Abstract
Consider a stereo display containing two vertically-aligned rectangles, one with zero disparity and the other slanted about the vertical axis. Their relative slant is underestimated when the rectangles are amodally unified into a single object behind a fronto-parallel occluder (either real or illusory) than in the baseline condition in which they are perceived as separate objects. This effect is at odds with the enhancement of perceived slant by stereoscopic boundaries (Gillam et al., 1988), but is consistent with the hypothesis that amodal unification involves a process of approximation, not simply an interpolation based on the extrapolation of specified fragments. Approximation is characterized by the modification of image parts in limiting cases of unification; for instance, when the smooth unification of non-coplanar surfaces and the minimization of their deviation from coplanarity requires torsion. Relative slant was measured in three conditions: real, illusory, and no occluder. Observers viewed disparate images on a monitor through a double-mirror stereoscope and adjusted a gauge figure to match the perceived slant of the upper rectangle. The lower zero-disparity rectangle was constant. Different groups of observers judged 12 experimental and 6 control displays for each condition. In experimental displays 6 slant amounts and 2 vertical alignments (symmetric vs. asymmetric) were combined. In control displays both monocular images included the same upper rectangle (zero disparity) with a width equal to the average of corresponding experimental rectangles. As predicted by the approximation hypothesis, the sign of slant was detected better without the occluder. Relative slant was overestimated in the no-occluder condition and underestimated in both occlusion conditions. Slant underestimation in occlusion conditions increased as a direct function of disparity (i.e., of simulated relative slant).
Grant MIUR-COFIN 2003115470 to Walter Gerbino