Reconstruction of 3D Face from a Single 2D Image for Face Recognition

Abstract

In this paper, a synthetic exemplar based framework for face recognition with variant pose and illumination is proposed. Our purpose is to construct a face recognition system only according to one single frontal face image of each person for recognition. The framework consists of three main parts. First, a deformation based 3D face modeling technique is introduced to create an individual 3D face model from a single frontal face image of a person with a generic 3D face model. Then, the virtual faces for recognition at various lightings and views are synthesized. Finally, an Eigenfaces based classifier is constructed where the virtual faces synthesized are used as training exemplars. The experimental results show that the proposed 3D face modeling technique is efficient and the synthetic face exemplars can significantly improve the accuracy of face recognition with variant pose and illumination.