Abstract
High resolution stereopsis in the Panum's fusional area (patent stereopsis) is often assumed to depend on matching corresponding image points. There is, however, no obvious way for the visual system to do this, a confound referred to as the "correspondence problem". The alternative we consider is that patent stereopsis depends on binocular information derived from monocular neurons with receptive fields that have the same retinotopic locus. To test this hypothesis, we evolved artificial neural networks stimulated by luminance that fell within the receptive fields of corresponding contralateral and ipsilateral monocular neurons. The responses triggered in this way were conveyed to an associated binocular cell tasked with reporting whether surfaces at its preferred eccentricity were nearer or further than surfaces impinging on nearby retinotopic loci, and by how much. Consistent with the results of visual physiology, we found that: a) far-tuned artificial binocular neurons evolved contralateral ocular dominance; b) near-tuned binocular neurons evolved ipsilateral dominance; and c) the degree of evolved dominance was not correlated with disparity tuning. These observations suggest that vision circumvents the correspondence and false target problems by means of retinotopy and ocular dominance, providing a plausible functional role for the latter.
Meeting abstract presented at VSS 2014