Energy minimization methods have been shown to be particularly effective for solving a wide class of inverse problems in computer vision and image processing. Variational methods are energy minimization methods which are at the interface between several mathematical concepts : PDEs, calculus of variations, statistics and optimization. In this talk, we revisit the so-called photometric stereo problem, which is an extension of the "shape-from-shading" problem using several images, under a variational approach. In particular, we discuss the choice of an appropriate model (L2 vs L1 data term, Sobolev vs TV regularization, etc.) and efficient numerics based on augmented Lagrangian and splitting methods (ADMM, split-Bregman schemes, etc.) for performing the minimization. As an application of the variational approach to photometric stereo, 3D-reconstruction results on real-world data are presented.