Coupled statistical face reconstruction

Abstract

We present a coupled statistical model that can be used to accurately recover facial surfaces from single images by jointly capturing variations in surface normal direction and surface height. The model is trained on range data. By fitting the model to surface normal data, the surface height function is implicitly recovered without having to integrate the recovered field of surface normals. We show how the coupled model can be fitted to image brightness data using geometric constraints on surface normal direction furnished by Lambert's law. © Springer-Verlag Berlin Heidelberg 2005.