On Learning Shapes from Shades
Abstract
Shape from Shading (SFS) is one of the most extensively studied problems in Computer Vision. However, most of the approaches only deal with Lambertian or other specific shading models, and are hence limited in their applicability to real images. In this contribution we propose a general unified framework in which it is possible to incorporate different illumination and shading models. We effectively deal with the usual multiple local minima problem of the SFS domain through some minimal user interactions. Features such as pyramidal refinement, parallelizability of solution evolution, global smoothness of solution give our framework a definite edge over most other existing SFS schemes. Results on real images demonstrate the efficacy of our approach.