Accurate human face pose recovery from single image through generic shape regularization

In this paper, we propose a novel approach to recovering face pose from a single face image. The projection from 3D space to image plane is modeled by a scaled orthogonal projection, which contains the unknown pose parameters and face shape. To remove the ambiguity, the projection is regularized by prior information about generic face shape. Given a set of image features and a set of 3D features selected from a generic face, we can solve the unknown pose parameters conveniently through a Newton method. Owing to the shape regularization, the efficiency and accuracy of the proposed method precede the existed approach.