Exploring material recognition for estimating reflectance and illumination from a single image

In this paper, we propose a novel approach for recovering illumination and reflectance from a single image. Our approach relies on the assumption that the surface geometry has already been reconstructed and a-priori knowledge in form of a database of digital material models is available. The first step of our technique consists in recognizing the respective material in the image using synthesized training data based on the given material database. Subsequently, the illumination conditions are estimated based on the recognized material and the surface geometry. Using this novel strategy we demonstrate that reflectance and illumination can be estimated reliably for several materials that are beyond simple Lambertian surface reflectance behavior because of exhibiting mesoscopic effects such as interreflections and shadows.