Self shadows and cast shadows in estimating illumination distribution

This paper compares two methods for estimating illumination distributions from shadows: methods using self shadows and cast shadows. The cast shadow method provides an effective framework for recovering illuminations in a real scene utilizing variations of image brightnesses inside shadows cast by an occluding object of known shape. The shadow surface is typically the same plane on which the object is placed. A new self shadow method is proposed in this paper in which the shadow surface is on the occluding object itself. We show that it works even for lighting environments where the cast shadow method fails while clarifying the stability issues related to sampling resolution, and the selection of lighting model. We also present an automatic framework for estimation of near lighting by a depth identification, as well as the directional analysis of the illumination distribution, which was not addressed in the cast shadow method. We prove the effectiveness of our new method with experiments using real scenes.