Abstract
Binocular disparities are informative about the three-dimensional shape of objects as well as their spatial layout in the scene. Is the visual system equally efficient in representing shape and depth? This question is difficult to answer because of their non-commensurate dimensions, namely depth is measured as a distance from the observer whereas shape can be represented as surface curvature or local surface orientation. We addressed this issue by comparing two stimuli, a “step” stimulus representing fronto-parallel surfaces in different depth planes and a “slant” stimulus representing surfaces oriented in depth. Importantly, the stimuli differed only by a simple depth-irrelevant manipulation. More specifically, stimuli were composed of a series of vertical lines at crossed (C) or uncrossed (U) disparities in the order C-C-U-U-C-C-U-U etc. Small horizontal lines connected consecutive pairs of same disparity to produce the step stimulus, or they connected consecutive pairs of dissimilar disparity to produce the slant stimulus. It is important to note that the horizontal lines did not add any new binocular disparity, so the two stimuli were identical in terms of the amount of depth displayed. Discrimination thresholds were measured for both stimuli (front/back or left/right slant judgments). We found that disparity thresholds were at least ten times smaller for step as compared to slant stimuli, even though the two stimuli contained the same binocular disparity information. In addition, disparity thresholds for slant stimuli increased linearly with object size whereas thresholds for step stimuli remained roughly constant for the range of sizes used. These results suggest that at least two three-dimensional representations are extracted from binocular disparities, one representing the distance from the observer, the other surface orientations, and that most of the binocular information is lost to generate the latter.
Chaire d'excellence from the French Ministry of Research