Abstract
In natural vision, the luminance in each local region of a matte surface is primarily determined by its local orientation relative to the direction of illumination. However, it is also possible to obtain compelling perceptions of shape from shading for range images, in which the intensity at each pixel location is determined by the relative depth of the closest surface region that projects to it. Although there are some situations such as spheres with point source illuminations where range images can be identical to a rendering based on Lambert’s law, they are more commonly quite different from one another. Consider, for example, a valley that is bounded on both sides by ridges. In a cross-section through a Lambertian rendering there will be three intensity maxima – one for each ridge and one for the valley. For a range image, in contrast, there will only be intensity maxima on the ridges, and the valley will be marked by a local minimum of image intensity. The present research was designed to investigate observers’ shape judgments for these two types of displays, and to model the expected pattern of performance using a shape from shading algorithm based on an assumed Lambertian reflectance function and a distant point source illumination. The stimuli included both Monge surfaces and deformed spheres, and we examined several different reflectance models and patterns of illumination. The results reveal some systematic differences among these conditions, but they are much smaller than what would be expected from classical shape from shading algorithms. These findings cast doubt on the common belief that the perceptual mechanisms for determining shape from shading are based on an implicit assumption that visible surfaces have Lambertian reflectance functions.
Meeting abstract presented at VSS 2012