500 likes | 656 Views
Automatic Modeling of 3D Human Face. Supervisors :. Marco Andolfi. I. Ragnemalm Institute of Technology of Linkoping. M. Schaerf “La Sapienza ” University of Rome. Co-Supervisor :. M. Fratarcangeli “La Sapienza ” University of Rome. The problem : Face modeling.
E N D
Automatic Modeling of 3D Human Face Supervisors: Marco Andolfi I. Ragnemalm Institute of Technology of Linkoping M. Schaerf “La Sapienza” University of Rome Co-Supervisor: M. Fratarcangeli “La Sapienza” University of Rome
The problem: Facemodeling Modeling System Input Output Our goal
Our system:Input - Output - Features Modeling System Input Output 1 - FiveOrthogonalPhotos 3D colored model of the subject’s head in photo (no hair) 2 – A “Generic” Mesh 3 – A Landmark set points No particularinstrument
The Input:(2) – A “generic” Mesh Withoutexpression Without Gender Features Without Race Features
The Input:(3) – The landmark set ofpoints (almost) MPEG-4landmarkpoints
Inside the system Modeling System Coloring Sub-System Morphing Sub-System
Morphing sub system Morphing Sub-System MeshFeature Extraction Photo Feature Extraction Camera Error Correction 2D – 3D Conversion MorphingExecution
Meshfeatureextraction Manuallypicking Preprocessingoperation
Morphing sub system Morphing Sub-System MeshFeature Extraction Photo Feature Extraction Camera Error Correction 2D – 3D Conversion MorphingExecution
Photo featureextraction Manuallypicking Noisy Data
Morphing sub system Morphing Sub-System MeshFeature Extraction Photo Feature Extraction Camera Error Correction 2D – 3D Conversion MorphingExecution
Camera Errors: thecorrectedones Differentdistance Camera-subject Scaling
Camera Errors: thecorrectedones Differentdistance Camera-subject Scaling Camera orientation Translation
Camera Errors: thecorrectedones Differentdistance Camera-subject Scaling Camera orientation Translation Wrong Head Position: Zaxis Rotation
Camera Errors: thenegligibleones Wrong Head Position: Xaxis Rotation Angle?
Camera Errors: thenegligibleones Wrong Head Position: Xaxis Rotation Angle? Wrong Head Position: Yaxis Rotation No info
Camera Errors: thenegligibleones Wrong Head Position: Xaxis Rotation Angle? Wrong Head Position: Yaxis Rotation No info Prospective error Negligible
Camera Errors: Summary Corrected Ignored Differentdistance Camera-subject Wrong Head Position: Xaxis Rotation Angle? Scaling Camera orientation Wrong Head Position: Yaxis Rotation No info Translation Wrong Head Position: Zaxis Prospective error Rotation Negligible
Getreferencepoint: Why? Rotation Scaling
Getreferencepoint:Why? Translation
Getscalingfactor: example on Y Noisy Data Bad Result Fy Ly Fy ScalingFactor = ----------- Ly
Getscalingfactor: example on Y Fy1 Fy2 Ly1 Ly2 Fy1 + Fy2 + ... + FyN ScalingFactor = --------------------------------------- Ly1 + Ly2 + ... + LyN
Morphing sub system Morphing Sub-System MeshFeature Extraction Photo Feature Extraction Camera Error Correction 2D – 3D Conversion MorphingExecution
2D to 3D convertion Severalvaluesforeach coordinate WeightedAverage More importanceto front photo
Morphing sub system Morphing Sub-System MeshFeature Extraction Photo Feature Extraction Camera Error Correction 2D – 3D Conversion MorphingExecution
Single stepmorphing Noisy Data Interpolation RBF Morphingfunction Lowstiffnessparameter (asinterpolation) Bad Result
Twodifferentlevel of quality High levelqualitypoints Lowlevelqualitypoints
Doublestepmorphing Low and Highlevelqualitypoints RBF Morphingfunction High stiffnessparameter (asapproximation) First step c RBF Morphingfunction Lowstiffnessparameter (asinterpolation) Secondstep High levelqualitypoints
Inside the system Modeling System Coloring Sub-System Morphing Sub-System
Coloring sub-system Coloring Sub-System Texture coordinate computing Weight computing Bottomtexture generation Texturemapping Remove ghost effect Solve Overlapartifact
Texture coordinate computing Whatabout the cameracorrection error? Not a simpleorthogonalprojection
Coloring sub-system Coloring Sub-System Texture coordinate computing Weight computing Bottomtexture generation Texturemapping Remove ghost effect Solve Occlusion Problem
Weight computing γ β α Avoidingoverlapping Why? The weightastransparencylevel NT Weightforvertexrespect front photo NP Proportionaltoα NR NL WF = NF∙ NP Set NULL negative weights NF WR = NR∙ NP < 0 → WR = 0
Coloring sub-system Coloring Sub-System Texture coordinate computing Weight computing Bottomtexture generation Texturemapping Remove ghost effect Solve Overlappartifact
Bottomtexture generation AIM: obtain a texturecontainigonlyskin Little squareonlyskin Repetition of the little square Final bottomtexture No mole, Noscar Bad quality? (toomanysquares...)
Coloring sub-system Coloring Sub-System Texture coordinate computing Weight computing Bottomtexture generation Texturemapping Remove ghost effect Solve Overlappartifact
TextureMapping The weightastransparencylevel
Coloring sub-system Coloring Sub-System Texture coordinate computing Weight computing Bottomtexture generation Texturemapping Remove ghost effect Solve Overlappartifact
Coloring sub-system Coloring Sub-System Texture coordinate computing Weight computing Bottomtexture generation Texturemapping Remove ghost effect Solve Overlappartifact
Solvingocclusionproblem Detectvertexbetweeninnereyepoints Givingweightto side photos Removingweightto side photos