Establishing Point Correspondence of 3D Faces Via Sparse Facial Deformable Model

Establishing Point Correspondence of 3D Faces Via Sparse Facial Deformable Model

Outline

IntroductionSparse facial deformable model Solving Shape constraint by face deformationCorrespondence constraint: patch-based

sparse representationExperiments Conclusions

Introduction

Recent progresses in 3D digital acquisition techniques allow 3D data to be accurately captured in real time. This otivates extensive researches on 3D data in computer vision and computer graphics communities

This paper aims at building an anthropometric dense correspondence between 3D faces. We assume that original 3D faces are represented as triangle meshes. Other forms of 3D faces can be easily changed to meshes

Introduction

Outline

IntroductionSparse facial deformable model Solving Shape constraint by face deformationCorrespondence constraint: patch-based

sparse representationExperiments Conclusions

Sparse Facial Deformable Model

Outline

IntroductionSparse facial deformable model Solving Shape constraint by face deformationCorrespondence constraint: patch-based

sparse representationExperiments Conclusions

Solving Shape Constraint By Face Deformation

Solving Shape Constraint By Face Deformation

Outline

IntroductionSparse facial deformable model Solving Shape constraint by face deformationCorrespondence constraint: patch-based

sparse representationExperiments Conclusions

Correspondence Constraint: Patch-BasedSparse Representation

Sparsity Threshold

-linear function

-exponential function

Correspondence Constraint: Patch-BasedSparse Representation

Outline

IntroductionSparse facial deformable model Solving Shape constraint by face deformationCorrespondence constraint: patch-based

sparse representationExperiments Conclusions

Experiments

Rn computes the average distance between the vertices on M’ and their nearest vertices on M, which exhibits the reconstruction accuracy.Rl f computes the average distance between the vertices on M’ and their corresponding vertices on Ml f , which implies the accuracy of overall correspondence.Rl computes the average distance between the landmarks on M and their corresponding landmarks on M’, which shows the accuracy of landmark correspondence.Dl shows the accuracy of the anthropometric correspondence from the view of face structure.

Experiments

Experiments

Experiments

Experiments

Experiments

Outline

IntroductionSparse facial deformable model Solving Shape constraint by face deformationCorrespondence constraint: patch-based

sparse representationExperiments Conclusions