Abstract
It has been shown that binocular perception of depth intervals is both inaccurate and unreliable. On the other hand, binocular discrimination of depth order (called stereoacuity) is extremely reliable. Our recent psychophysical experiments showed that human binocular 3D shape recovery of symmetric polyhedra is also extremely reliable and accurate. These results suggest that binocular shape mechanism relies on binocular judgment of depth order, rather than of 3D distances. Our computational model provided a possible explanation of the underlying perceptual mechanisms by showing how a 3D symmetry constraint interacts with the depth order information to produce a 3D metric shape. The question arises as to whether the stereoacuity thresholds can actually account for the 3D shape recovery results. The study of Norman & Todd (1998) showed that stereoacuity thresholds are substantially elevated when the points, whose depth order is judged, are superimposed on the image of a smoothly curved surface. If these results generalize to the case of vertices of a symmetric polyhedron, will the elevated stereoacuity thresholds account for veridical 3D shape recovery? In order to answer this question we measured thresholds for depth order discrimination between two vertices of a polyhedron in the presence and in the absence of the line drawing of a polyhedron. The threshold was almost twice as big when the polyhedron was present, compared to when the two points were shown in isolation. These results were used to revise our model of binocular 3D shape recovery. We conclude by discussing the role of depth vs. shape information in 3D shape recovery.
National Science Foundation, US Department of Energy, Air Force Office of Scientific Research.