From binocular input correlation to disparity tuning and ocular dominance in cortex
Presented at "Vision in a 3-D world", AVA annual meeting, (Applied Vision Association), 19 March 2003, London
Stereo images are highly redundant; the left and right frames of typical scenes are very similar. Li and Atick (1994, Network, computations in neural systems Vol.5 157-174) have measured the binocular correlation of the input signals and hypothesized that the striate cortex is concerned with, among other things, removing binocular correlations in the inputs to achieve more efficient input coding. We can thus show the principle components in the stereo input signals and derive the most efficient coding strategies that achieve binocular decorrelation. It is shown that multiscale coding combined with a binocular decorrelation strategy leads to a rich diversity of cell types: monocular/binocular and disparity selective cells, among which one can identify cells that are tuned-zero-excitatory, near, far, and tuned inhibitory. Consequences on cortical ocular dominance column formation from abnormal developmental conditions such as strabismus and monocular eye closure can also be explained (Li 1995, The neurobiology of computation P. 397-402. Ed. J. Bower, Kluwer Academic Publishers). The theory also provided testable predictions on the correlations between the following receptive field properties: ocular dominance, receptive field size, optimal orientation, and disparity selectivities. In particular, it predicted that striate cortical cells tuned to near horizontal orientations are more likely binocular than cells tuned to near vertical orientations. This prediction has been subsequently confirmed using data from cats. A psychophysical test of the prediction will also be shown.