Reconstruction of Archaeological Finds using Shape from Stereo and Shape from Shading

We propose a novel algorithmic framework for the refinement of sparse 3D models using shape from shading. Starting from an initial model obtained by shape from stereo we use a global optimization scheme in order to refine the surface. The constraints we use are based on the shading in the image, the initial 3D points obtained by stereo and the smoothness of the surface. In contrast to many previous approaches that assume that the photometric properties of the scene are known we iteratively update the light source direction and several parameters of the reflectance map. A careful choice of several boundary conditions on the reflectance map, the use of a multigrid scheme, the discretization of the object on a pseudo-hexagonal grid and the frequent switching between different refinement modules allows us to optimize the highly underconstrained problem. We provide results on synthetic data and reconstruction examples of frieze parts from an excavation site in Sagalassos (Turkey).