Abstract
Westheimer (1979) showed that stereoacuity thresholds were an order of magnitude poorer when only absolute disparity information was available compared with relative disparity information. He proposed that this reflected an early mechanism that measured disparity differences, and was hence insensitive to fluctuations in absolute disparity. To examine this possibility explicitly, 3 subjects performed a stereoacuity task with a centrally fixated, circular, dynamic RDS stimulus (2.8 deg diam.) consisting of centre and surround regions of equal area between which a relative disparity was introduced, all against an uncorrelated background. This avoided monocular cues to the depth in the stimulus and allowed for simultaneous variation of both absolute and relative disparities. Normally distributed noise (s.d. 0 to 4 arcmin) was added to the centre and surround on each refresh frame in such a way as to disrupt either absolute disparities only (absolute noise) or relative disparities also (relative noise). The average relative disparity of the noise added to the stimulus over each 1-sec trial was arranged to be zero, so the noise would not disrupt the performance of an ideal observer. Absolute and relative disparity noise could also be combined to look at interactions. Absolute disparity noise was much better tolerated than relative disparity noise, consistent with Westheimer. For noise of 4 arcmin s.d., thresholds were an order of magnitude higher for relative noise than for absolute noise. 2D regression analysis showed there was no interaction between the effects of absolute and relative disparity noise: they summed linearly. This result is consistent with the presence of a specialized mechanism for processing relative disparity that is insensitive to absolute disparity over the range tested, and the disruptiveness of relative disparity noise shows that subjects were unable simply to integrate disparity information over the 1-sec trial to estimate relative disparity.