A Study Of Statistical Methods For Facial Shape-from-shading

Shape-from-shading addresses the problem of recovering a 3D surface from a single 2D image. In this project, we concentrate on reconstructing faces from a single 2D image by incorporating a statistical model in to a Shapefrom-shading algorithm. We build on the work presented by WAP Smith and investigate the reconstructive power of computing principal components from directed data. We also present a novel reconstruction technique that recovers a set of texture and shape coefficients in order to reconstruct an input image.The novel reconstruction technique presented produces results that are similar in visual appearance to WAP Smith’s reconstructions. We also found that projecting the normals in to a Euclidean space improved the ability of the statistical model to recover facial shape. Using unprojected normals, however, still produced acceptable results.