Abstract
When interpreting 3D surfaces, the human visual system is often confronted with regions with little or no structure. It has been shown that the stereoscopic system rapidly and effectively interpolates depth estimates across ambiguous regions to create the percept of a continuous plane. In previous studies, we have found reduced estimates of perceived depth magnitude for rows of elements with smooth disparity gradients (Deas & Wilcox, 2015). Here we evaluate if such distortions are also seen for extended surfaces, and assess the extent to which the interpolation process is influenced by the coherence of disparate elements. We used stimuli comprised of sparse, randomly positioned elements whose disparity map depicted two overlapping Gaussian bumps, with laterally separated peaks, but variable amplitude and width. The Gaussians were constrained so that the peaks were aligned horizontally, and the stimulus was encircled by a textured annulus positioned on the screen plane. Two horizontal line segments, one on each side of the circle, indicated the middle of the stimulus: the region that the observers were asked to estimate. Subjects viewed the stimuli stereoscopically, and on each trial toggled between the stimulus and a response screen. On the response screen, subjects indicated the perceived height of the surface along the horizontal midline, using lines with nodes that could be adjusted vertically, effectively drawing a cross-section of the surface. We found that depth estimates were accurate in the fronto-parallel regions at the edges of the bumps, but that for all test conditions, subjects systematically and significantly underestimated depth within the region containing depth variation. The depth distortions were remarkably consistent across observers, were not influenced by the percentage of elements displaced off the surface, and appear to have been constrained by the regions of maximum surface curvature.
Meeting abstract presented at VSS 2017