Surface Gradients, Contours and the Perception of Surface Attitude in Images of Complex Scenes

Abstract

Sophisticated computer graphics were used to generate images of three-dimensional blocks-world scenes to investigate the perception of surface attitude. Three types of surface primitive (planar blocks, cylinders, and ellipsoids) were combined to form structured settings. The experiments were designed to investigate whether surface-based information such as gradients in shading and texture provide any significant advantage in attitude judgments over information derived from object contours. Images of shaded, textured, and line-drawn surfaces formed the stimulus set. The subjects' task consisted of setting an attitude probe on different parts of the scene so that the probe appeared to be locally coplanar with the perceived surfaces. Analysis of settings according to attitude components, slant and tilt, shows remarkable agreement in slant settings for the shaded and line-drawn scenes but poor correlation between shaded and textured scenes. Similarly, tilt was also easily judged in shaded and line-drawn scenes and the experiments indicate that explicit surface boundaries are important for stable tilt perception. In general, the results suggest that, for the simple surfaces employed here, surface cues provide little extra information beyond that which is derived from contours.