Abstract
Recent models of human stereopsis include the constraint that binocularly matched edges must have the same contrast-polarities in the two eyes (Marr & Poggio, Pollard et al., Ohzawa et al.). The psychophysical evidence supporting this constraint is based on studies in which interocular reversals of edge-contrast were created by reversing the relative luminances of image regions (Treisman, Kaufman & Pitblado, Cogan et al.). These studies thus confound edge contrast-polarity with the relative luminance of adjoining surface regions. Here we simulate a common real-world viewing situation in which background surfaces are differentially occluded in the two eyes (Da Vinci stereopsis). In this situation, interocular reversals of edge-contrast can occur without changing the relative luminance of visible surface regions in the two eyes. Although ‘qualitative’ stereopsis has been previously reported using opposite-contrast edges, the stereo thresholds reported for opposite-contrast edges are roughly ten times larger than for same-contrast edges, leading to the view that opposite-contrast edges are processed by coarser mechanisms. Using our stereograms, however, we obtain stereo thresholds for opposite-contrast edges that are close to the smallest previously reported thresholds for same-contrast edges. We also show that the quantitative gain in depth as a function of binocular disparity is similar for our same-contrast and opposite-contrast stimuli. Matches between opposite-contrast edges in our stereograms could be signalled by tuned inhibitory cells (Poggio, 1995), or by cells sensitive to reversed contrast-phase in the two eyes, as reported by Anzai, Ohzawa & Freeman (1999). We also show that the ecologically-valid constraint of interocular consistency of relative surface lightness can be critical for stereopsis.