Approximating 3D facial shape from photographs using coupled statistical models

Abstract

In this paper we focus on the problem of developing a coupled statistical model that can be used to recover surface height from frontal photographs of faces. The idea is to couple intensity and height by jointly modeling their combined variations. We perform Principal Component Analysis (PCA) on the shape coefficients for both intensity and height training data in order to construct the coupled statistical model. Using the best-fit coefficients of an intensity image, height information can be implicitly recovered through the coupled statistical model. Experiments show that the method can generate good approximations of the facial surface shape from out-of-training photographs of faces. © Springer-Verlag Berlin Heidelberg 2006.