A Piece-Wise learning approach to 3D facial animation

Abstract

Web-based 3D facial animation is an alternative of face to face communication. Animating 3D human faces is also a difficult task because of the substantive non-rigid facial motions and utmost human familiarity with facial expressions. This paper presents a novel piece-wise learning approach to 3D facial animation driven by facial motion capture data. The pipeline of our algorithm comprises three major parts: (1) data pre-processing, (2) facial region segmentation, and (3) facial deformation. We first present an effective pre-processing algorithm for non-rigid motion extraction and data alignment. Second, based on the statistical and kinematical analysis of motion capture data and the topological analysis of facial mesh, our system segments the facial regions by a two-layer clustering algorithm. The edges of segments are well considered using an adapted plane/space partition algorithm. During runtime, the stream of motion capture data and the 3D face model are efficiently fused by cluster-wise optimization. The experimental results show that our algorithm is not only realistic but also fast enough for real time applications. © 2008 Springer-Verlag Berlin Heidelberg.