In this paper, we propose a fully automatic, effective and efficient framework for 3D face reconstruction based on a single face image in arbitrary view. First, a multi-view face alignment algorithm localizes the face feature points, and then EM algorithm is applied to derive the optimal 3D shape and position parameters. Moreover, the unit quaternion based pose representation is proposed for efficient 3D pose parameter optimization. Compared with other related works, this framework has the following advantages: 1) it is fully automatic, and only one single face image in arbitrary view is required; 2) EM algorithm and unit quaternion based pose representation are integrated for efficient shape and position parameters estimation; 3) the correspondence between 2D contour points and 3D model vertexes are dynamically determined by normal direction constraints, which facilitates the 3D reconstruction from arbitrary view image; 4) a weighted optimization strategy is applied for more robust parameter estimation. The experimental results show the effectiveness of our framework for 3D face reconstruction.
展开▼