Tomoki Ichikawa

Polarization is a fundamental property of light, alongside intensity and color. Unlike intensity and color, it is invisible to the human eye and has often been overlooked. Today, polarimetric cameras allow us to readily capture linear polarization. In computer vision, polarization has mainly been exploited as an auxiliary cue for various tasks.

Rather than treating polarization simply as an additional cue, this talk explores two complementary directions: modeling polarization and designing polarization. I first present two works on modeling how polarization behaves in the real world. The first models outdoor polarimetric reflection to exploit the unique polarization pattern of a clear sky for surface normal recovery. The second introduces the Fresnel Microfacet BRDF, a physically based reflectance model that unifies surface and body reflections with the same microgeometry and describes both their radiometric and polarimetric behavior. It enables accurate reflectometry from a single polarimetric image.

In the final part, I turn to designing polarization and present Structured Polarization, a projector-camera sensing framework that encodes illumination patterns in polarization rather than in intensity or color, as conventional structured light does. Since these polarization patterns are invisible to the human eye, they preserve the natural appearance of the target while enabling joint recovery of shape and reflectance. I first present SPIDeRS, which realizes this concept, and then Spatial Polarization Multiplexing, which extends it to single-shot recovery.