Performance-based Virtual Face Gestures

The face motion plays a special role in sustaining the human gestures. The emerging media applications are conveying the human expressions by means of virtual faces which are typically controlled by a parameterized description. In this context, we developed a Facial Analysis and Synthesis Scheme able to extract a spatial-temporal semantic face description from an image sequence. This research proposes new methods for detection and tracking of the face and facial features, for estimating the face movements and the facial expressions, and for extracting efficiently stored facial animation. By combining vision, graphics and mechanical models, we can solve the inverse problem: estimating the 3-dimensional face deformation from the 2-dimensional image measurements. Animating the face model with the extracted parameters, results in a faithful reproduction of the original gestures.