Qualitative shape from shading

When human observers are asked to describe the shape of a surface, they often identify an arrangement of surface features like bumps, dimples, ridges, or valleys. The central hypothesis of the present research is that the perceptual representation of three-dimensional shape has a graph-like structure that is defined by patterns of surface curvature, and that this is the structure that artists depict when they produce line drawings of objects. Two experiments were performed, in which observers marked the boundaries of bumps on a shaded surface, or the locations of ridges and valleys. Although they were not specifically instructed about where those features were located, the observers' responses corresponded quite closely with the curvature extrema on each depicted object, and their judgments exhibited a high degree of constancy over changes in the pattern of illumination. The relationship is much weaker between the perceived locations of ridges and valleys and the local extrema of luminance in an image. Although variations of luminance are strongly influenced by the pattern of surface curvature, they are also influenced by local variations in illumination caused by multiple light sources, cast shadows, or indirect reflections. Human observers can somehow distinguish between those two components of luminance variation, but the visual information that makes that possible has yet to be determined.