LBF Based 3D Regression for Facial Animation
Congquan Yan, Lianghao Wang, Jianing Li, Dongxiao Li, Ming Zhang · 2016
This paper presents a system for performance-driven avatar animation by estimating facial pose and expression parameters from single image. In this system, a 3D shape prediction model is trained based on local binary feature (LBF) algorithm, which use random forest to extract image features and learns a linear regression model mapping these features to 3D shape. With the help of this model, the 3D positions of facial vertexes can be estimated from a web camera image. The facial pose and expression parameters can be calculated by solving an optimization problem that fitting a set of blend shapes models to these 3D vertexes. Experiments show that our system can estimate accurate facial parameters from single image and generate similar looking avatar animation.