We previously described the spatiotemporal requirements for binocular correlation in stereopsis using sinusoidal gratings-in-depth (Lankheet and Lennie, 1996 Vision Research36 527 – 538). The use of smooth sinusoidal surfaces emphasised the effects of spatial and temporal integration. Binocular correlation, however, depends not only on integration, but also on segregation at depth discontinuities. In the present experiments we therefore investigated segregation-in-depth, using random dot stereograms that depicted two transparent frontoparallel planes positioned symmetrically on either side of a binocular fixation point. Sensitivity for segregating the two planes was established by adding Gaussian distributed disparity noise to the disparities specifying the planes, and finding the noise amplitude that rendered transparency just detectable. The stimuli consisted of dynamic random-dot displays (dot lifetime 4 frames, at a frame rate of 67 Hz), generated in real time by a Macintosh computer, displayed on a television monitor, and viewed through a stereoscope. We used a method of constant stimuli and a 2AFC procedure. Two transparent planes were presented in one interval, and a single plane, with Gaussian distributed disparity values spanning the same range, was presented in the other. Segregation of stationary patterns was optimal for disparity differences of about ±9 min arc. Differences smaller than ±3 min arc and larger than about ±18 min arc could not be resolved. Motion contrast between the two patterns greatly facilitated segregation in depth. The facilitating effect increased with the difference in motion directions. The optimal speed varied with the difference in disparity.
Shape from depth discontinuities under orthographic projection