Abstract
Stereopsis requires the ability to align the eyes. We describe a two phase developmental theory undergirded by neurons of qualitatively different character. Recent work has found that, across multiple species, orientation maps in V1 are characterized by a remarkable property: the density of singularities per hypercolumn area is constant. By no means a feature of all orientation maps, simulations show that this is a universal property of maps that arise from dynamics that have long-range inhibition (Kaschube et al., 2010 Science, 330: 1113). We predict that long-range inhibition will also be a defining characteristic of ocular dominance maps. In our theory, segregation of the inputs to the eyes in V1 is followed by a set of randomly directed inhibitory connections between ocular dominance columns. The result of the long-range interactions are neurons excited primarily through one eye and inhibited primarily through the other. Across the population, there is sensitivity to all directions of binocular disparity, and inhibition gives rise to a broad disparity-dependent response modulation that is suitable as an error signal for aligning the eyes. For example, neurons on the horizontal meridian sensitive to vertical disparity can be combined to provide signals for torsional error correction. In addition, the asymmetric ocularity conserves eye-of-origin information and so provides a foundation for the representation of monocular regions in occlusion-based stereopsis. The characteristics described above are precisely those of a subset of the neurons termed Near and Far cells in earlier studies. Therefore, we postulate that the key aspects of the first stage of stereopsis are: (i.) eye-of-origin conservation, and (ii.) a qualitative disparity map that provides error signals for ocular alignment. Strikingly, this depends on apparently monocular neurons. Stage 2 involves correlation cells underlying fine stereopsis. Implications for squirrel monkey will be discussed.
Meeting abstract presented at VSS 2013