An Efficient Approach for Multi-view Face Animation Based on Quasi 3D Model

摘要

Multi-view face animation is widely required in various applications nowadays, but most of existing relative techniques have burdensome computational load in data registration, model construction and animation. Addressing those problems, this paper proposes an efficient approach via employing quasi 3D model, which is the fusion of a 3D geometry model with 2D facial textures. 3D Point Distribute Model (PDM) serves to model geometrical deformation in the shape model. By preserving depth information, the proposed approach is convenient to manipulate pose variation and deal with new-coming subjects. By taking advantages of image-based techniques, data collection is facile and the number of our 3D model vertices is reduced to less than one hundred, while animated faces still keep expressive by incorporating with partial Expression Ratio Image (ERI). The primary experiments demonstrate that our approach efficiently achieved individual animated face among viewpoint range of [-60, 60] based on only 2 input facial images.