Automatic 3D face texture mapping framework from single image

Abstract

This paper proposes a novel face texture mapping framework for 3D face reconstruction from a single frontal view or half-profile view facial image. Face reconstruction method that originates from the proposed framework, unlike most of the existing ones, is novel in the sense that it is not tightly coupled to a specific face model, and yet it simplifies the pose estimation problem which is pivotal to the success of face reconstruction. This paper details the proposed framework, and illustrates how it addresses the ill-posed pose estimation problem, of which the solution is optimal in the least square sense. With accurate pose estimation of face, precise texture mapping is thus made possible to allow photo realistic rendering of face images in the 3D space. Experimental results demonstrates that reliable and photo realistic 3D face reconstruction can be easily realized in our framework by utilizing a generic 3D face model, standard Haar-like feature based detector and active appearance model. With proposed frame work, face recognition systems could be more robust to pose changes by reconstructing frontal faces from non-frontal ones.