The Shape--from--Shading problem is a well known ill-posed problem. Several contributions have addressed the case of Lambertian surfaces improving the model with the introduction of perspective deformations or studying the corresponding photometric stereo problem. In our study we focus the attention on a different improvement which is intended to reduce the assumptions on the properties of the surface dealing with more general (and real) non-Lambertian surfaces. Our goal is to find a unique model which should be flexible enough to handle many different kinds of real images. As a starting point for this rather big project, we consider the basic model of a single nonlinear partial differential equation (PDE) where we need to introduce new terms to tackle the general non-Lambertian case. In particular, in this talk we will consider the non-Lambertian diffusive Oren-Nayar reflectance model and the specular Phong model, we will construct semi-Lagrangian approximation schemes for the corresponding nonlinear PDEs and we will compare their performances with the classical Lambertian model in terms of some error indicators on a series of benchmarks images.