Shading and stereo in early perception of shape and reflectance

Abstract

Recent experiments involving shaded 2-D stimuli have shown that early-vision mechanisms are capable of interpreting 3-D shape from shading. In particular, target discrimination tasks suggest that a target 'pops out' when background distractors, but not the target, can be interpreted as convex and lit from above or top-left. Since the problem of extracting 3-D shape from shading is intrinsically ill-defined, early vision may need to make these twin assumptions of convexity and top-left lighting in order to constrain the problem. Would these assumptions be recognized as unnecessary and consequently discarded when 3-D shape could be unambiguously defined by some other cue, like stereo disparity? A 2AFC stimulus onset asynchrony paradigm with masking was used in target discrimination experiments. The performance of five naive subjects on tasks where only shading cues were present was compared with that on tasks involving shading as well as stereo cues that define shape unambiguously. The results show that although stereo disparity information is incorporated by early-vision 3-D mechanisms, it is not used to overturn the default assumptions of lighting and shape. Stereo information is interpreted within the framework of top-left lighting, and a consistent preference for convexity is seen over concavity.