Abstract
Purpose: Random dot stereograms containing two disparity populations can be perceived as two surfaces separated in depth (e.g. Akerstrom & Todd, 1988), a far opaque surface being perceived through a near transparent surface. This poses a more complex stereo correspondence problem than a single surface stimulus, as it violates the continuity constraint. We computed the efficiency of depth discrimination for stereoscopically defined transparent surfaces, and compared this to a similar opaque surface task. Methods: We presented random dot stereograms with disparities defining a ‘near’ and a ‘far’ surface (relative to fixation). In the transparent condition the two surfaces were presented simultaneously, in the opaque condition each surface was presented in one of two consecutive temporal intervals. The observers' task in both conditions was to respond which surface, ‘near’ or ‘far’, was farther in depth from the fixation plane. Performance was limited by adding different levels of disparity noise chosen to constrain performance around a d' value of 1. We measured human sensitivities across a range of dot densities and disparity differences. To compute efficiency, the squared ratio of human and ideal sensitivity, we ran simulations of the ideal observer in the same conditions. Results: We found 1) across these ranges human and ideal sensitivities were consistently lower for the transparent conditions; 2) efficiencies were also consistently lower for the transparent conditions; 3) efficiencies depended on both dot density and disparity; 4) efficiencies were consistently low compared to efficiencies reported for other visual tasks. Conclusions: There is a significant processing limitation for depth discrimination of stereoscopic transparent surfaces, greater than that for depth discrimination of opaque surfaces. We suggest that models of stereo correspondence should exhibit similar limitations.