Scanning Iridescent Reflectance

Acquisition of high-dimensional spatially-varying bidirectional reflectance distribution functions (SVBRDFs) is extremely challenging for scenes that exhibit high-frequency angular effects-in particular, iridescence-that requires dense sampling across spatial and angular dimensions of reflectance. This is further complicated by the need to illuminate and observe the material along grazing angles, where the effects of Iridescence are often most prominent. This paper proposes an imaging system for acquiring the reflectance associated with such iridescent materials. Our system uses an imaging setup, consisting of an ellipsoidal mirror and a light field camera, that can acquire a dense slice of the SVBRDF across spatial and observation angle axes from a single image, including at grazing observation angles. We use an active illumination setup to sparsely sample the incident illumination angles, thereby providing an acquisition setup that only requires a few (light field) photographs. To compensate for the lack of density in the incident angles, we train a neural model that implicitly represents the SVBRDF, creating a light-weight data-driven reflectance function that enables interpolation of the missing measurements. We show that our imaging system can reconstruct textures that exhibit spatially-varying iridescence stemming from diffraction as well as structural coloration.