Several successful methods for shape reconstruction from image sequences have been developed using a variational formulation. In this work we utilize the same framework, which leads to a Partial Differential Equation (PDE) describing the motion of an initial surface to a refined surface that is a better match to the input images. Motivated by the primary goals of reconstructing the object and the parameters to a reflectance model, we take advantage of known lighting conditions in the error measure. In particular, we assume that there is light variation, due to object rotation relative to the light source, allowing the recovery of shape in both textured and textureless regions. Additionally, we propose a method to filter out specular highlights, which allows the recovery of surfaces having non-Lambertian reflectance. Following recent work using an explicit surface parameterization, we apply the PDE refinement to a deformable mesh. We test our method using images obtained from an easy to use capture setup. The setup is accessible to the average PC user, because the only hardware requirements are a camera, a light source, and a glossy white sphere. The capture setup provides images, camera calibration, light calibration, and silhouette images to be used in the refinement. The visual hull is used as a starting point for the PDE evolution. At the end of the refinement, our method outputs a triangulated mesh and the parameters of a Phong reflectance model represented in texture space. Results on real and synthetic images demonstrate that this method is capable of recovering the geometry of textureless surfaces, and moderately textured surfaces, but is unstable in the recovery of deep concavities. Several examples on real sequences illustrate the applicability of our models in computer graphics applications, where the recovered objects are composed and rendered under novel lighting and view conditions.