We argue that de64257ning a modeling operation by asking for rigidity of the local transformations is useful in various settings Such formulation leads to a nonlinear yet conceptually simple energy formulation which is to be minimized by the deforme ID: 3125
Download Pdf The PPT/PDF document "To appear at the Eurographics Symposium ..." 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.
O.Sorkine&M.Alexa/As-Rigid-As-PossibleSurfaceModeling originalmeshuniformweightscotanweightsuniformweightscotanweightsFigure2:Demonstrationoftheimportanceofproperedgeweightingintheenergyformulation( 3 ).Deformationusinguniformweighting(wij=1)leadstoasymmetricalresults,whereascotangentweightingenablestoeliminatetheinuenceofthemeshingbias. LetusdenotebySithecovariancematrixSi=åj2N(i)wijeije0Tij=PiDiP0Ti;(5)whereDiisadiagonalmatrixcontainingtheweightswij,Piisthe3jN(vi)jcontainingeij'sasitscolumns,andsimilarlyforP0i.ItiswellknownthattherotationmatrixRimaximizingTr(RiSi)isobtainedwhenRiSiissymmetricpositivesemi-denite(ifMisapsdmatrixthenforanyor-thogonalR,Tr(M)Tr(RM)).OnecanderiveRifromthesingularvaluedecompositionofSi=UiSiVTi:Ri=ViUTi;(6)uptochangingthesignofthecolumnofUicorrespondingtothesmallestsingularvalue,suchthatdet(Ri)0.2.2.ThelocalrigidityenergyOursimpleideaformeasuringtherigidityofadeformationofthewholemeshistosumupoverthedeviationsfromrigiditypercell,asexpressedby( 3 ).Thus,weobtainthefollowingenergyfunctional:ES0=nåi=1wiE(Ci;C0i)= (7) =nåi=1wiåj2N(i)wij p0ip0jRi(pipj) 2;wherewi,wijaresomexedcellandedgeweights.NotethatES0dependssolelyonthegeometriesofS,S0,i.e.,onthevertexpositionsp;p0.Inparticular,sincethereferencemesh(ourinputshape)isxed,theonlyvariablesinES0arethedeformedvertexpositionsp0i.ThisisbecausetheoptimalrotationsRiarewell-denedfunctionsofp0,aswasshownintheprevioussection.Thechoiceofper-edgeweightswijandper-cellweightswiisimportantformakingourdeformationenergyasmesh-independentaspossible,asdemonstratedinFigure 2 .Theweightsshouldcompensatefornon-uniformlyshapedcells andpreventdiscretizationbias.Wethereforeusethecotan-gentweightformulaforwij[ PP93 , MDSB03 ]:wij=1 2cotaij+cotbij;whereaij;bijaretheanglesoppositeofthemeshedge(i;j)(foraboundaryedge,onlyonesuchangleexists).Wefur-thernotethatthedeviationfromrigidity,asdenedby( 3 ),isanintegratedquantity,sothatthecellenergyispropor-tionaltothecellarea,andwecansetwi=1.Analternativeexplanationforthiswouldbeusingthearea-correctededgeweightsw0ij=(1=Ai)wij,whereAiistheVoronoiareaofcellCi[ MDSB03 ],andthenalsosettingthecellweightstobetheVoronoiarea:w0i=Ai.Theareatermsimplycancelsout,andweareleftwiththesymmetriccotangentweightswij.3.ModelingframeworkInamodelingframeworkweneedtosolveforpositionsp0ofS0thatminimizeE(S0),undersomeuser-denedmodel-ingconstraints.Thismeanswedonotknowtherigidtrans-formationsfRigaprioriandthusneedtosolveforthemaswell.Therefore,werstinterpretE(S0)asafunctionofp0andfRigandinanymodelingsituationweseekthemini-mumenergyunderthevariationinbothsets.Tosolveforthenextlocalminimumenergystate(start-ingfromagiveninitialvectorofpositionsandrotations),weproposetouseasimplealternatingminimizationstrat-egy.Thismeans,foragivenxedsetofrigidtransforma-tions,wendpositionsp0thatminimizeE(S0).Then,wendtherigidtransformationsfRigthatminimizeE(S0)forthegivensetofpositionsp0.Wecontinuetheseinterleavediterationsuntilthelocalenergyminimumisreached.LetusrstlookhowtondoptimalrigidtransformationsfRigforagivensetofmodiedpositionsp0.Eachterminthesum( 7 )involvesonlytheper-cellrigidtransformationRi,i.e.,wecancomputeanoptimaltransformationforeachcellwithoutregardfortheothercellsandtheirrigidtrans-formations.Thus,weseekanRithatminimizesthepercellenergyin( 3 ).Thesolutiontothis,however,isdetailedinSection 2.1 ,namely,Equation( 6 ). c TheEurographicsAssociation2007. O.Sorkine&M.Alexa/As-Rigid-As-PossibleSurfaceModeling initialguess1iteration2iterationsinitialguess1iteration4iterationsFigure3:Successiveiterationsoftheas-rigid-as-possibleeditingmethod.TheinitialguessisthenaiveLaplacianeditingresult(asin[ LSCO04 ]butwithoutanylocalrotationestimation).TheoriginalstraightbarmodelisshowninFigure 6 . Inordertocomputeoptimalvertexpositionsfromgivenrotations,wecomputethegradientofE(S0)withrespecttothepositionsp0.Letuscomputethepartialderivativesw.r.t.p0i.NotethattheonlytermsinE(S0)whosederivativedoesnotvanisharethoseinvolvingverticesiandj2N(i):¶E(S0) ¶p0i=¶ ¶p0i åj2N(i)wij p0ip0jRi(pipj) 2++åj2N(i)wji p0jp0iRj(pjpi) 2!==åj2N(i)2wijp0ip0jRipipj++åj2N(i)2wjip0jp0iRjpjpi:Usingthefactthatwij=wji;wearriveat¶E(S0) ¶p0i=åj2N(i)4wijp0ip0j1 2(Ri+Rj)pipj:Settingthepartialderivativestozerow.r.t.eachp0iwearriveatthefollowingsparselinearsystemofequations:åj2N(i)wijp0ip0j=åj2N(i)wij 2Ri+Rjpipj:(8)Thelinearcombinationontheleft-handsideisnon-otherthanthediscreteLaplace-Beltramioperatorappliedtop0;thesystemofequationscanbecompactlywrittenasLp0=b;(9)wherebisann-vectorwhoseithrowcontainstheright-handsideexpressionfrom( 8 ).Wealsoneedtoincorporatethemodelingconstraintsintothissystem.Inthesimplestform,thosecanbeexpressedbysomexedpositionsp0j=ck;k2F;(10)whereFisthesetofindicesoftheconstrainedvertices.Theseconsistofstaticandhandlevertices,interactivelyma-nipulatedbytheuser.Incorporatingsuchconstraintsinto( 9 )simplymeanssubstitutingthecorrespondingvariables,ef-fectivelyerasingrespectiverowsandcolumnsfromLandupdatingtheright-handsidewiththevaluesck.NotethattherigidtransformationsfRigonlyinuencetheright-handsideofthesystem,whereasthesystemma-trixonlydependsontheinitialmesh.Thus,wecanemploy adirectsolver,andthesystemmatrixhastobefactoredonlyonceforminimizingE(S0).Moreover,sincep0consistsofthreecolumns(forthethreecoordinatefunctions),weonlyneedtoperformthreetimesback-substitutiontosolveforeachcoordinate,usingthesamennfactorization.SinceLissymmetricpositivedenite,thesparseCholeskyfac-torizationwithll-reducingreorderingisanefcientsolverchoice[ Tol03 ].Tosummarize,theoverallminimizationofE(S0)pro-ceedsasfollows.Firstlythecoefcientswijareprecomputedandthesystemmatrixof( 9 )ispre-factored.Givenaninitialguessp00,thelocalrotationsRiareestimated,asdescribedinSection 2.1 .Newpositionsp01areobtainedbysolving( 9 ),pluggingRiintotheright-handside.Thenfurtherminimiza-tionisperformedbyre-computinglocalrotationsandusingthemtodeneanewrighthand-sideforthelinearsystem,andsoon.Thisleadstoanefcientsolutionofthenon-linearproblemathand,sinceonlyback-substitutionsarenecessary.4.ResultsanddiscussionWehaveimplementedtheas-rigid-as-possibledeformationtechniqueusingC++onaPentium42.16GHzlaptopwith2GBRAM.WeusedthesparseCholeskysolverprovidedwiththeTAUCSlibrary[ Tol03 ]andstandardSVDimple-mentation(usedforpolardecompositionof33matrices)from[ PTVF92 ].WepresentsometypicaldeformationresultsobtainedwithourtechniqueinFigures 1 , 4 8 .Notethatnaturalde-formationsareobtained,evenwhenthemanipulationhandleisonlybeingtranslated,becausetheoptimizationautomati-callyproducesthecorrectlocalrotations.TheCactus(Fig-ure 7 )isaparticularlychallengingexample,especiallyforlinearvariationaldeformationmethods,duetoitslongpro-trudingfeatures[ BS07 ].TheresultsofourmethodcanbecomparedwithPRIMO[ BPGK06 ],astate-of-the-artnon-lineartechnique,aswellasvariouslinearvariationaltechniques,byobserv-ingthecanonicalexamplesinFigures 5 , 6 , 7 .Suchde-formationsappearinthecomparisontablein[ BS07 ];itisevidentthatourmethodperformsequallywelltoPRIMOandisgenerallysuperiortolinearmethods,especiallywhenhandletranslationisinvolved.Toemphasizethispoint,wecomparetheresultsofourmethodwithPoissonmeshedit-ing[ YZX04 , ZRKS05 ]inFigure 5 .Notethatsincethe c TheEurographicsAssociation2007. O.Sorkine&M.Alexa/As-Rigid-As-PossibleSurfaceModeling Figure4:Largedeformationofthespikyplane(seeFigure 2 fortheoriginalmesh).Notethatthehandle(inyellow)wasonlytranslated,withoutspecifyinganyrotation. PoissonourmethodPoissonourmethodFigure5:ComparisonwithPoissonmeshediting.Theorig-inalmodelsappearinFigures 2 and 7 .Theyellowhan-dlewasonlytranslated;thisposesaproblemforrotation-propagationmethodssuchas[ YZX04 , ZRKS05 , LSLCO05 ]. handlewasonlytranslated,Poissoneditingcannotgener-ateaproperrotationeld(sincethereisnohandlerota-tiontopropagate),whichresultsindetaildistortionandlackofsmoothnessneartheconstraints.Thesametranslation-insensitivitywouldbeobservedinthemethodofLipmanetal.[ LCOGL07 ];ourtechniquehandlestranslationwellbyoptimizingforthelocalrotations,atthepriceofaglobalnon-linearoptimization.Itisworthnotingthoughthattherequirednumericalmachineryandthesetupofthelinearsys-temisalmostidenticaltothelinearvariationalmethods.Theaccompanyingvideoshowsseveralshorteditingses-sionscapturedlive.Ourunoptimizedcoderunsinterac-tively(at10-30fps)forregionsofinterest(ROI)ofupto10Kvertices,using2-3iterationsperedit.Anumberofimprovementsarepossibletospeedupconvergence:afasterpolardecompositionroutine(i.e.,onethatreusespre-viousframecomputationsratherthanstartingfromscratcheachtime)andamultiresolutiontechnique,suchastheonein[ BPGK06 ]or[ HSL06 ],toallowtheoptimizationtorunonacoarseversionofthemeshinordertoquicklypropagatethedeformationacrosstheROI.Animportantimplementationissueistheinitialguesswhichstartstheoptimization;sincetheenergyweminimizeisnon-linear,multiplelocalminimamayexist,andtheso-lutiondependsontheinitialguessinsuchcase.Itisimpor-tanttouseareasonable-qualityinitialguess(i.e.,nottoofarfromtheinitialshapeandtheintuitivelyexpectedresult)toallowquickconvergence,yetitisdesirabletocomputeitquickly.Weexperimentedwithseveralpossibilities,whichcanbeusedindifferentscenarios:Previousframe(forinteractivemanipulation):Iftheuserinteractivelymanipulatesthecontrolhandle(s),itisreason-abletousetheresultofthepreviousframeastheinitial Figure6:Twistandrotationdeformations. ModelFigureRelativeRMSerror DinoFig. 1 0.024 SpikyplaneFig. 4 left0.034 SpikyplaneFig. 4 right0.016 TwistedbarFig. 6 left0.095 ArmadilloFig. 8 (b)0.037 ArmadilloFig. 8 (c)0.013 ArmadilloFig. 8 (e)0.051 Table1:RelativeRMSerrorofedgelengthsforvariousde-formations.Whenthemodelingconstraintsdonotnecessi-tatestretching,theerrorisverylow.Thetwistexampledoesinvolvesomeslightstretchingbecausethetopofthebarisconstrainedtoremainatthesameheight,hencethehigherrelativeerrorinthiscase. guess,sincethehandlemovementand/ordeformationisex-pectedtobecontinuous.Therefore,inthiscasewesimplytakethepreviousframeandassigntheuser-denedpositionstotheconstrainedvertices.Thisapproachwasusedforalltheguresinthispaper,unlessexplicitlymentionedother-wise,andisalsodemonstratedintheaccompanyingvideo.Theuserexperienceremindsalotofinteractingwithphysi-calmaterial.NaiveLaplacianediting:ThestartingguessisobtainedbysimplelinearminimizationofkLp0dk2undertheposi-tionalmodelingconstrains( 10 ),whered=Lparethedif-ferentialcoordinatesoftheinputmesh.Althoughthisguessproducesdistortedresultsforlargedeformations,thesubse-quentiterationsmanagetorecover,asdemonstratedinFig-ure 3 .Forsignicantlydistortedinitialguesstheconver-gencemaybeslow,however.Rotation-propagation:Ifthemanipulationofthehandlein-volvesexplicitrotation(alongwithtranslation),onecanuseanyofthetechniquesthatexplicitlypropagatethespeciedrotationtotheunconstrainedregions,suchas[ LSLCO05 , ZRKS05 , LCOGL07 ].Subsequentoptimizationofouren-ergyallowstoconsolidatetheotherwisedecoupledrotationandtranslationandimprovestheresults;convergenceistyp-icallyveryfastsincethestartingrotationalcomponentofthedeformationisalreadygood.Aninterestingpropertyofouras-rigid-as-possiblesur-facedeformationisedgelengthpreservation,totheextentallowedbythemodelingconstraints.Ifthemodelingcon-straintsdonotimposestretchingonthesurface,theopti-mizationalwaysstrivestoconvergetoastatewheretheedgelengtherrorissmall.Thisisclearlyvisibleinthedeforming c TheEurographicsAssociation2007. O.Sorkine&M.Alexa/As-Rigid-As-PossibleSurfaceModeling (a)(b)(c)(d)(e)(f)Figure7:BendingtheCactus.(a)istheoriginalmodel;yellowhandlesaretranslatedtoyieldtheresults(b-f).(d)and(e)showsideandfrontviewsofforwardbending,respectively.Notethatin(b-e)asinglevertexatthetipoftheCactusservesasthehandle,andthebendingistheresultoftranslatingthatvertex,norotationconstraintsaregiven. (a)(b)(c)(d)(e)(f)Figure8:EditingtheArmadillo.(a)and(d)showviewsoftheoriginalmodel;therestoftheimagesdisplayeditingresults,withthestaticandhandleanchorsdenotedinredandyellow,respectively. planeexample(Figure 4 ),forinstance,whichbehavessim-ilartorubber-likematerial.Table 1 summarizesroot-mean-squareedgelengtherrormeasurementsforseveraldeforma-tionspresentedinthispaper;itcanbeseenthattherelativeRMSerrorisverylow.5.ConclusionsTheimportantfeaturesofourapproachare(1)robustness,resultingfromtheminimizationprocedurethatisguaranteedtonotincreaseenergyineachstep;(2)simplicity,aseachstepoftheminimizationisconceptuallysimilartoLapla-cianmodeling;and(3)efciency,becausetheLaplacesys-temmatrixisconstantthroughouttheiterationsandhastobefactoredonlyonce.Wehavelearnedduringourexperimentsthatthiscom-binationisnotevident,i.e.,simplyupdatingtheright-handsideofadiscreteLaplacesysteminaseeminglyreasonablewaywouldfailtoconvergeinalmostallcases.Convergenceinourapproachistheresultofderivinganenergythatcan-notincreaseineachstepoftheiterations.Notethattheo-retically,thelocalminimumfoundbydecreasingtheenergymightnotbeunique,i.e.,therecouldbeaconnectedsetofminimumenergystates.However,wehavenotexperiencedthisproblemandbelievethatifitexistsatallthenonlyforparticularlyderivedexamples.Thefactthateachstepintheiterationscanbeperformedbysolvingalinearsystemwithaconstantmatrixthroughout theminimizationprocedurereallyistheresultofacarefuldesignoftheenergyfunctional.Thenumberofiterationsrequiredtogetreasonablyclosetoaminimumdependsontheconditionnumberofthe(anchored)Laplacianmatrix,whichisgenerallyproportionaltothemeshsize.Speci-cally,ifwekeeptheboundaryconditionsthesameandre-nethemesh,theconditionnumberwillgrowproportion-ally,eveniftheshapeofthemeshelementsisperfect(fordetailedanalysisandboundsontheconditionnumberoftheuniformanchoredLaplacianmatrix,see[ CCOST05 ];theuniformLaplaciancoincideswiththecotangentLaplacianfortessellationswithequilateraltriangles,andinothercasestheboundsforthecotangentLaplacianareprobablymorepessimistic).Thismeansasthemeshesarerenedstabilitydeteriorates,andtypicallymoreiterationsareneededuntilconvergence(inadditiontothefactthateachiterationbe-comesmorecostly).Thispracticalefciencyproblemcouldbeeasilyalleviatedwithmulti-resolutiontechniques.Anotherinterestingqualityofourapproachisthatittriv-iallyextendstovolumetriccells,e.g.,tetrahedra.Astherigidityismeasuredbasedontheedgesineachcell,nothingwouldhavetobechangedinthesetupoftheenergyonewouldonlyhavetoplug-intheconnectivityofavolumet-ricgrid.So,ifpreservationofvolumeisofconcernratherthanpreservationofsurface,thenthiscouldeasilybeac-complished.Ofcourse,aswithotherrecentapproaches,theoptimizationcouldbeappliedtoacoarsevolumetricgrid c TheEurographicsAssociation2007. O.Sorkine&M.Alexa/As-Rigid-As-PossibleSurfaceModeling whichcontrolstheshapeembeddedinit,ratherthandirectlytothediscretesurfaceorvolume.Infuturework,wewishtoexperimentwithseveraldegreesoffreedomthatourmodelingframeworkoffers:changingthesizeandrelativeweightspercell,soastocon-troltheoverallandrelativelocalrigidityofthesurface.AcknowledgementWewishtothankMarioBotschandLeifKobbeltforin-sightfuldiscussionsandtheanonymousreviewersfortheirvaluablecomments.ThisworkwassupportedinpartbytheAlexandervonHumboldtFoundation.References [ACOL00] ALEXAM.,COHEN-ORD.,LEVIND.:As-rigid-as-possibleshapeinterpolation.InProceedingsofACMSIGGRAPH(2000),pp.157164. [Ale01] ALEXAM.:Localcontrolformeshmorphing.InPro-ceedingsofSMI(2001),pp.209215. [ATLF06] AUO.K.-C.,TAIC.-L.,LIUL.,FUH.:DualLapla-cianeditingformeshes.IEEETVCG12,3(2006),386395. [BK04] BOTSCHM.,KOBBELTL.:Anintuitiveframeworkforreal-timefreeformmodeling.ACMTOG23,3(2004),630634. [BPGK06] BOTSCHM.,PAULYM.,GROSSM.,KOBBELTL.:PriMo:Coupledprismsforintuitivesurfacemodeling.InPro-ceedingsofSGP(2006),pp.1120. [BS07] BOTSCHM.,SORKINEO.:Onlinearvariationalsurfacedeformationmethods.IEEETVCG(2007).Toappear. [BSPG06] BOTSCHM.,SUMNERR.,PAULYM.,GROSSM.:Deformationtransferfordetail-preservingsurfaceediting.InProceedingsofVMV(2006),pp.357364. [CCOST05] CHEND.,COHEN-ORD.,SORKINEO.,TOLEDOS.:Algebraicanalysisofhigh-passquantization.ACMTOG24,4(2005),12591282. [GSS99] GUSKOVI.,SWELDENSW.,SCHRÖDERP.:Multires-olutionsignalprocessingformeshes.InProceedingsofACMSIGGRAPH(1999),pp.325334. [Hor87] HORNB.K.P.:Closed-formsolutionofabsoluteorien-tationusingunitquaternions.JournaloftheOpticalSocietyofAmerica4,4(1987). [HSL06] HUANGJ.,SHIX.,LIUX.,ZHOUK.,WEIL.-Y.,TENGS.,BAOH.,GUOB.,SHUMH.-Y.:Subspacegradientdomainmeshdeformation.ACMTOG25,3(2006),11261134. [IMH05] IGARASHIT.,MOSCOVICHT.,HUGHESJ.F.:As-rigid-as-possibleshapemanipulation.ACMTOG24,3(2005),11341141. [JSW05] JUT.,SCHAEFERS.,WARRENJ.:Meanvaluecoor-dinatesforclosedtriangularmeshes.ACMTOG24,3(2005),561566. [KCVS98] KOBBELTL.,CAMPAGNAS.,VORSATZJ.,SEI-DELH.-P.:Interactivemulti-resolutionmodelingonarbitrarymeshes.InProceedingsofACMSIGGRAPH(1998),ACMPress,pp.105114. [KS06] KRAEVOYV.,SHEFFERA.:Mean-valuegeometryen-coding.IJSM12,1(2006),2946. [LCOGL07] LIPMANY.,COHEN-ORD.,GALR.,LEVIND.:Volumeandshapepreservationviamovingframemanipulation.ACMTOG26,1(2007). [LSCO04] LIPMANY.,SORKINEO.,COHEN-ORD.,LEVIND.,RÖSSLC.,SEIDELH.-P.:Differentialcoordinatesforinter-activemeshediting.InProceedingsofSMI(2004),pp.181190. [LSLCO05] LIPMANY.,SORKINEO.,LEVIND.,COHEN-ORD.:Linearrotation-invariantcoordinatesformeshes.ACMTOG24,3(2005),479487. [MDSB03] MEYERM.,DESBRUNM.,SCHRÖDERP.,BARRA.H.:Discretedifferential-geometryoperatorsfortriangulated2-manifolds.InVisualizationandMathematicsIII,HegeH.-C.,PolthierK.,(Eds.).Springer-Verlag,Heidelberg,2003,pp.3557. [PP93] PINKALLU.,POLTHIERK.:Computingdiscreteminimalsurfacesandtheirconjugates.Experiment.Math.2,1(1993),1536. [PTVF92] PRESSW.H.,TEUKOLSKYS.A.,VETTERLINGW.T.,FLANNERYB.P.:NumericalRecipesinC:TheArtofScienticComputing.CambridgeUniversityPress,1992. [SLCO04] SORKINEO.,LIPMANY.,COHEN-ORD.,ALEXAM.,RÖSSLC.,SEIDELH.-P.:Laplaciansurfaceediting.InProceedingsofSGP(2004),pp.179188. [SMW06] SCHAEFERS.,MCPHAILT.,WARRENJ.:Imagede-formationusingmovingleastsquares.ACMTOG25,3(2006),533540. [Sor06] SORKINEO.:Differentialrepresentationsformeshpro-cessing.ComputerGraphicsForum25,4(2006),789807. [Tol03] TOLEDOS.:TAUCS:ALibraryofSparseLinearSolvers,version2.2.Tel-AvivUniversity,Availableonlineat http://www.tau.ac.il/~stoledo/taucs/ ,Sept.2003. [TPBF87] TERZOPOULOSD.,PLATTJ.,BARRA.,FLEISCHERK.:Elasticallydeformablemodels.InProceedingsofACMSIG-GRAPH(1987),pp.205214. [WBH07] WARDETZKYM.,BERGOUM.,HARMOND.,ZORIND.,GRINSPUNE.:Discretequadraticcurvatureenergies.CAGD(2007).Toappear. [XZWB05] XUD.,ZHANGH.,WANGQ.,BAOH.:Poissonshapeinterpolation.InProceedingsofSPM(2005),pp.267274. [YZX04] YUY.,ZHOUK.,XUD.,SHIX.,BAOH.,GUOB.,SHUMH.-Y.:MesheditingwithPoisson-basedgradienteldmanipulation.ACMTOG23,3(2004),644651. [ZRKS05] ZAYERR.,RÖSSLC.,KARNIZ.,SEIDELH.-P.:Harmonicguidanceforsurfacedeformation.InComputerGraphicsForum(ProceedingsofEurographics)(2005),pp.601609. [ZSS97] ZORIND.,SCHRÖDERP.,SWELDENSW.:Interac-tivemultiresolutionmeshediting.InProceedingsofACMSIG-GRAPH(1997),pp.259268. c TheEurographicsAssociation2007.