Convergence of Model Based Shape from Shading

Three dimensional models are used by a wide variety of applications such as face recognition and model based coding for teleconferencing. In face recognition and model based coding the generic 3D model is used to generate images of the individual from any viewpoint. Recent work has been focussed on using shape from shading for refinement of the generic model toward the 3D shape of the individual. However, the previous methods have been based on iteration schemes in which convergence could not be proven. The main goal of this paper is to introduce a provably convergent algorithm for solving the problem of model based shape from shading. Bounds on the number of iterations for attaining error of order h are also given.