Deriving facial articulation models from image sequences

Human facial articulation models are derived from frontal and side view image sequences using a connected vibrations non-rigid motion tracking algorithm. First, a 3D head geometric model is fitted to the face subject in the initial frame. The face model is masked with multiple planar membrane patches that are connected with each other. Then, the in-plate facial motions in the image sequences are computed from an over-determined system. Finally, this information is exploited for creating or customizing a facial articulation model.