Point Clouds Based 3D Facial Expression Generation

Our research in this paper is a new method for facial expression generation based upon extrinsic and intrinsic information with the advantages of direct processing of Point Clouds (PCs). In our pipeline of facial expression generation, there is no need to set up a face model or facial expression database in advance or to construct any mesh or surface models. The basic idea of our method is that the animation of facial expression is achieved by a combination of Global Face Motion and Local Face Motion. In the former, Global Face Motion can be decided by the ICP algorithm, while the latter takes charge of partial movements on the face. Finding correct corresponding points is important for generating an intermediate face from two different facial expressions. This process yields in a good result of the interpolated PCs for facial expression generation between two different facial PCs. Finally, the experimental result shows fairly satisfactory facial expression generation.