Combining curvature evidence to shape-from-shading

This paper describes a new curvature consistency method for shape-from-shading. Our idea is to combine evidence for the best surface normal direction. To do this we transport surface normals across the surface using a local estimate of the Hessian matrix. The evidence combination process uses the normal curvature to compute a weighted average surface normal direction. We experiment with the resulting shape-from-shading method on a variety of real world synthetic data.