Abstract
Human subjects perceive depth when viewing binocularly correlated stereograms. Binocular anti-correlation of an entire stereogram abolishes depth perception, while anti-correlation only of the center part of the stimulus accompanied by a correlated surrounding area reverses the direction of perceived depth. Here we developed a computational model which explains the effects of the surround on depth perception. The model consisted of input units responding to disparity in either the center or surround of the stimuli. Anti-correlation of the stimuli inverted the disparity tuning curves of the input units, mimicking V1 neurons. Integration of the input units' responses with threshold operation resulted in relative disparity selective units. The model output was the difference between the responses of two relative disparity selective units preferring either near or far disparity. The model reproduced the effects of surround area binocular correlation on depth perception. We tested the model with psychophysical experiments using random dot stereograms consisting of center and surround areas. In each trial, all dots in the center had the same disparity (−0.32 or +0.32 deg). Dots in the surround were divided into two groups. Different disparities of equal magnitude but opposite sign were assigned for dots in each group (0, ±0.1, …, or ±1.0 deg). Each area was either binocularly correlated or anti-correlated. Four human subjects discriminated the depth of the center against the surround. When both the center and the surround were correlated, subjects reported the depth based on the minimum relative disparity between the center and the surround. Stimuli with correlated center and anti-correlated surround caused reversed depth when the magnitudes of the surround disparities were small (0.2 deg). The results were agreement with our model. We suggest that relative disparity computation between the center and surround is crucial for the effects of surround area binocular correlation on depth perception.