Efficient, Robust and Accurate Fitting of a 3D Morphable Model

《Efficient, Robust and Accurate Fitting of a 3D Morphable Model》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、Efficient,RobustandAccurateFittingofa3DMorphableModelSamiRomdhaniThomasVetterUniversityofBasel,ComputerScienceDepartmentBernoullistrasse16,CH-4056Basel,Switzerlandsami.romdhani,thomas.vetter
@unibas.chAbstractRobustagainstnon-Gaussiannoise.Theassumptionofnormalityofthedifferencebetwe

2、entheimagesyn-3DMorphableModels,asameanstogenerateimagesthesizedbythemodelandtheinputimageisgenerallyofaclassofobjectsandtoanalyzethem,havebecomein-violatedduetothepresenceofaccessoriesorartifactscreasinglypopular.Theproblematicpartofthisframework(glasses,hair,specularhighlight).isthe

3、registrationofthemodeltoanimage,a.k.a.thefitting.Thecharacteristicfeaturesofafittingalgorithmareitsef-ficiency,robustness,accuracyandautomation.Manyaccu-Accurate,aswehavealreadypointedout,theaccu-ratealgorithmsbasedongradientdescenttechniquesexistracyiscrucialfortheapplicationwhichistou

4、sethewhichareunfortunatelyshortontheotherfeatures.Re-fittingresults(and,generally,thelevelofaccuracyre-cently,anefficientalgorithmcalledInverseCompositionalquireddependsthereon.)ImageAlignment(ICIA)algorithm,abletofit2Dimages,wasintroduced.Inthispaper,weextentthisalgorithmtofit3DMorphable

5、ModelsusinganovelmathematicalnotationAutomatic:Thefittingshouldrequireaslittlehumanwhichfacilitatestheformulationofthefittingproblem.Thisinterventionaspossible,optimally,noinitialization.formulationenablesustoavoidasimplificationsofarusedintheICIA,beingasefficientandleadingtoimprovedfit-t

6、ingprecision.Additionally,thealgorithmisrobustwithoutAnalgorithmcapableofanyofthefouraforementionedsacrificingitsefficiencyandaccuracy,therebyconformingfeaturesisdifficulttosetup.Analgorithmcapableofalltothreeofthefourcharacteristicsofagoodfittingalgo-fourfeaturesistheholygrailofmodelbase

7、dcomputervi-rithm.sion.Accuratealgorithmsexistalreadysinceafewyears(e.g.[5])buttheyareshortonthethreeotherfeatures.Re-cently,BakerandMatthews[2]presentedanefficiental-gorithmforfitting2Dcorrespondence-basedmodels,called1.IntroductionInverseCompositionalImageAlignment(ICIA).Theyde-riveda

8、firsto