Shape from shading and intensity gradient

We present a novel algorithm for shape estimation, using both brightness and brightness gradient as input data. Our algorithm is an improvement of the recently introduced Green's function approach to shape-from-shading (GSFS). In GSFS, we assume that the single brightness image will be matched to a second image through a uniform disparity field, and solve for the matching image via Green's function. When a linear expansion of the reflectance map is considered, the matching image can be related to surface gradient, leading to a closed-form depth map whose free parameters are easily estimated. The author shows that the same procedure can be repeated with the gradient image as input; a second depth estimate thus results which takes into account higher-frequency components of the imaged surface. Extensive experimentation with synthetic and real images corroborates the advantage of the new method.