Specular Photometric Stereo for Surface Normal Estimation of Dark Surfaces

Abstract

This paper presents Specular Photometric Stereo (SPS), which is a Photometric Stereo (PS) technique incorporating specular reflection. The proposed SPS uses multiple images of a surface under different lighting conditions to obtain surface normals similarly to the conventional PS, but uniquely utilizes specular components of a dark surface, which reflects little diffuse light. The proposed framework consists of two sequential numerical steps, which are the conversion of a highly non-linear specular reflection model to a non-linear equation with only one non-linear parameter, and then the iterative removal of the diffuse components. The proposed SPS can estimate normals of dark surfaces, which is not possible by the conventional PS. The proposed SPS was examined using synthesized data and then tested with real-world surfaces. The results of surface normal estimation show that the capability of the proposed SPS over the existing PS in both accuracy and computational cost.