Reillumination-driven shape from shading

摘要

Most existing SFS techniques aim to recover accurate surface shape or normal estimates. In this paper, we opt instead to place constraints directly on the images that can be produced from the shape information, thus decoupling the smoothness constraint from the surface normal estimates. We show that this approach offers advantages over existing techniques in terms of generating plausible re-illuminated images of simple, piecewise-smooth objects under variable light conditions. Such techniques could have applications in graphics and object recognition.