Abstract

In this paper, we aim to recover the 3D shape of a human face using a single image. We use a combination of symmetric shape from shading by Zhao and Chellappa and statistical approach for facial shape reconstruction by Atick, Grifin and Redlich. Given a single frontal image of a human face under a known directional illumination from a side, we represent the solution as a linear combination of basis shapes and recover the coeficients using a symmetry constraint on a facial shape and albedo. By solving a single least-squares system of equations, our algorithm provides a closed-form solution which satisfies both symmetry and statistical constraints in the best possible way. Our procedure takes only a few seconds, accounts for varying facial albedo, and is simpler than the previous methods. In the special case of horizontal illuminant direction, our algorithm runs even as fast as matrix-vector multiplication.