K.U.Leuven > ESAT > PSI > Visics > Research > Topics > Item 1.10

3D face expression

L. Van Gool

The human face is among the shapes that is the most difficult to animate. In 3D animation movies, human characters are few. Movie makers have tended to focus on virtual and artificial characters like animals or toys; objects of which we have only weak preconceptions how they should talk and gesture. With human faces on the other hand, the smallest glitch is plain to see to everyone, as we are all experts in the perception of the face and its dynamics. For us, it is easy to see whether emotional expressions are genuine, for instance. The goal of this project is to measure in great detail the 3D shapes of faces and the temporal changes thereof, to characterize these 3D face deformations, and to reverse the process and animate static 3D face models. For the 3D acquisition stage a system is used that was developed by VISICS and extracts the 3D coordinates of thousands of points on the face, and that can produce 3D reconstructions at 40ms time intervals. This allows performing a very detailed analysis of the face deformations during an actor's performance. PCA analysis suffices to yield a convincing and very compact representation of the deformations that come with mood or speech. The first experiments are ongoing where the learnt face dynamics are transplanted onto faces for which only a static model is available. The goal is very realistic animation. The preliminary results are very encouraging.