Polarisation photometric stereo

Abstract

This paper concerns a novel approach to fuse two-source photometric stereo (PS) data with polarisation information for complete surface normal recovery for smooth or slightly rough surfaces. PS is a well-established method but is limited in application by its need for three or more well-spaced and known illumination sources and Lambertian reflectance. Polarisation methods are less studied but have shown promise for smooth surfaces under highly controlled capture conditions. However, such methods suffer from inherent ambiguities and the depolarising effects of surface roughness. The method presented in this paper goes some way to overcome these limitations by fusing the most reliable information from PS and polarisation. PS is used with only two sources to deduce a constrained mapping of the surface normal at each point onto a 2D plane. Phase information from polarisation is used to deduce a mapping onto a different plane. The paper then shows how the full surface normal can be obtained from the two mappings. The method is tested on a range of real-world images to demonstrate the advantages over standalone applications of PS or polarisation methods.