Abstract
Absolute disparities, which humans detect imprecisely, can support a high sensitivity to relative disparity. Absolute disparities confound the disparity of the stimulus with the vergence state of the eyes; differencing two absolute disparities eliminates the vergence component, leaving a quite precise measure of relative disparity. If the relative disparity of stimuli A and B is computed as a difference between their absolute disparities, sensitivity should be limited by noise associated with the disparity of each binocular stimulus. It should be unaffected by stimulus differences (e.g., A has one orientation and B another) that conserve the precision of absolute disparity signals. I tested this prediction and found it to be false. Observers judged the depth of a central Gabor patch as ‘Near’ or ‘Far’ relative to that of a surrounding annular grating, whose disparity was zero. The two gratings had the same spatial frequency (3 c/d) and independently variable orientations. Their spatial separation was also varied. Nonius alignment preceded the 150 ms presentations. When both gratings were vertical, stereo discrimination thresholds were nearly as low as when a central high-contrast fixation point was present. Rotating the annulus 45 elevated thresholds by a factor of about 5—as high as when the center grating was presented alone, with no annulus. When the center grating was then rotated to be parallel to the annulus, thresholds fell to the low value found when both were vertical. The orientation tuning of relative disparity is quite sharp; disparity sensitivity is halved by orientation differences of about 15 . This tuning is little affected by moderate spatial separations between center and annulus. I conclude that (1) disparity is extracted with approximately equal precision from stimuli ranging in orientation from vertical to at least 45; (2) relative disparity is not computed by simply differencing absolute disparities; (3) it is computed within orientation bands.
Supported by NIH Grant EY12286