This project recovers fine surface detail — shape, surface normals and reflectance — from images captured under varying illumination. A recurring goal is to lift photometric stereo beyond its classical single-viewpoint setting: by combining object silhouettes with images taken under changing lighting from multiple viewpoints, the methods recover camera motion, build a visual hull, and produce complete reconstructions “in the round” of shiny, textureless objects that are difficult for conventional stereo.
A second strand addresses moving and deforming surfaces. Multispectral (coloured-light) photometric stereo recovers a dense normal field from untextured surfaces in a single shot, enabling the capture of bends and wrinkles on non-rigid objects and their registration over time from video. The project also tackles one of the principal failure modes of the technique — shadows — showing how to resolve the orientation ambiguity that arises when light is occluded in three-source setups, using integrability and noise-aware reasoning.