Abstract : Proximal splitting algorithms play a central role in finding the numerical solution of convex opti- mization problems. This paper addresses the problem of stereo matching of multi-component images by jointly estimating the disparity and the illumination variation. The global formulation being non-convex, the prob- lem is addressed by solving a sequence of convex re- laxations. Each convex relaxation is non trivial and in- volves many constraints aiming at imposing some reg- ularity on the solution. Experiments demonstrate that the method is efficient and provides better results com- pared with other approaches.