Abstract
A fundamental problem for the perception of 3D shape from shading is to achieve some level of constancy over variations in the pattern of illumination. The present experiment was designed to investigate how changes in the direction of illumination influence the apparent shapes of surfaces. The stimuli included 3D objects with Lambertian reflectance functions that were illuminated by rectangular area lights. The radial positions of these lights were systematically manipulated to allow five different directions of illumination. All stimuli had exactly the same bounding contours so that those contours provided no information for distinguishing the different possible surfaces. Observers judged the 3D shapes of these objects in two phases: First, they marked critical points (e.g. local depth minima, maxima, and inflection points) along a designated scan line in an image. These were then used to position control points on a spline curve located adjacent to the image, and observers adjusted the shape of that curve to match the apparent profile in depth along the designated scan line. The results revealed that parts of the surface appeared to shift slightly toward the direction of illumination, but these changes were much smaller than what would be expected based on differences in the pattern of luminance among the stimulus images. Regions of high curvature where the surface abruptly changed from flat to curved remained much more stable over changes in illumination than regions with more gradual curvature. These findings demonstrate that there is a substantial amount of illumination constancy in the perception of 3D shape from shading, but that it is not perfect. Several hypotheses are considered about how this constancy could potentially be achieved.
Meeting abstract presented at VSS 2017