Skip to Main Content
Attempting to model face animation with both picture-perfect appearance and natural head rotation, this paper describes a multi-view face animation model for photo-realistic face animation in novel views from videos of a talking person using machine learning techniques. For modeling face, we express face by a multi-view face texture space model with the aid of 3D point distribution model space to express the shape variation in compact way, which can result in synthesis of the novel view face texture with various realistic expressions and poses. The appearance space and shape space are connected by bridge of 2D mesh structures. Following the idea of analysis-by-synthesis, we employed model fitting analysis using Levenberg-Marquardt optimization and trajectory training for smooth and continuous animation. The encouraging experimental results are shown.
Date of Conference: 17-19 May 2004