Keiko Abe, Takahiro Doi, Maki Takano and Ichiro Fujita
Computation Underlying Stereoscopic Depth Perception Depends on the Temporal Frequency of Visual Patterns: A Model Incorporating Temporal Frequency Channels
The primate visual system encodes binocular disparity by using two distinct
computations. One computation resembles a cross-correlation between left and
right retinal images (correlation computation). The other computation detects
matched patterns between left and right retinal images, and encodes binocular
disparity between these patterns (matching computation). In a previous psychophysical
experiment, we found that the correlation and matching computations change
their relative contributions to stereoscopic depth perception depending on
the temporal frequency of the visual stimuli. However, more needs to be known
about the mechanism the stereoscopic system uses to control these contributions.
Here, we hypothesized that temporal frequency channels control the relative contributions such that each temporal frequency channel processes a specific temporal frequency component of the visual stimuli. We constructed a computational model in which we incorporated temporal frequency channels to be temporal receptive fields of the primary visual cortical neurons. In our model, band-pass temporal frequency channels fed the input to the correlation computation, while low-pass temporal frequency channels fed the input to the matching computation.
We examined the model performance using the same tasks as those used in for our psychophysical experiments (near/far depth discrimination task). We manipulated the task difficulty by varying the similarity between the left and right retinal images (the proportion of visual elements whose luminance contrast was binocularly matched). We expressed the discrimination performance as a function of the proportion of contrast-matched elements. (psychometric function).
Our model successfully reproduced the experimental data. When the stimulus temporal frequency was high (43Hz), the shape of the model psychometric function followed the prediction of the correlation computation. As the temporal frequency became lower, the shape of the model psychometric function deviated from the prediction of the correlation computation, and partially followed the prediction of the matching computation.
These results suggest that temporal frequency channels control the relative contributions of the correlation and the matching computation in the stereoscopic system.