Image-Based Modeling and Rendering Techniques A Su

Image-BasedModelingandRenderingTechniques:ASurveyManuelM.Oliveira1Resumo:Arecenteconvergênciaentrecomputaçãográficaevisãocomputacionallevouàcriaçãodeumconjuntodetécnicasdenominadasmodelagemerenderingbaseadosemimagens(MRBI).Modelagembaseadaemimagensserefereaoprocessodeutilizaçãodeimagensparareconstruçãodemodelosgeométricos3D.Renderingbaseadoemimagensbuscaaobtençãodeumaltograuderealismo,reduçãodotempodeprocessamentoesimplificaçãodatarefademodelagem,atravésdousodeimagenscomoprimitivasdemodelagemerendering.Taisabordagenstemopotencialdesimplificararepresentaçãodecenascomplexasepermitiraobtençãodeefeitossubtospresentesemcenasreais,masdedifícilreproduçãopormeiodetécnicasconvencionais.Esteartigoapresentaumlevantamentosistemáticodoestado-da-arteemMRBI,discutindoosprincípiosbásicosdecadatécnica,suasvantangenselimitações.Palavras-Chave:ModelagemeRenderingBaseadosemImages,ComputaçãoGráfica,Realismo.Abstract:Arecentconvergenceofcomputergraphicsandcomputervisionhasproducedasetoftechniquescollectivelyknownasimage-basedmodelingandrendering(IBMR).Image-basedmodelingreferstotheuseofimagestodrivethereconstructionof3Dmodels.Inthecaseofimage-basedrendering,thegoalistoachievemorerealisticandfasterrenderingsandtosimplifythemodelingtaskbyusingimages,asopposedtopolygons,asbothmodelingandrenderingprimitives.Image-basedapproachescanpotentiallyeliminatethelabor-intensivetaskusuallyrequiredformodelingdetailedgeometricstructures.Theycanalsohandlesubtlereal-worldeffectscapturedbyimages,butdifficulttoreproducewithconventionalgraphicstechniques.Thispapersurveysthestate-of-the-artinimage-basedmodelingandrenderingtechniques,discussingthefundamentalprinciplesbehindeachtechnique,theirstrengthsandlimitations.Keywords:Image-BasedModelingandRendering,ComputerGraphics,Realism.1InstitutodeInformática,UFRGS,CaixaPostal15064,CEP91501-970,PortoAlegre,RS,Brasil{[e-mail]}Image-BasedModelingandRenderingTechniques:ASurvey38RITA•VolumeIX•Número2•20021IntroductionInrecentyears,image-basedmodelingandrendering(IBMR)techniqueshavegainedconsiderableattentioninthegraphicscommunitybecauseoftheirpotentialtocreateveryrealisticimages.Oneofthemajorbenefitsofthesetechniquesistheabilitytocapturesubtlereal-worldeffectsanddetailsrelatedtotheimperfectionsoftherealworldthatgraphicsresearchersstilldonotknowhowtomodelandrender[12].Byusingimagesasbothmodelingandrenderingprimitives,image-basedapproachescanhelptoalleviatetwoimportantandlongstandingproblemsincomputergraphics:theneedforsimplermodelingtechniquessuitableforrepresentingcomplexscenes,andtheeverneedforrenderingacceleration.Theformercanbeachievedbyreplacingconventional(geometric)modelswithimage-basedrepresentations.Renderingspeedupsareobtainedbydetachingrenderingtimefromscenecomplexity,andbyre-samplingpre-shadedimages.Image-basedrendering(IBR)usesimages,asopposedtopolygons,asmodelingandrenderingprimitives.Inpractice,manyIBRapproachescorrespondtoimage-geometryhybrids,withthecorrespondingamountofgeometryrangingfromper-pixeldepthtohundredsofpolygons.Image-basedmodeling(IBM),ontheotherhand,referstotheuseofimagestodrivethereconstructionofthree-dimensionalgeometricmodels.Despitetheirpotential,IBMRtechniquesarestillintheirinfancyandseveralchallengesstillneedtobeovercome.Thistutorialsurveysthestate-of-the-artinimage-basedmodelingandrenderingtechniques,discussingtheirstrengthsandlimitations,andenumeratingsomepossibleresearchopportunities.ThestructureofthesurveyisorganizedaroundthediagrampresentedinFigure1.Thetechniquesarefirstclassifiedasimage-basedrendering(IBR)orimage-basedmodeling(IBM)andfurtherrefinedaccordingtotheirrelativepositionsalongtheimage-geometryspectrum(indicatedbyadot).Thus,forinstance,pureimage-basedapproaches(i.e.,pureIBRandpureIBM)fallontheleftpartofthespectrum,whereashybridtechniquesarepositionedaccordingtotheamountofgeometricinformationrequired.ThegoalofprovidinganintuitivedescriptionofthefundamentalideasbehindmostcurrentIBRandIBMtechniquesinasingledocumentisveryambitious.Itshopeistoofferausefulreferenceforstudentsandresearchersinterestedinasolidintroductiontotheunderlyingprinciplesofthisyoungfield.Unfortunately,duetospacelimitations,sometechniqueshadtobeleftoutinordertoguaranteepropercoverageofthematerialtreatedhere.Thechoiceofthepresentedtechniquestookintoaccounttheirfitnessinaprogressionthathelpsthereadertograduallyassimilateconceptsandunderstandthefieldevolution.Thisbynomeansimplythelackofimportanceoftheuncoveredmethods.TechniquessuchasImage-BasedVisualHulls[24,47],SurfaceLightFields[45],LightFieldMapping[8]andVoxelColoring[48]deserveproperrecognitionbutcouldnotbeincludedinthissurveyduetothelackofspace.Theyappear,however,inthediagramshowninFigure1.Image-BasedModelingandRenderingTechniques:ASurveyRITA•VolumeIX•Número2•2002392PureImage-BasedRenderingTechniquesPureIBRtechniquesutilizeonlyintensityimages2ofthescenesforrenderingnovelviews.Samplesoftheenvironmentarecapturedassetsofphotographs(orvideossequences)andresampledduringrenderingtime.Inthiscase,no3Dmodelingisrequiredandtherenderingspeedisn