Generating anatomical substructures for physically-based facial animation.

Abstract

Physically-based facial animation techniques are capable of producing realistic facial deformations, but have failed to find meaningful use outside the academic community because they are notoriously difficult to create, reuse, and art-direct, in comparison to other methods of facial animation. This thesis addresses these shortcomings and presents a series of methods for automatically generating a skull, the superficial musculoaponeurotic system (SMAS – a layer of fascia investing and interlinking the mimic muscle system), and mimic muscles for any given 3D face model. This is done toward (the goal of) a production-viable framework or rig-builder for physically-based facial animation. This workflow consists of three major steps. First, a generic skull is fitted to a given head model using thin-plate splines computed from the correspondence between landmarks placed on both models. Second, the SMAS is constructed as a variational implicit or radial basis function surface in the interface between the head model and the generic skull fitted to it. Lastly, muscle fibres are generated as boundary-value straightest geodesics, connecting muscle attachment regions defined on the surface of the SMAS. Each step of this workflow is developed with speed, realism and reusability in mind.