We introduce a new Bayesian/Geometric method for estimating orientation and depth from shade and/or texture. The method is based on a prior distribution on the space of all posible (random) surfaces, and a variant of the annealing algorithm for constraint global optimization problems. The method is also useful for partially classifying image contours.
"A Bayesian/Geometric Framework For Reconstructing 3-D Shapes In Robot Vision", Proc. SPIE 1058, High Speed Computing II, (17 May 1989); doi: 10.1117/12.951670; https://doi.org/10.1117/12.951670