Abstract
Julesz (1960) demonstrated, using RDS, that binocular depth perception can be based solely on binocular disparity, defined as a difference between spatial positions of points in the two retinae. It has been conjectured, however, that there is a second binocular mechanism, based on the average of spatial positions of points in the two retinae (Rose & Blake, 1988; Regan, 2000). We will present a new visual phenomenon, which demonstrates the operation of both. Images were presented on a monitor and viewed by using a modified Elliot's stereoscope. The simulated object was a 3-D cube. In condition (i), the cube was represented by 9 visible edges, and in condition (ii), by 6 visible edges and a dot at the vertex of the visible Y junction. In (i) there were strong perspective cues to the three-dimensionality of the stimulus, but in (ii) perspective cues were largely absent. One image of the stereo pair was stationary. The second image changed; the changes were produced by an oscillation of the simulated object around the vertical axis going through the center of the cube. For both stimuli, depth calculations based on disparity predict a percept of a non-rigid object whose Y junction vertex moves toward and away from the eye that views the stationary image. In (i) the percept is of a cube oscillating around the vertical axis, with minimal or no non-rigidity, in violation of the disparity calculations. The amplitude of the perceived oscillation is smaller than that of the simulated object. In (ii), the percept does correspond to a non-rigid cube stretching in depth, but the cube sometimes oscillates around the vertical axis for some subjects. These results are inconsistent with any theory that is based exclusively on absolute or relative disparity. A model that can account for these results requires a combination of (1) monocular depth cues, (2) binocular disparity, as well as (3) a weighted average of spatial positions of points and features in the two retinae.