by Sara Schvartzman
28 July 2016
Data Optimization for Facial Simulation
Siggraph Poster 2016
by Sara C. Schvartzman, Marco Romeo
Digital characters are common in modern films visual effects and the demand for digital actors has increased during the past few years. The success of digitally created actors is related to their believability and, in particular, the realism of the animation and simulation of their faces. Facial expressions in computer graphics are commonly obtained through linear vertex interpolation techniques such as blend shapes. These enable high artistic control and fast interaction, but cannot properly reproduce collisions or other physical phenomena such as gravity and inertia. These effects can be achieved by applying simulation techniques over the animated facial geometry (e.g. muscle simulation), but could potentially alter the look of the desired facial expression and produce inconsistencies with the work approved in animation. Moreover, animating such muscle rigs can be very cumbersome.
Figure 1: From left to right columns: two input target poses, their scene where the resulting deformable interacts with a rigid sphere.
In this poster, we propose an optimization algorithm that uses digitally sculpted blend targets to find the forces and material properties to produce a Finite Element Method (FEM) simulation that is faithful to the facial expressions produced by animators. Our work extends the results of [Bickel et al. 2009] by adding the optimization of forces, similar to what is described in [Skouras et al. 2013], but without relying on a library of predefined materials.
The result of our work will enable a creative workflow that lets artists sculpt the required shapes and create the main animation for the character. The system then replaces the animation with a simulated version that preserves their artistic decisions and enriches the results to convey more realism. Furthermore, this approach re- moves the need to create and setup complex muscle rigs, can be applied to fictitious creatures and is artistically controllable by nature. Finally, we believe it could be used to generate more target shapes, as a simulated sculpting tool.