Abstract
The distribution of the preferred disparities of binocularly-tuned neurons is highly non-uniform, with many more neurons tuned to small disparities close to the horopter. These neurons have been classified as tuned-excitatory, tuned-inhibitory, near and far (Poggio 1995), depending on the way that their responses vary with disparity. Near and far cells respond to a broad range of crossed or uncrossed disparities, respectively, while the responses of tuned excitatory or inhibitory cells are modulated by disparities within a narrow range. This distribution of disparity tuning is optimised for the efficient coding of the disparities found in natural scenes, which are strongly peaked around zero (Liu, Bovik and Cormack, 2008; Sprague, Cooper, Tošić and Banks, 2015). We assessed whether (i) this efficient encoding of binocular information might account for human sensitivity to disparity differences and (ii) whether this encoding adapts to short-term changes in the distribution of disparities. Test stimuli were random-dot stereograms in which one patch was shifted forwards in depth from a pedestal value, and the other shifted backwards. We used a 2AFC procedure to measure sensitivity to depth differences as a function of the pedestal value. Disparity thresholds increased with pedestal disparity, consistent with the prediction of an efficient coding model. To assess whether this sensitivity adapts to short-term changes in the distribution of disparities, observers were then exposed to a small (0 arc min) or large (±20 arc min) range of disparities. Sensitivity was measured again after adaptation. Thresholds were not affected by short-term adaptation. These results show that both the physiological encoding of disparity, and observer's sensitivity to depth differences, are consistent with an efficient coding of the distribution of disparities found in the natural environment. Finally, disparity encoding does not appear to adapt to short-term changes to the distribution of disparities.
Meeting abstract presented at VSS 2017