaymemory,inthemed-resandlo-resmodesthepixelscan havejust2/4bitsofcolorinformation.Thiswouldallowfor palettesof4/16colors.Fortunately,AtarihasmadetheSTmuch moreversatile,generouslyincludingthree3bitA/Dconverters inthevideocontroller.Thus,foreachprimarycolorthereare8 greyvaluesavailable,allowingapaletteof8*8*8=512colors. Ifthepixelscontaininformationforthedefinitionof4/16 colorsonly,howdoesthevideocontrollerknowwhatcolorsto choosefromthepaletteof512?Theanswerisintheoperating system;ithasfacilitiestodefinea'sub-palette'of4/16 colors.Eachcolorinthesub-paletterepresentsaparticular combinationofgreyvaluesforred,greenandblue.These combinationsarestoredinalookuptableof4/16elements.The 2/4bitcolorinformationforeachpixelreferstoapositionin thislookuptableandthevideocontrollerconvertstheRGB- combinationstoredthereintothecorrespondingintensitycontrol voltages. ÇWhenacolormonitorisusedtodisplayamonochromegreyvalue image,theoperatingsystemautomaticallyloadsthelookuptable with4/8(whynot16?)greycombinationsformed-resorlo-res displays,respectively. Themonitors. Basically,avideomonitorconsistsofaCRTwithsomeelectronic circuitrytocontrolthedeflectionandtheintensityofthe electronbeam.Amonochromemonitorusesoneelectronbeam.A colormonitorrequiresthreebeams,oneforeachprimarycolor. Forintensitycontrolthemonitorshave1or3inputs.The voltageontheseinputsdeterminestheintensityofthelightspot onthescreen. Note,thatinprinciplebothmonochromeandcolormonitorsare capableofdisplayinggreyvalues:thereisacontinuousrange overwhichtheintensitiescanbecontrolled.However,aswehave seen,theotherdisplayhardwarereducesthenumberofpossible greyvaluesand/orcolorsconsiderably. Thedeflectionoftheelectronbeam(s)iscontrolledbythe monitoritself;forsynchronizationitneedsatriggeringsignal ('sync')whichisproducedbythevideocontrollerandsimply addedtotheintensitysignal(s). ëAIMIMAGESANDTHEIRDISPLAYONTHEATARISTMONTORSÇ TheAIMworkspaceallowssimultaneousstorageof4images:A,B, CandD.Eachimage(monochromeorcolor)occupies64kB.A monochromeimage(storedondiskwiththeextension.IM)consists of256*256pixels,having8bitsofgreyvalueinformation(i.e. 256greyvaluespossible).Colorimages(extension.COL)consist of128*128pixels,having24+8=32bitsofcolorandluminance information.ForthedisplayofcolorimagesAIMsupportsthe ST'slo-resmodeonly(320*200pixelswithapaletteof16 colors). YoumayhavewonderedwhyAIMuses4Bytesofinformationfor pixelsof.COLimages.Wouldn'tacolorimagebeperfectly definedbyjustusing1Byteforeachcolorsub-image?Theanswer is:yes,itwould.Fortechnicalreasons,however,wehavechosen toaddenextrabyteforluminancedefinition.Thisformat allows,forexample,themonochromedisplayof.COLimages. WewilldiscussthedisplayofAIMimagesin4different situations: Monochrome(.IM)images-monochromedisplay(hi-res) Color(.COL)images-monochromedisplay(hi-res) Monochrome(.IM)images-colordisplay(lo-res) Color(.COL)images-colordisplay(lo-res) ê ÇMonochrome(.IM)images-monochromedisplay(hi-res). ü ÇAsanextensiontoourcheckerboardmodelwemightintroducea stackof8checkersplacedoneachfield,representingthe8grey valuebitsofthecorrespondingpixel.Awhitecheckerrepresents a'1',ablackonea'0'.Dependingonitspositioninthestack theweightofawhitechecker(i.e.thecontributiontothegrey value)is128,64,32,16,8,4,2or1.Thus,ifthelowerthree andtheuppertwocheckersarewhite,thecorrespondinggrey valueis(128+64+32+0+0+0+2+1=)227. ë ÇInourmodelwehave8horizontalplanesof256*256checkers, eachplanecontributingitsownweighttothetotalimage.Note thateachplaneitselfactuallycorrespondstoabinaryimage, hencethename:übitplanesÇ.Theupperplane(b1)istheleast significantbitplane,thelowerplane(b8)isthemost significantone.InAIMthe8bitplanesofagreyvalueimagecan beprocessedas8separatebinaryimages.Bitplanesofoneimage canbecopied(command:BCOPY)tooneormorearbitrarilychosen bitplane(s)ofanother.Usingthisfacility,agreyvalueimage canbeusedasa(temporary)storagefor8independentbinary images. ë ÇAsdeterminedbytheSThardware,inthehi-resmodeonlybinary imagescanbedisplayed.AIMcircumventsthislimitationbyusing atrickcalledüsigma-deltatechniqueÇ:the256greyvalues availableintheimagearerepresentedaswellaspossiblebydot densitiesinthedisplay.Thisyieldsa256*256*1bitdisplay withareasonablesuggestionofgreyvaluestothehumaneye. Aspecialsituationoccursifwewanttoprocessmonochrome picturesfromthedrawingpackageDEGAS.MonochromeDEGAS pictures(extension.PI3),beingbinary640*400pixelimages, cannotbeprocesseddirectlybyAIM.Whenreadfromadiskthese imagesareconvertedtoa320*200pixelimagewith5levelsof greyinformation.The640*400pixelsaregroupedin320*200non- overlapping2*2fields.Eachpixelinthefinal320*200pixel imageisattributedtheaveragegreyvalueofthecorresponding 2*2field: 0pixelswhite->greyvalue:0 1pixelswhite->greyvalue:63 2pixelswhite->greyvalue:127 3pixelswhite->greyvalue:191 4pixelswhite->greyvalue:255 Subsequently,theconvertedDEGASimagesgetthesamesigma-delta treatmentastheregular.IMimagesforthedisplayofgrey values.Notethatinthisprocessinformationislost;the resultingimagehaslessthan320*200*3=192,000bitsof information(originally:256,000bits).Consequently,eventhough AIMcanre-convert(loosinginformationagain!)theseimagesback toDEGASformat,theoriginalpicturescanneverberecovered completely. Color(.COL)images-monochromedisplay(hi-res).ê Ç ÇForthedisplayofa.COLimageonamonochromemonitor,onlythe 128*128pixelluminancesub-imageisused.Thesigma-delta techniqueagainconvertsthisluminanceinformationtodot densities. Monochrome(.IM)images-colordisplay(lo-res).ê Ç Monochrome(.IM)imageshave8bits(256levels)ofgreyvalue information.Whenan(.IM)imageisreadfromdiskinthelo-res mode,AIMwillloadthe16colorpalettewiththemaximumof8 levelsofgreyandappliesthesigma-deltatechniqueto interpolatebetweenthese8levels. Color(.COL)images-colordisplay(lo-res). Forthedisplayof.COLimagesinthelo-resmodeAIMfirst determinesanoptimumpaletteof16colorsfromthepaletteof 256*256*256colorsavailableintheoriginalimage.The8bitsof luminanceinformationisnotused. Thepaletteforaparticularscreenisderivedfromtheimagein theactivewindowonly;thismaynotbetheoptimumpalettefor otherimagesdisplayedsimultaneously. InthismodecolorpicturesfromtheDEGASandNeochromedrawing packages(extensions.PI1and.NEO,resp.)canbereadbyAIM also.Again,sizereduction(to160*100pixels),greyvalue attributionandthesigma-deltatechniqueareappliedforthe processinganddisplayoftheseimages. ë IMAGEPROCESSINGCAPABILITIESOFAIM Ç TheimageprocessingsoftwarepackageAIMhasbeenwritteninthe Clanguage,usingtheMarkWilliamsC-compiler.Itmakes extensiveuseoftheGEMwindowstructurefortheuserinterface andthedisplayofimages.AIMallowsprocessinganddisplayof monochrome,binaryandgreyvalueimagesaswellasofcolor images.Ithasaninterface(viatheAtariST'sDMAbus)to monochromeandcolorframegrabbers,developedinthePattern RecognitionGroup. Theuserinterface. Imageprocessingoperationscanbechosenfromdrop-downmenues, usingdialogboxes,aswellasbytypingtheappropriatecommand inthecommandbox.Inanycasethecommandsaretransferredas stringsbytheWindowManagementSystemtotheCommand Interpreter.ManyoftheAIMcommandscannotbeinvokedfrom drop-downmenues.Theywillbedesignatedbyanasterisk(*). Thecommandinterpreterstartstheimageprocessingroutine.It isnotnecessarytotypethecompletecommandname;anynumberof charactersthatdoesn'tgiveambiguitieswilldo. ÇSequencesofcommandsmaybecombinedinacommandfileormacro. Commandfilescanbenestedto16levels.Theyareautomatically createdbyactivatingtheLoggingfeature;anyoperationyou performafter'Logging'ischecked,willbelogged.Macroscanbe savedtodiskforre-use.Theextensionforthefilenameis.AIM. Amacroisexecutedbytypingthename(withoutextension), precededbythecharacter'@'.Clickingthemenu-entryMacroand thenameinthedialogboxwilldothesame.Pausingthe executionispossiblebyhittingtheEsc-key.Proceedwiththe macrobyhittingReturnorstopbyclickingintheappropriate box. ThecommandsintheCommandWindowcanbeeditedsimply.Delete andbackspaceworklikeinmostwordprocessors.Control-Xdeletes awholeline. Allcharacterstypedafterastar(*)willbeignoredbythe commandinterpreterandthuscanbeusedforcomments. Typingaquestionmark(?)and<Return>willdisplayalistof availableoperations.Thesyntaxandashortexplanationforthe commandscanbefoundinHelp-files.ChooseHELP!!!inthe Utilitiesmenu.Note,however,thatthisversion(2.5)ofAIMhas beenextensivelyrevised.Manyoperationshavebeenaddedand someoftheexistingoperationsallowmoreparameterstobe specifiedthanthosegiveninthecurrentHelp-filesanddialog boxes.Fortunately,alloperationsareupwardcompatible.The AIMANUAL.DOCintheCOMPRESSfoldercontainsanupdated descriptionofalltheavailableoperations. ëEXAMPLESANDDEMOSÇ TheexamplesontheAIMdiskareself-explaining.Theonly prerequisiteisareasonableunderstandingoftheprevious chapters.Becausethediskspacewaslimited,wecouldonly includesomeofthemostrelevantexamples.Consequentlythey coverjustafewofthepossibleoperations;mostofthe operationswillhavetobeexploredbyyourself! ThecommentedcommandfilesEXAM1.AIM->EXAM5.AIMhavebeen includedsothatyoucancheckyourmanuallyenteredexercises. Besidesthepreparedexamplesthreeextrademos(macros)have beenincludedonthedisk:DEMOCER.AIM,DEMOHFE.AIMand DEMOSCH.AIM.Theycontainsomeoperationsandstrategiesthatmay beofinteresttoyouafterhavingcompletedtheexamples successfully. Finallytwoexercisesareaddedto(self-)testyourabilities.If thingsgototallywrong,don'tfeelembarrassedbypeepinginto themacrosPRAK1.AIMandPRAK2.AIM;theycontainpossible solutionsfortheseexercises.Theexercisesarecompressedfiles andcanbefoundinthefolderCOMPRESS. Incaseyoumissedapoint:inagreyvalueimage'black' correspondswithavalue0,'white'withavalue255.Inabinary image(orabitplaneofagreyvalueimage)thesevaluesare0 and1,respectively. ÇAllexamplesandexerciseswillworkinbothmodes,lo-rescolor andhi-resmonochrome.Forbinaryoperationsamonochromemonitor isrecommended.Somedemosaremoreimpressiveincolor,though! üExampleEXAM1.AIMÇ Operationsused:Histogram,Equalize FirstwewillretrievetheimageMOON.IM.Clicktheentry'Read Image'underthe'File'menu.Thedefaultsintheselectorbox thatpopsup('Window:A'and'STANDARD')areacceptable,so click'OK'(orhittheReturnkey).Thoseofyouwhoprefera commandlineinterpretermightaswelltypein'readA'instead ofusingthemouse. NextaGEM'ITEMSELECTOR'appears.Choose'MOON.IM'andaftera fewsecondsapictureappearsinWindowA,thatshowsthefoot ofmoonlandersuroundedbymoondust.Youwillagreethat, normally,youwouldthrowsuchapictureinthetrashcanand probablydeclarethemissiontothemoonaflop.AIMcanhelpyou dealwithsuchfrustations! Justtogetafeelofthepoweraccumulatedinthattiny3.5", trytheoperation'Equalize'underthemenu'GreyOps'.ChooseA astheinputimageandBastheoutput.Looksbetter,doesn'tit? Remember,theinformationwasalreadyintheimage;evenwithout AIMthemissiontothemoonwasn'tinvain!Buttheapplication ofAIMcertainlyaddstothesucces. Tounderstandwhathashappenedtotheoriginalimage,trythe entry'Histogram'under'Utilities'.Thiswillshowagraphof thefrequencyofoccurrenceofaparticulargreyvalue(or rather:asmallrangeofgreyvalues)intheimage,plotted againstthegreyvalue.Frequenciesareplottedvertically,grey values(0->255)horizontally.Thehistogramindicatesthatthe greyvaluesintheoriginalimagewereconcentratedina relativelysmallrange(centeredaroundgreyvalue50). Fromthehistogramwecanconcludethatthegreyscalefrom0to 255hasnotbeenusedefficiently;nearlyallpixelshaveagrey valuebetween,say,25and75.Theresultisthat,inaway,we arelookingattheprintingonatoyballoon;onlyifweinflate theballoonsomeofthesmallerdetailswillshow.Ifsomehowwe couldstretchtheoriginalgreyvaluerange,sothat25would shiftto0and75to255,thesameeffectmightbeimposedonour moonimage.Throwingawaytheranges0->25and75->255doesn't domuchharm,becauselittleinformationwillbelost. Equalizationistheairpumpthatinflatesagreyvaluerange. Unlikeanairpump(andthesuggestionintheprevious paragraph!)itdoessointelligently.Thatis,thestretching effectonaparticularrangeofgreyvaluesdependsontheheight ofthehistogramforthatrange.Rangeswithhighvaluesinthe histogramwillbeexpandedatthecostofthecompressionofless prominentranges.Note,thattheremaybemultiplepeaksand valleysinthehistogram.Toseewhatwemean,determinethe histogramofimageBandcompareitwiththeoriginalone.The frequencies(verticallines)aredividedmoreevenlyoverthe availablerangeofgreyvalues. ÇRangeswithhighfrequenciesofoccurrenceintheoriginal histogramhavebeenstretchedandthosewithlowfrequencies compressed.Ifyouarenotcompletelysatisfied,performthesame operationsontheimageTRUI.IM,whichhasmultiplepeaksinthe histogram. Equalizationisaniceexampleofimageprocessing:important features(inthiscasegreyvalueranges)areemphasizedatthe costoflessimportantfeatures.Yet,inthisparticular operationnoinformationislost;a'de-emphasis'procedure(not availableinAIM)couldrestoretheoriginalimage. Hint:ifyouwanttogetridofthehistograminaparticular display,use'Gdisplay'intheUtilitiesmenu. üExampleEXAM2.AIMÇ Operations:Gradx,Grady,Enhance ReadtheimageTRUI.IMin(thedefault)WindowA.Thisimagehas becomeakindofreferenceformanygroupsinvolvedinimage processing. Performtheoperation'Gradx'intheGreyOpsmenu.Acceptallthe defaultsproposedbyAIM.WorryabouttheMultiplyandAddvalues later,orbetterstill,experimentwiththem. TheGradx('gradientinx-direction')operationdeterminesthe rateatwhichgreyvalueschangeintheoriginalimageand attributescorrespondinggreyvaluestothepixelsoftheoutput image.Anoutputpixelgetsagreyvalueaccordingtothe formula: grey-out=(gradient-in)*Multiply/1024+Add. Thisoperationshowsverticaledgesintheimage.Anedge correspondstoastepingreyvalue.Notetheeffectonedgesin thegirl'shat. NowperformGrady.ThistimesendtheoutputtoWindowC.Note theresultforhorizontaledges(mouth,eyebrows,etc.). Arelatedoperation,whichisgenerallyusedtosharpen-up picturesistheEnhance-operation(inmenuGreyOps).It determinesthesecondderivative(inbothdirections)ofthe grey-valuesintheinputimage.LikeGradx/yitemphasizeshigh frequenciesintheimage.Highfrequenciesoccuratsharpedges. TrytheEnhancefilterontheoriginalTRUI.IM.andenjoythe improvement.Butwhatdidwedotohereyes?Seemswecreatedher contactlenses... üExampleEXAM3.AIMÇ Operation:Medianfilter ReadimageBNOISE.IM.Thisimagehasintentionallybeendegraded by'shotnoise'.Thistypeofnoiseoccursofteninthe transmissionofsignals.Someofthepixelsdifferverymuchin greyvaluefromthesurroundingpixels.Inthiscasetheyare white.Wewanttousefilteringtoremovethenoise. Usuallyitisdifficulttopredicttheeffectsoffiltering. Variousparametersmustbespecified,requiringmuchintuition. Try,forexampletheentry'Filter'intheGreyOpsmenu! Agoodfilterfortheproblemofshotnoiseistheso-called medianfilter.InthisoperationAIMdeterminesthegreyvalues in3*3neighborhoodsforeachpixelintheinputimage.These valuesaresortedintableswith9entries.The5thentryineach table(the'median')becomesthegreyvalueofthecorresponding outputpixel.Thegreyvalueofashotnoisespeckislikelyto beintheextremesofthetablesinwhichitappears,andthusit willberemoved. Performtheoperation'Median'intheGreyOpsmenu.Onlyafew stubbornnoisespeckswillsurvive.Theymayberemovedbya repeatedapplicationofthisfilter. üExampleEXAM4.AIMÇ Operations:Copy,Lmin,Lmax,(*)mindev,Gdisplay OncemorewecallonTRUI.IM;readitfromdisk.Wewillneeda secondcopyofthisimageinWindowB.Therearetwooptions: a.ReadtheimageagainfromdiskandhaveAIMputitinWindowB b.Use'Copy'inmenuGreyOpstocopyWindowAinWindowB. Notethatthereisa'Copy'entryintheBinOpsmenutoo.This binaryoperationcopiesonlybitplanes,asopposedtoagrey valuecopy,whichcopiesall8bitplanesofanimageatonce. Ifallwentwell,younowhavethefabulousTruitwinsonscreen. Thefirstoperationtobeperformedis'Lmin'intheGreyOps menu.Thisoperationlooksforthelowestgreyvalueinann*n neighborhood(ntobespecifiedbytheuser)oftheinputpixel andattributesittothecorrespondingoutputpixel.Thisfilter hasatendencyofexpandingdarkerareasintheimage.Tryit withn=9andputtheresultinWindowC.Experimentalittlewith differentvaluesforn.Observetheeffectsonnostrilsandeyes! Inanycaseendwitha9*9Lmin-nedTRUI.IMinWindowC. Nextapplythe'Lmax'filteronWindowA.Choosen=9andputthe resultinWindowD.Thisfilterexpandslighterarea'sinthe image.Wehopethatuntilnowyouacquiredenoughintuitiontobe abletopredicttheeffect:gonearethosenostrilsandgoneis thatmouthinthe'lightershadeofpale'thatsurroundedthem. ÇButnowtherealimageprocessorarises.Wearegoingtousean operationthatusesthreeinputimages:(*)mindevC,D,A.This operationcomparescorrespondingpixels(i.e.pixelswiththe samecoordinates)intheoriginalimage(A)andthetwoother images(CandD).Theoutputpixelgetsthegreyvalueofthe correspondingpixelinCüorÇD,dependingonwhichoneofthetwo isclosesttothegreyvalueinA.Bigdeal,butwhatisthe effect? Let'stryitfirst.The(*)mindev('minimaldeviation') operationcannotbeinvokedfromthemenu,sowehavetotypeit indeCommandWindow:typemindevC,D,A<Return>.Theresultwill beshowninWindowA.Compareitwiththeoriginalimage,which isstillluringinWindowB.Apparentlytheedgesintheimage (cheeks,hair)havebeensharpened.TheneteffectofLmin,Lmax andmindevis,thatnearedgesthereisabacklashingreyvalue. Inotherwords,greyvaluesarereluctantagainstchanges; changesoccursuddenlyasafunctionofposition.  üExampleEXAM5.AIMÇ Operations:(*)Clear,Threshold,Invert,Propagation,Bdisplay, Reset,Exorand(*)Measure. ReadtheimageCERMET.IMfromdisk. Ourpurposewiththisexampleistodosomemeasurementsonthe objectsinthisimage.The'(*)Measure'operationrequiresthe objectsofinteresttobewhiteinablackbackground.Inthe menu'BinOps'wehavesomeusefuloperationsatourdisposalto achievethatgoal. Inordertobeabletoapplybinaryoperations('BinOps'),we firstwillmakethegreyvalueimageinWindowAbinary.The operation'Threshold'intheGreyOpsmenumakespixelswitha greyvaluebelowthethresholdblackandpixelswithagreyvalue aboveitwhite.Whentheoperationisinvoked,itproposesa thresholdwhichistheaveragegreyvalueintheimage;itisnot necessarilythebestguess,butusuallyityieldssatisfactory results. Stop!TheThresholdoperationwillaskforthebitplanewherethe binaryimageistobestored(itdefaultstob1inWindowB). However,asourpreviousexerciseswillhaveleftsomegarbagein theotherbitplanesofWindowB,wefirsthavetoclearthose. Theoperation'(*)Clear'isprovidedtoclearallbitplanesofa windowatonce(clear=greyvalue0=black!).Type:clearB <Return>. NowwearereadyfortheThresholdoperation;clickonitinthe GreyOpsmenuandacceptthedefaults. Asweneedwhiteobjectsinablackbackground,thepicturein WindowBmustbeinverted.Clickon'Invert'intheBinOpsmenu (acceptthedefaults). ÇNotethatsomeoftheobjectsneartheframeofthewindoware onlypartiallydisplayed.Wedon'twantthe'(*)Measure' operationtotaketheseincompleteobjectsintoaccount,buthow canwegetridofthem?Howcanwedistinghuishthemfrom completeobjects?Theansweris:usethe'Propagation'operation (inCelOps). The'Propagation'operationallowsustoisolateobjects.White pixels(called'seeds)will'grow'(expand)underthecondition thatinagivenreferencebitplane(the'mask')thecorresponding pixelsarewhitetoo.Wecanvisualizethisoperationastaking placeinafieldwithseeds(=whitepixelsthatwewanttogrow toshapespresentinthemask).Thereferencebitplane(the 'mask')istheculturemedium;ithasfertileandinfertile regions.Ifaregionisfertile(i.e.ithaswhitepixelsthat coincidewiththeseedpixels),theseedcangrow.Inone 'iteration'(propagationstep)apixelgrowsinaccordancewith itsconnectivity:onewhitepixelisaddedinallconnecting directions.InAIMtheseedbitplaneiscalledCLP(Cellular LogicPlane). Inthenextiterationtheseedbitplane(CLP)thatresultedfrom thefirstiterationiscomparedoncemorewiththemaskbitplane. CoincidingwhitepixelsinCLPandmaskwill,oncemore,grow. Thiswillcontinueuntilallseedshavegrowntotheedgesofthe fieldsinthemask;wehavecreatedanexactcopyofallthe fieldsthatinthefirstiterationcoincidedwithatleastone seed.Thetrickofisolatingparticularfieldsinthemaskisto createaCLPwithseedsthatcoincidewiththosefields. Let'sreturntoourproblemofisolatingtheincompletewhite objectsatthebordersofbitplaneb1inWindowB.Itwouldbe niceifwehadabitplaneavailable(tobeusedastheCLP)with whitepixelsalongtheframeüonlyÇ.ThePropagationoperation wouldmakethoseborder-pixelsgrowtotheexactshapesofthe intersectedobjectsinthemask(inourcase:b1).Theresult wouldbe:twoimageshavingtheunwantedobjectsincommon. Subsequentapplicationofthelogicaloperation'Exor'(inthe BinOpsmenu)onthesetwoimageswillfinallyremovetheunwanted objects.Exoristheinequalityoperator:itcompares correspondingpixelsinboth(binary!)imagesandoutputsawhite pixelif(andonlyif)theyaredifferent. Nowthatwehopefullyunderstanditwecannotwaittoputour knowledgetothetest.Theonlyimageweneedisbitplaneb1in WindowB.ClickonPropagation.Wewillhavetochangealmostall ofthedefaults!Ourmaskbitplaneisb1insteadoftheproposed b2.Furthermore,wewilluseb2asourCLPinsteadofb1.The connectivityisnotimportant,butchangeitto8.Thenecessary whiteedgeforourCLPcanbechosenwiththeentry'Edgevalue'. Choose1(forwhite).Makethenumberofiterations100;thatis toomany,butAIMwillstopautomaticallywhenallgrowthhas stopped. CheckallentriesintheDialogBox:ifsomethingiswrongthere isariskthatyouhavetostartalloveragain!YourSTwon't complain,butwedon'twanttofrustrateprospectiveimage processorsinsuchanearlystage. ÇHitReturn.Ifalliswell,youwillseeanimageinbitplaneb2 ofWindowBthathastheintersectedobjectsonly.Wrong?Don't worry,justkeeptrying.Don'tforgettoclearbitplaneb2prior toeveryPropagationoperation. NextinvokeoperationExor(inBinOps).Useb1andb2inWindowB asinputsandb3asoutput. Atlastourimageisreadyforthe(*)Measureoperation.This operationassumesthattheimageisinb1ofWindowB,sowemust copyb3tob1first.Click'Copy'inBinOpsandchoosethe appropriatebitplanes. (*)Measureperformsanumberofmeasurementsontheobjectsand displaystheresultsintheCommandWindow.Itassignsnumbersto theobjects(left->right,up->down)andlabelseachobject withagreyvalueaccordingtothisnumber.Thiscanbeobserved inWindowD. Typemeasure(meawilldo!)<Return>.NotethattheCommand Windowcanbesizedtobringallmeasureddataonscreen. üPRACTICALEXERCISESÇ PRAK1. StudytheimagePRAK1.IM.Yourassignmentistoseparatethenuts fromthebolts.Inotherwords:useAIMtogeneratetwoimages, onecontainingthenuts(includingtheonethatrestsonits side!)andtheotheronethebolts.Counttheobjectsafterwards. PRAK2. TheimagePRAK2.IMshowsahandfulofelectroniccomponents.Try toseparatethem.Useyourfantasyandsurprisehusbands,friends andneighbors!