Abstract
The perception of the shape of surfaces defined by binocular disparity requires the matching of regions between the left and right images. The stereoresolution for discriminating such surfaces is relatively low, and it has been argued that this is constrained by (i) the size of the smallest window available in the matching process and (ii) an assumption that disparity is constant within this window (Banks et al, 2004, Journal of Neuroscience, 24, 2077–2089). We investigated, in addition to the constraints imposed by this matching process, the impact of other factors determining the detection of surfaces. To do this, we presented observers with disparity-defined surfaces that were corrupted by binocular decorrelation, and determined the degree of correlation required for their detection. This was done using a 2AFC procedure, in which one interval contained a target surface, and the other contained uncorrelated noise. Detection thresholds were measured for (i) slanted surfaces, over a range of slants and tilts and (ii) sinusoidally corrugated surfaces, over a range of spatial frequencies, amplitudes and orientations. We found that performance differed depending on the type of surface presented. For corrugated surfaces performance peaked at low spatial frequency and amplitude. This result can be explained by a model that predicts a matching process in which disparity is assumed to be constant within a local region. However, we found that for slanted planar surfaces, performance improved at larger magnitudes of slant. We conclude that this result cannot be attributed to the limitations imposed by the sampling process.