Fully Automatic Generation of Anatomical Face Simulation Models

How to read this

Category

Method: automatic anatomical face-model generation

Contributions

• A fast, fully automatic method to create simulatable facial models from only a target surface mesh
• Automatic detection of 17 landmarks and feature curves plus Poisson-equation-based morphing of a high-fidelity anatomical template (skeleton and 41 facial muscles) to the target
• Resulting models carry complete internal anatomy and can be simulated to generate a wide range of expressions

Context

Builds on anatomical/physics-based facial simulation (Sifakis et al., automatic determination of facial muscle activations), automating the otherwise manual construction of the flesh-and-muscle model.Builds on: Automatic Determination of Facial Muscle Activations from Sparse Motion Capture Marker Data

Correctness

Relies on a single high-fidelity template morphed to each target, so fidelity for human and humanoid faces depends on landmark/curve detection and how well target anatomy matches the template; faces far from the template's anatomy may be approximated rather than truly individualized.

Clarity

Clear for a simulation audience; a first pass conveys the template-morph pipeline, a second covers the Poisson morphing and muscle setup.

How to read it

Focus on the landmark/feature-curve detection and the Poisson-based template morph; second pass on the anatomical template and simulation step if you intend to drive expressions.