A Different(ial) Approach to the Photometric Stereo Problem

Photometric Stereo (PS) is a fundamental problem in Computer Vision aimed at reconstructing the 3D shape given multiple images of the object under different illuminations. It can be expressed as a non-linear system of first order Hamilton Jacobi equations when 3 or more images are given. We talk about realistic PS modeled by PDEs starting from the problem when only 2 images are available. We extend our previous results on this problem, considering the more realistic perspective projection of surfaces during the photographic process and point light source parametrization. We prove that the new differential formulation is well-posed leading to quasi-linear PDEs.