Departmen of Computer Science Univ ersit of Massac usettsAmherst Abstract raditional generativ Mark random 57356elds for seg men ting images mo del the image data and corresp onding lab els join tly whic requires extensiv indep endence assumptions f ID: 31784
Download Pdf The PPT/PDF document "SIGN DETECTION IN NA TU RAL IM GES WITH ..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
SIGNDETECTIONINNATURALIMAGESWITHCONDITIONALRANDOMFIELDSJerodWeinman,AllenHanson,andAndrewMcCallum{weinman,hanson,mccallum}@cs.umass.eduDepartmentofComputerScienceUniversityofMassachusetts-AmherstAbstract.TraditionalgenerativeMarkovrandomeldsforseg-mentingimagesmodeltheimagedataandcorrespondinglabelsjointly,whichrequiresextensiveindependenceassumptionsfortract-ability.Wepresenttheconditionalrandomeldforanapplicationinsigndetection,usingtypicalscaleandorientationselectivetex-tureltersandanonlineartextureoperatorbasedonthegratingcell.Theresultingmodelcapturesdependenciesbetweenneighbor-ingimageregionlabelsinadata-dependentwaythatescapesthedicultproblemofmodelingimageformation,insteadfocusingef-fortandcomputationonthelabelingtask.Wecomparetheresultsoftrainingthemodelwithpseudo-likelihoodagainstanapproxima-tionofthefulllikelihoodwiththeiterativetreereparameterizationalgorithmanddemonstrateimprovementoverpreviousmethods.INTRODUCTIONAppearinginIEEEInternationalWorkshoponMachineLearningforSig-nalProcessing,S~aoLus,Brazil,Sep.2004.Imagesegmentationandregionlabelingarecommonproblemsincomputervision.Inthiswork,weseektoidentifysignsinnaturalimagesbyclassifyingregionsaccordingtotheirtexturalproperties.Ourgoalistointegratewithawearablesystemthatwillrecognizeanydetectedsignsasanavigationalaidtothevisuallyimpaired.Genericsigndetectionisadicultproblem.Signsmaybelocatedanywhereinanimage,exhibitawiderangeofsizes,andcontainanextraordinarilybroadsetoffonts,colors,arrangements,etc.Forthesereasons,wetreatsignsasageneraltextureclassandseektodiscriminatesuchaclassfromthemanyotherspresentinnaturalimages.Thevalueofcontextincomputervisiontaskshasbeenstudiedinvariouswaysformanyyears.Twotypesofcontextareimportantforthisproblem:labelcontextanddatacontext.Intheabsenceoflabelcontext,localregionsareclassiedindependently,whichisacommonapproachtoobjectdetection.Suchdisregardforthe(unknown)labelsofneighboringregionsoftenleadstoisolatedfalsepositivesandmissingfalsenegatives.Theabsenceofdatacon-textmeansignoringpotentiallyhelpfulimagedatafromanyneighborsofthe regionbeingclassied.Bothcontextsaresimultaneouslyimportant.Forin-stance,sinceneighboringregionsoftenhavethesamelabel,wecouldpenalizelabeldiscontinuityinanimage.Ifsuchregularityisimposedwithoutregardfortheactualdatainaregionandlocalevidenceforalabelisweak,thencontinuityconstraintswouldtypicallyoverridethelocaldata.Conversely,localregionevidencefora\sign"labelcouldbeweak,butastrongedgeintheadjoiningregionmightbolsterbeliefinthepresenceofasignatthesitebecausetheedgeindicatesatransition.Thus,consideringboththelabelsanddataofneighboringregionsisimportantforpredictinglabels.Thisisexactlywhattheconditionalrandomeld(CRF)modelprovides.Theadvantageofthediscriminativecontextualmodeloveragenerativeonefordetectiontaskshasrecentlybeenshownin[8].Wedemonstrateatrainingmethodthatimprovespredictionresults,andweapplythemodeltoachallengingreal-worldtask.Firstthedetailsofthemodelandhowitdiersfromthetypicalrandomeldaredescribed,followedbyadescriptionoftheimagefeaturesweuse.Weclosewithexperimentsandconclusions.RANDOMFIELDSModelFormanycomputervisiontasks,thepriorprobabilityofthedatabeingob-servedisinconsequential.Imageshappen.Weareprimarilyinterestedinwhatmaybeinferredwhengiventheimages.However,probabilitydistribu-tionsoverlabelsyandanimagexhavetraditionallybeenmodeledjointly,withtheimagepriorprobabilitybeingignoredatclassicationtime.Forthatreason,generativejointmodelsrequireunnecessarymodelingeortandmorecomputationthantheirconditionalcounterparts.MarkovrandomeldsareprobabilitydistributionsparameterizedbyagraphtopologyG=(V;E).Fortractabilityreasons,typicalgenerativeran-domeldstreattheinteractionbetweenlocaldataanditslabelindependentlyoftheinteractionbetweenneighboringlabels.Thejointdistributionisthusfactoredintotheprioronlabelassignmentsandtheprobabilityoflocallyobserveddata,conditionedonthesinglesitelabel:p(y;x)=p(xjy)p(y),1ZYC2C C(yC)Yv2V v(yv;xv);(1)where ()arecompatibilityfunctions,Zisanormalizingconstantmakingtheexpressionaprobabilitydistribution,Cisafamilyofcliquesofthegraph,andyCarethevariablesinagivencliqueCV.Inthismodel,objectsx(e.g.,patchstatistics,salientfeatures,etc.)fromeachclassy2Yaregener-atedbyaclass-conditionalprobabilitydistributionp(xjy).Thisrequiresnotonlyamodelforeveryclasswewishtodistinguish,butanaccurategenerativebackgroundmodelevenforclassesofnointerest;anon-trivialtaskbecausetherealworldcontainsamyriadofimage\classes"(regiontypes,textures, x3y3x4y4x2y2x1y1xyyyy1234Figure1:Left:Traditionaljointrandomeldoverdataxandlabelsy(cf.Eq.1).Right:Conditionalrandomeldwheredataisobserved(cf.Eq.2).objects,etc.).Inshort,itisgenerallymorediculttoexplaintheprocessesthatgenerateclassdatathanitistomodeltheboundariesbetweenclasses.Inthelatterapproach,onlyboundariesamongclassesofinterestmustbedis-tinguished,withtheremaindereasilycollapsingintoasingle\background"class.Modelingtheinteractionsbetweendataandlabelsseparately,as(1)does,isoftentoolimitingformanycomputervisiontasks;wethereforeusearecentlyproposedmodelthathandlestheinteractionbetweensitelabelsinacontext-dependentway[9],describingitnext.Therandomeldgraphtopologycommonlyusedforjointimagelabelingproblemsisthelattice,(Figure1),wherecliquesaresinglenodesandedges.Weuseahomogeneous,anisotropicrandomeld.Thus,cliquesofthesameclassusethesamecompatibilityfunctionsregardlessofimagelocation,buthorizontalandverticaledgesareconsidereddierentclassesandthushavedistinctcompatibilityfunctions.Anisotropyallowsthemodeltolearnanyorientationalbiasofthelabels.Ourconditionalrandomeldhastheformp(yjx)=1Z(x)Yv2V V(yv;x)Y(u;v)2E E(yu;yv;x)(2)=1Z(x)exp0Xv2VF(yv;x)+X(u;v)2EG(yu;yv;x)1;(3)whereZisnowanobservation-dependentnormalizer.Thecompatibilitiesarefunctionsofcliquelabels,allowingneighboringlabelinteraction,buttheyarealsofunctionsoftheentireobservation.Thisdiersmarkedlyfrom(1)byallowingdata-dependentlabelinteraction(seeFigure1).FandGarevector-valuedfeaturefunctions,andandarevectorsofparametersfornodesandedges,respectively.Nodelabelscomefromadiscrete,nitealphabetY.Weuseonesetofobservationfeaturesfornodesandedgesandtransformthemintofeaturefunctions(observation,labelpairs)usingtherelationshipfky(yv;x)=(y;yv)fk(x)gjy;y0(yu;yv;x)=(y;yu)(y0;yv)gj(x);wheref= fkk=1:::Kisavectorofnodefeatures(i.e.,texturestatisticsof aregion)andg= gjj=1:::Jisavectorofedgefeatures(i.e.,dierencesbetweenstatisticsofneighboringregions),sothatF= fkyk=1:::K;y2YandG=gjy;y0j=1:::J;y;y02YY:Thus2RKjYjand2RJjYj2.WhenE=;,themodelusesnolabelcontextandiscommonlycalledaconditionalmaximumentropyclassier(hence,MaxEnt),orlogisticregression.TrainingandInferenceParametersforprobabilisticmodelslikeCRFsaregenerallysetbymaximiz-ingthelikelihoodofadatasample.Unfortunately,inferenceforanyrandomeldwiththelatticetopologyisintractableduetoZ(anexponentialsum).MarkovchainMonteCarlo(MCMC)(seee.g.,[16])isoftenusedtoapproxi-mateZinsimilargenerativemodels.However,inourconditionalmodelZisdependentontheimagedataxandmustbeestimatedforeachobservationinthesample.Asimplerapproximationistomaximizethepseudo-likelihood(PL)[1],whichistheproductoftheprobabilitiesofnodesgiventheirneigh-boringlabels.Thenormalizersarethensummationsoverlabelsatasinglenode,ratherthanthepossiblelabelingsofallnodes.ArelativelynewalternativetoMCMCandPLforapproximatinglike-lihoodiscalledtreereparameterization(TRP)[15].Inferenceingraphicalmodelswithoutcycles(unlikethelattice)isveryecient,i.e.,duetothejunctiontreealgorithm(e.g.,[10]).Animportantconsequenceofthejunc-tiontreealgorithmisthatmarginaldistributionsarerevealedonpairsofneighboringnodes,inducinganalternativefactorizationofthejointdistribu-tion.TRPoperatesbyusingjunctiontreetocomputetheexactmarginalsonaspanningtreeofthecyclicgraph.Thespanningtree'sfactorizationisthenplacedbackintotheoriginalgraph,andtheprocessrepeatswithdierentspanningtreesuntiltheparameterizationconverges,leavingthemarginals.WedemonstrateimproveddetectionperformanceusingTRPtoapproximatethelikelihoodoverpseudo-likelihood.Thelikelihoodfunctionisconvexandmaybeoptimizedgloballyviagradientascent.Pseudo-likelihoodissensitivetoinitialization,however,sonodeparametersareoptimizedrst.Weusethequasi-newtonL-BFGSalgorithmformaximization.Topreventtrainingproceduresfromoverttingparametersinconditionalmodels,apriorisintroduced,andtheposteriorismaximizedratherthanlikelihood.Weemployadiagonalzero-centeredGaussianprioronparameters[2](similartoweightdecayinneuralnetworksorridgeregression);variancesareexperimentallydeterminedthroughcross-validation.Giventheimagedata,ourmodelsimplyyieldsajointposteriordistribu-tiononlabelings.Wheninterestedinpickingahardandfastlabelforeachregionoftheimage(nodeinthegraph),thequestionbecomeswhattodowiththatdistribution.Asimple,oft-usedansweristonditsmaximum.Thatis,usemaximumaposteriori(MAP)estimation:^y=argmaxy2YjVjp(yjx): Thissearchspaceisintractable.However,aslightalterationofTRPallowsMAPestimatestobequicklycalculated.Asimpleralternativeistosearchforalocalmaximumoftheposterior,anestimatecallediteratedconditionalmodes(ICM).Givensomeinitiallabelingy0,subsequentlabelsaregivenbyyk+1v=argmaxyv2YpyvjykN(v);x;8v2Vuntilyk+1=ykoraniterationlimitisexceeded.Often,theinitiallabelingcomesfromthelocalcompatibilitymaximumy0v=argmaxyv2Y (yv;x):Likemanypointestimates,theMAPestimationhasanimportantcaveat:poorpredictionscanresultwhenthemaximumoftheposteriorisnotrepre-sentativeofmostoftheotherlikelylabelings[6].Analternativemethodforpredictioniscalledmaximumposteriormarginal(MPM)estimation,^yv=argmaxyv2Yp(yvjx);8v2V;whichaccountsfortheprobabilityofalllabelings,notsimplythemaxi-mal(joint)labeling,bychoosingthelabelateachnodethatmaximizesitsmarginalprobability.MAPandMPMareequivalentintheMaxEntclassi-ersincenodelabelsareindependent.MarginalizationsuersfromthesamecomputationalcomplexityproblemsasMAP,butsinceTRPreveals(ap-proximate)marginalsonthenodes,itiseasilyusedforMPM.ComparisonsbetweenICMandMAPestimatedwithTRParegivenintheexperiments.IMAGEFEATURESFORSIGNDETECTIONTextandsigndetectionhasbeenthesubjectofmuchresearch.Earlierap-proacheseitheruseindependent,localclassications(i.e.,[5,7,11])oruseheuristicmethods,suchasconnectedcomponentanalysis(i.e.,[4,14]).Muchworkhasbeenbasedonedgedetectorsormoregeneraltexturefeatures,aswellascolor.Ourapproachcalculatesajointlabelingofimagepatches,ratherthanlabelingpatchesindependently,anditobviateslayoutheuristicsbyallowingtheCRFtolearnthecharacteristicsofregionsthatcontaintext.Ratherthansimplyusingfunctionsofsinglelters(e.g.,moments)oredges,weusearicherrepresentationthatcapturesimportantrelationshipsbetweenresponsestodierentscale-andorientation-selectivelters.Tomeasurethegeneraltexturalpropertiesofbothsignandespeciallynon-sign(hence,background)imageregions,weuseresponsesofscaleandori-entationselectivelters.Specically,weusethestatisticsoflterresponsesdescribedin[13],wherecorrelationsbetweensteerablepyramidresponsesofdierentscalesandorientationsaretheprominentfeatures.Abiologicallyinspirednon-lineartextureoperatorfordetectinggratingsofbarsataparticularorientationandscaleisdescribedin[12].Scaleandorientationselectivelters,suchasthesteerablepyramidorGaborlters,respondindiscriminatelytobothsingleedgesandoneormorebars.Grating FTIPrimarySimpleFiltersSecondaryOrthogonalFilterWeightedSimpleFiltersReceptiveFieldMaximumsGratingAND-likeIndicatorFinalGratingResponseQPMMMMMMFTIRFigure2:Gratingcelldata\rowforasinglescaleandorientation.TwoboxesatI,T,andFrepresentcenteronandcenterolters,whiletheboxesatMareforthesixreceptiveelds.cells,ontheotherhand,respondselectivelyonlytomultiple(threeormore)bars.Thispropertyisanidealmatchfordetectingtext,whichisgener-allycharacterizedbya\grating"ofstrokes.Theoriginalmodeliscontrast-normalized,butweexpecttextinsignstohavehighcontrastforreadability,soweomitanynormalizationwhencalculatingI;!;,theresponseofanin-putimagetoalterwithpreferredorientation,spatialfrequency!,andphase(Figure3,upper-right).Furthermore,lettershavealimitedaspectratio,thusthebarsintexthaveboundedheight.ForthisreasonwesubjecttheresponsesI;!;toasecondroundoflteringwithoutputT;!;,where;!;stillindicatestheparametersoftheprimarysimplelter.Thesec-ondarylterhasanorthogonalorientation+2,acenter-onphaseof,andshouldhaveafrequencyofnomorethan!/2.Toelicitstrongerresponsesfrombarsoflimitedheight,theoriginalsimplelterresponseisweightedbytheperpendicularresponsewiththeSchurproductF;!;,I;!;T;!;.Oncetheweightedresponsesarecalculated,abinarygratingcellsubunitQ;!indicatesthepresenceofagratingateachimagelocation.Tomakesuchadetermination,alternatingstrongmaximumcenter-on(=)andcenter-o(=0)responsesM;!;narerequiredinreceptiveeldregionsR;!;n( 3n2)oflength1/(2!)alongalinewithorientation(Figure3,bottom).WeletthenaloutputP;!bethemeanresponseamongthereceptiveeldswhereQ;!indicatesagratingandzeroelsewhere.Thisalsodiersfromtheoriginalmodel,whichsimplygivesthespatialaverageofthegratingindicator.Useofactuallterresponsesintheoutputisimportantbecauseitrepresentsthestrengthofthegrating,ratherthanonlyitspres-ence.Aftertakingmaximumresponsesoverasetofscales,weusethemean,max,variance,skewandkurtosisoftheoutputsinaregionasfeatures.Additionally,histogramsofpatchhueandsaturationareused,whichalsoallowsustomeasurecolordiscontinuitiesbetweenpatches.Usinganalgorithm[3]thatranksdiscriminativepowerofrandomeldmodelfeatures,wefoundthetopthreeintheedge-less,context-freeMaxEntmodeltobe(i)thelevelofgreenhue(easilyidentifyingvegetationasback-ground),(ii)meangratingcellresponse(easilyidentifyingtext),and(iii)correlationbetweenaverticallyanddiagonallyorientedlterofmoderatescale(thesinglemostusefulother`textural'feature). Figure3:Gratingoperatorontext.UpperLeft:Inputimage.UpperRight:Center-onandcenter-osimplelterresponses(=0).Bottom:Sliceofsimplelterresponsesandreceptiveregionsforamarkedpoint.Figure4:Multi-scaletextdetectionwithgratingcells.Left:Inputimagewithsignareasoutlined.Right:Gratingcellresponses.EXPERIMENTSOursignexperimentsarebasedonahand-labeleddatabaseof309imagescollectedfromaNorthAmericandowntownareawithastillcamera.1Weviewthe1024x768pixelimagesasan8x6gridof128x128pixelpatchesoverwhichthefeaturesarecomputed.Thisouterscalewaschosentobalancecomputationalburdenagainsttypicalsignsize;somepatchescontainmoresignthanothers.Letfprepresentthestatisticsofthesteerablepyramid,fgthegratingstatistics,andfh,fsthehueandsaturationhistograms,respec-tively.Ournodefeaturesaretheconcatenatedvectorsf=hfp;fg;fh;fs;1i1Availableathttp;://v;is-w;ww.c;s.um; ss.;íu/;proj;ìts;/vid;i000;. ClassierPredictionRecallPrecisionF1MaxEntMAP48.3668.0256.45ICM49.4270.2357.90PLMAP49.5370.2357.97MPM49.9769.7758.13CRFICM54.0166.5459.49TRPMAP54.5866.0759.57MPM54.5866.0759.65Table1:Predictionresultsforsigns.PLindicatestrainingwithpseudo-likelihood,andTRPtrainingwithapproximatedfulllikelihood.MAPandMPMfortheCRFisestimatedwithTRP.00.20.40.60.8100.20.40.60.81False Alarm RateDetection Rate50%- 75% : 0.909100.20.40.60.8100.20.40.60.81False Alarm RateDetection RateMaxEnt : 0.7061Figure5:Discriminativepower.Left:ROCcurvesandareasforMaxEntonpatchesthatarenearlyall(75%-100%)signormostly(50-75%)sign.Right:ROCcurvesandareasforMaxEntandCRFinpatchescontaininglessthan25%sign.plusabiasfeature.EdgefeaturesaretheL2normsofdierencesbetweenstatisticsatneighboringpatches,g=\n\rfp f0p\r;kfh f0hk;kfs f0sk;1.Theimagesetissplitevenlywithhalfeachfortrainingandtesting.Ta-ble1containstheaveragepredictionresultsof20suchsplits.Sincethisisadetectiontask,wereportprecisionandrecall(commonininformationretrieval)foreachpredictionmethod.LetDSbethenumberoftruesignpatchesdetectedwithDthetotalnumberofdetectionsandStheactualnumberoftruesignpatches.PrecisionisP=DS/D,thepercentageofdetectionsthatarecorrect(notthecomplementoffalsealarm).RecallisR=DS/S,thedetectionrate.TheharmonicmeanofrecallandprecisionF1=2PR/(P+R)re\rectsthebalance(orlackthereof)betweentherateandaccuracyofdetections;higherF1indicatesbetteroverallperformance.MAPandICMarepointestimatesoftheunwieldyjointposteriorprob-ability,butthemarginalposteriorofalabel(i.e.,\sign")atanodeisarealquantitythatmaybeeasilyvaried.Figure5(left)demonstratesthatoveralldiscriminationisverygoodeveninthecontext-freeMaxEntclassierwhen Figure6:Exampledetectionresults.Left-Right:MaxEnt,CRFICMandMAP.apatchcontainsnearlyallsign,butperformancedegradesastheamountofsigninapatchdecreases.Figure5(right)showsthataddingcontextwithaCRFimprovestheabilitytoidentifyallregionsofasign,especiallythoseontheborderwherethepatchcontainsmorebackground.UsingaCRFsignicantlyimprovesF1overthelocalMaxEntclassier.2TrainingwithTRPalsoimprovesrecall.Becauseitisgiventrueneighboringlabels,whichareunavailableattesttime,PLtrainingtendstobeovercon-dentwithedgeparameters,leadingtohigherprecision(exceptingMPM)asaresultofover-smoothingthelabels.TRPtrainingyieldshigherF1andrecalloverPLforallpredictionmethods.CONCLUSIONSconditionalrandomeldisapowerfulnewmodelforvisionapplica-tionsthatdoesnotrequirethestrongindependenceassumptionsofgenera-tivemodels.Withit,wedemonstratesigndetectioninnaturalimagesusingbothgeneraltexturefeaturesandspecialfeaturesfortext.Addingcontextincreasesthedetectionratefasterthanthefalsealarmratebydrawingonbothobserveddataandunknownlabelsfromneighboringregions.Thecomplexityissuesofcyclicrandomeldsarewellknown.Althoughtrainingtimesaregreater,predictionwithaCRFstillonlyrequiresabout3secondsona3GHzdesktopworkstation.Wehaveshownthesuperiorityoftreereparameterizationoverthepseudo-likelihoodapproximationforparam-eterestimationandpredictionintheCRFmodelforourdetectiontask.Weplantoaddmoreedgefeaturestoincreaseouruseofthemodel'scontextualpowerbyincorporatingfeatureselectionandinductionmethods.Overttingremainsaconstantprobleminsuchahigh-dimensionalmodel,soregularizationisanimportantareaforstudy.AcknowledgmentsThankstoAronCulotta,KhashayarRohanimanesh,andCharlesSuttonfortheirassistivediscussions.ThisworkwassupportedinpartbyNFSgrant#IIS-0100851,inpartbytheCenterforIntelligentInformationRetrieval,andinpartbyTheCentralIntelligenceAgency,theNationalSecurityAgencyandNationalScienceFoundationunderNSFgrant#IIS-0326249.2Claimsofrelativeperformancearebasedonatwo-sided,pairedsigntest(p4e 5). References[1]J.Besag,\Statisticalanalysisofnon-latticedata,"TheStatistician,vol.24,no.3,pp.179{195,1975.[2]S.ChenandR.Rosenfeld,\AGaussianpriorforsmoothingmaximumen-tropymodels,"Techn.ReportCMU-CS-99-108,CarnegieMellonUniver-sity,1999.[3]S.DellaPietra,V.DellaPietraandJ.Laerty,\InducingFeaturesofRan-domFields,"IEEETransactionsonPatternAnalysisandMachineIntelligence,vol.19,no.4,pp.380{393,1997.[4]J.GaoandJ.Yang,\AnAdaptiveAlgorithmforTextDetectionfromNaturalScenes,"inProceedingsofthe2001IEEEConferenceonComputerVisionandPatternRecognition,December2001,vol.2,pp.84{89.[5]C.GarciaandX.Apostolidis,\TextDetectionandSegmentationinComplexColorImages,"inProceedingsof2000IEEEInternationalConferenceonAcoustics,SpeechandSignalProcessing(ICASSP2000),June2000,vol.4,pp.2326{2330.[6]D.Grieg,B.PorteousandA.Seheult,\Exactmaximumaposterioriestimationforbinaryimages,"JournaloftheRoyalStatisticalSociety,vol.51,no.2,pp.271{279,1989.[7]A.JainandS.Bhattacharjee,\TextsegmentationusingGaborltersforauto-maticdocumentprocessing,"MachineVisionApplications,vol.5,pp.169{184,1992.[8]S.KumarandM.Hebert,\DiscriminativeRandomFields:ADiscrimina-tiveFrameworkforContextualInteractioninClassication,"inProc.2003IEEEInternationalConferenceonComputerVision(ICCV'03),2003,vol.2,pp.1150{1157.[9]J.Laerty,A.McCallumandF.Pereira,\ConditionalRandomFields:Prob-abilisticModelsforSegmentingandLabelingSequenceData,"inProc.18thInternationalConferenceonMachineLearning,MorganKaufmann,SanFrancisco,CA,2001,pp.282{289.[10]S.L.Lauritzen,GraphicalModels,no.17inOxfordStatisticalScienceSeries,Clarendon,1996.[11]H.Li,D.DoermannandO.Kia,\AutomaticTextDetectionandTrackinginDigitalVideo,"IEEETransactionsonImageProcessing,vol.9,no.1,pp.147{156,2000.[12]N.PetkovandP.Kruizinga,\Computationalmodelofvisualneuronsspe-cialisedinthedetectionofperiodandaperiodicorientedvisualstimuli:barandgratingcells,"BiologicalCybernetics,vol.76,pp.83{96,1997.[13]J.PortillaandE.P.Simoncelli,\Aparametrictexturemodelbasedonjointstatisticsofcomplexwaveletcoecients,"InternationalJournalofCom-puterVision,vol.40,no.1,pp.49{71,2000.[14]R.M.VictorWuandE.M.Riseman,\FindingTextinImages,"inDL'97:Proceedingsofthe2ndACMInternationalConferenceonDigitalLibraries,Images,andMultimedia,1997,pp.3{12.[15]M.Wainwright,T.JaakkolaandA.Willsky,\Tree-basedreparameterizationframeworkforanalysisofsum-productandrelatedalgorithms,"IEEETrans-actionsonImageProcessing,vol.49,no.5,pp.1120{1146,2003.[16]G.Winkler,ImageAnalysis,RandomFields,andMarkovChainMonteCarloMethods,Berlin:Springer-Verlag,2ndedn.,2003.