Anda di halaman 1dari 3

10/5/2016

Fundamentalsofvoxelization

[anerroroccurredwhileprocessingthisdirective]
"VolumeGraphics"
IEEEComputer,Vol.26,No.7
July1993
pp.5164
[anerroroccurredwhileprocessingthisdirective]
Sidebar:FundamentalsofVoxelization

Introduction
Anindispensablestageinvolumegraphicsisthesynthesisofvoxelrepresentedobjects.Thisstage,whichis
calledvoxelization,isconcernedwithconvertinggeometricobjectsfromtheircontinuousgeometric
representationintoasetofvoxelsthatbestapproximatesthecontinuousobject.Asthisprocessmimicsthe
scanconversionprocessthatpixelizes(rasterizes)2Dgeometricobjects,itisalsoreferredtoas3Dscan
conversion.In2Drasterizationthepixelsaredirectlydrawnontothescreentobevisualizedandfilteringis
appliedtoreducethealiasingartifacts.However,thevoxelizationprocessdoesnotrenderthevoxelsbut
merelygeneratesadatabaseofthediscretedigitizationofthecontinuousobject.
Intuitively,onewouldassumethatapropervoxelizationsimplyselectsallvoxelswhicharemet(ifonly
partially)bytheobjectbody.Althoughthisapproachcouldbesatisfactoryinsomecases,theobjectsit
generatesarecommonlytoocoarseandincludemorevoxelsthanarenecessary.Forexample,inFigure13a
2Dcurveisrasterizedintoaconnectedsequenceofpixels.Althoughthediscretecurvedoesnotcoverthe
entirecontinuouscurve,itisconnectedandconciselyandsuccessfullyseparatesbothsidesofthecurve.
Onepracticalmeaningofseparationisapparentwhenavoxelizedsceneisrenderedbycastingdiscreterays
fromtheimageplanetothescene.Thepenetrationofthebackgroundvoxels(whichsimulatethediscreteray
traversal)throughthevoxelizedsurfacecausestheappearanceofaholeinthefinalimageoftherendered
surface.Anothertypeoferrormightoccurwhena3Dfloodingalgorithmisemployedeithertofillanobject
ortomeasureitsavolume,surfacearea,orotherproperties.Inthiscasethenonseparabilityofthesurface
causesleakageofthefloodthroughthediscretesurface.
Unfortunately,theextensionofthe2Ddefinitionofseparationtothethirddimensionandtosurfacesisnot
straightforwardsincevoxelizedsurfacescannotbedefinedasanorderedsequenceofvoxelsandavoxelon
thesurfacedoesnothaveaspecificnumberofadjacentvoxels.Furthermore,thereareimportanttopological
issues,suchastheseparationofbothsidesofasurface,whichcannotbewelldefinedbyemploying2D
terminology.Thetheorythatdealswiththesetopologicalissuesiscalled3Ddiscretetopology.Wesketch
belowsomebasicnotionsandinformaldefinitionsusedinthisfield.

Figure13.Anexampleofa2D
discretecurve(shadedpixels)
thatintuitivelyseparatesitstwo
sidesevenwithoutcontainingall
thosepixelspiercedbythe
continuousline.
http://labs.cs.sunysb.edu/labs/projects/volume/Papers/Voxel/

Figure14.Thethreetypesofvoxeladjacenciesin3Ddiscrete
space:(1)thesixvoxelsthataare6adjacenttothevoxelat
thecenter(notseen),(2)theeighteenvoxelsthatare18
adjacenttothevoxelatthecenter,(3)thetwentysixvoxelsthat
are26adjacenttothevoxelatthecenter.
1/3

10/5/2016

Fundamentalsofvoxelization

Fundamentalsof3DDiscreteTopology
The3Ddiscretespaceisasetofintegralgridpointsin3DEuclideanspacedefinedbytheirCartesian
coordinates(x,y,z).Avoxelistheunitcubicvolumecenteredattheintegralgridpoint.Thevoxelvalueis
mappedinto{0,1}:thevoxelsassigned"1"arecalledthe"black"voxelsrepresentingopaqueobjects,and
thoseassigned"0"arethe"white"voxelsrepresentingthetransparentbackground.Outsidethescopeofthis
paperarenonbinaryapproacheswherethevoxelvalueismappedontotheinterval[0,1]representingeither
partialcoverage,variabledensities,orgradedopacities.Duetoitslargerdynamicrangeofvalues,this
approachmaysupporthigherqualityrendering.
Twovoxelsare26adjacentiftheyshareeitheravertex,anedge,oraface(seeFigure14).Everyvoxelhas
26suchadjacentvoxels:eightshareavertex(corner)withthecentervoxel(Figure14(1)),twelvesharean
edge(Figure14(2)),andsixshareaface(Figure14(3)).Accordingly,facesharingvoxelsaredefinedas6
adjacent,andedgesharingandfacesharingvoxelsaredefinedas18adjacent.
TheprefixNisusedtodefinetheadjacencyrelation,whereN=6,18,or26.Wesaythatasequenceof
voxelshavingthesamevalue(e.g.,"black")isanNpathifallconsecutivepairsareNadjacent.Asetof
voxelsAisNconnectedifthereisanNpathbetweeneverypairofvoxelsinA.AnNconnected
componentisamaximalNconnectedset.
Figure15showsa2Ddiscrete8connectedblackcurveandasequenceof8connectedwhitepixels(8
component)thatpassesfromonesideoftheblackcomponenttoitsothersidewithoutintersectingit.This
phenomenonisadiscretedisagreementwiththecontinuouscaseawherethereisnowayofpenetratinga
closedcurvewithoutintersectingit.Toavoidsuchscenario,ithasbeentheconventiontodefine"opposite"
typesofconnectivityforthewhiteandblacksets.Oppositetypesin2Dspaceare4and8,whilein3Dspace
6isoppositeto26orto18.

Figure15.Anexampleofa2D8
curvethatdoesnotseparateitstwo
sides.Thewhite8curvepenetrates
fromonesidetotheother.

Figure16.(1)a6connectedsurfacethatisnot6
separating.(2)a6connectedsurfacethatisnot18
separating.

Assumethatavoxelspace,denotedby ,includesonesubsetof"black"voxelsS.If SisnotN


connected,thatis, SconsistsofatleasttwowhiteNconnectedcomponents,thenSissaidtobeN
separatingin .Looselyspeaking,in2D,an8connectedblackpaththatdividesthewhitepixelsintotwo
groupsis4separatinganda4connectedblackpaththatdividesthewhitepixelsintotwogroupsis8
separating.Therearenoanalogousresultsin3Dspace.Forexample,Figure16showsa6connectedsetof
voxelsthatisnot6separating(Figure16(1))anda6connectedsetofvoxelsthatisnot18separating(Figure
16(2)).
LetAbeanNseparatingsurface.Avoxelp AissaidtobeanNsimplevoxelifApisstillNseparating.
AnNseparatingsurfaceiscalledNminimalifitdoesnotcontainanyNsimplevoxel.Acoverofa
continuoussurfaceisasetofvoxelssuchthateverypointofthecontinuoussurfaceliesinavoxelofthe
cover.Acoverissaidtobeaminimalcoverifnoneofitssubsetsisalsoacover.Thecoverpropertyis
http://labs.cs.sunysb.edu/labs/projects/volume/Papers/Voxel/

2/3

10/5/2016

Fundamentalsofvoxelization

essentialinapplicationsthatemployspacesubdivisionforfastraytracing[3].Thesubspaces(voxels)which
containobjectshavetobeidentifiedalongthetracedray.Notethatacoverisnotnecessarilyseparating,
whileontheotherhand,asmentionedabove,itmayincludesimplevoxels.Infact,evenaminimalcoveris
notnecessarilyNminimalforanyN.

VoxelizationAlgorithms
Inthepast,digitizationofsolidswasperformedbyspatialenumerationalgorithmswhichemploypointor
cellclassificationmethodsineitheranexhaustivefashionorbyrecursivesubdivision[7].However,
subdivisiontechniquesformodeldecompositionintorectangularsubspacesarecomputationallyexpensive
andthusinappropriateformediumorhighresolutiongrids.Instead,objectsshouldbedirectlyvoxelized,
preferablygeneratinganNseparating,Nminimal,andcoveringset,whereNisapplicationdependent.The
voxelizationalgorithmsshouldfollowthesameparadigmasthe2Dscanconversionalgorithmstheyshould
beincremental,accurate,usesimplearithmetic(preferablyintegeronly),andhaveacomplexitythatisnot
morethanlinearwiththenumberofvoxelsgenerated.
Theliteratureof3Dscanconversionisrelativelysmall.Danielsson[2]andMokrzycki[8]developedsimilar
3Dcurvealgorithmswherethecurveisdefinedbytheintersectionoftwoimplicitsurfaces.Voxelization
algorithmshavebeendevelopedfor3Dlines,3Dcircles,and.avarietyofsurfacesandsolids,including
polygons,polyhedra,andquadricobjects[4]Efficientalgorithmshavebeendevelopedforvoxelizing
polygonsusinganintegerbaseddecisionmechanismembeddedwithinascanlinefillingalgorithm[5],for
parametriccurves,surfaces,andvolumesusinganintegerbasedforwarddifferencingtechnique[6],andfor
quadricobjectssuchascylinders,spheres,andconesusing"weaving"algorithmsbywhichadiscrete
circle/linesweepsalongadiscretecircle/line[1].Figures29consistofavarietyofobjectsvoxelizedby
usingtheabovemethods.Thesepioneeringattemptsshouldnowbefollowedbyenhancedvoxelization
algorithmsthat,inadditiontobeingefficientandaccurate,willalsoadheretothetopologicalrequirementsof
separation,coverage,andminimality.

References
1.Cohen,D.andKaufman,A.,"ScanConversionAlgorithmsforLinearandQuadraticObjects",inVolume
Visualization,A.Kaufman,(ed.),IEEEComputerSocietyPress,LosAlamitos,CA,1990,280301.
2.Danielsson,P.E.,"IncrementalCurveGeneration",IEEETransactionsonComputers,C19,(1970),783
793.
3.Glassner,A.S.,"SpaceSubdivisionforFastRayTracing",IEEEComputerGraphicsandApplications,4,
10(October1984),1522.
4.Kaufman,A.andShimony,E.,'3DScanConversionAlgorithmsforVoxelBasedGraphics',Proc.ACM
WorkshoponInteractive3DGraphics,ChapelHill,NC,October1986,4576.
5.Kaufman,A.,"AnAlgorithmfor3DScanConversionofPolygons",Proc.EUROGRAPHICS'87,
Amsterdam,Netherlands,August1987,197208.
6.Kaufman,A.,"EfficientAlgorithmsfor3DScanConversionofParametricCurves,Surfaces,and
Volumes",ComputerGraphics,21,4(July1987),171179.
7.Lee,Y.T.andRequicha,A.A.G.,"AlgorithmsforComputingtheVolumeandOtherIntegralProperties
ofSolids:IKnownMethodsandOpenIssuesIIAFamilyofAlgorithmsBasedonRepresentation
ConversionandCellularApproximation",CommunicationsoftheACM,25,9(September1982),635650.
8.Mokrzycki,W.,"AlgorithmsofDiscretizationofAlgebraicSpatialCurvesonHomogeneousCubical
Grids",Computers&Graphics,12,3/4(1988),477487.
http://labs.cs.sunysb.edu/labs/projects/volume/Papers/Voxel/

3/3

Anda mungkin juga menyukai