Estimation of illumination direction from matte, Gaussian, wrinkled surfaces

Abstract

We have previously found that human observers use second order statistics in order to estimate the illumination direction of rough surfaces. They could estimate the azimuth modulo random 180 degrees flips and were very bad at elevation estimates. For that study we used photographs of actual surfaces. In the present study we use synthetic stimuli with exact shade and shadow, but without reflexes (multiple scattering). Question: Do observers use shadows to estimate illumination direction? If so, what is the cue they use? Do shadows have to be isolated and visible over their full extent? Result: Observers could estimate the azimuth of the light direction modulo 180 degrees in the shading regime (no shadows). When shadows are present they effectively resolve this ambiguity. They estimate the azimuth within about ten and the elevation within about five degrees. Discussion: Observers do use shadows to estimate the illumination direction. They don't need to see shadows over their full extent, nor is it a problem when multiple shadows run together. They appear to use the difference in structure between body shadow and cast shadow edges.