Abstract
To measure binocular disparity, the visual system must match the same object elements in the two eyes' retinal images. Solving this correspondence problem becomes increasingly difficult as similar elements are added to the images. In natural viewing, disparity is nearly always correlated with focus cues: blur in the retinal image and accommodation. We asked whether this correlation helps solve the correspondence problem. We presented a stimulus that specified two planes in depth: a frontoparallel near plane and a far plane that could be rotated about a vertical axis. The planes were textured with smooth sticks in random position (van Ee & Anderson, 2001). The sticks were the same color and angular size, so matches in the two eyes' images were frequently incorrect; incorrectly matched sticks did not appear as part of either plane. When the sticks all had similar orientations, false matches were common; when the sticks had a wider range of orientations, false matches were less common. We presented the stimulus in two ways: 1) on one image plane in a conventional 3d display, 2) on multiple image planes in a novel volumetric display (Akeley et al., 2004). The disparities were identical in the two conditions. In the first condition, focus cues specified one plane and were therefore inconsistent with the disparity-specified stimulus. In the second condition, focus cues specified two planes consistent with the disparity stimulus. The first condition mimics standard experimental technique in psychophysics; the latter mimics natural viewing. We measured slant-discrimination thresholds for the far plane. As expected, thresholds were lower in conditions with greater variation of stick orientation. More importantly, thresholds were systematically lower in the volumetric than in the conventional viewing condition. Thus, the visual system takes advantage of the natural correlation between disparity and focus cues when solving the correspondence problem.