Abstract
Crowding, the marked inability to identify shapes in peripheral vision when targets are flanked by other objects, has been widely studied; however, the mechanism of crowding remains unsettled. Here, we attempt to provide the rudiments of a model that accounts for several widely accepted characteristics of crowding: (a) the spatial extent of crowding scales to half the target eccentricity (Bouma, 1970); (b) the zone of crowding exhibits a marked radial-tangential anisotropy (Toet & Levi, 1992), and (c) crowding is asymmetric in that an outward flanker (away from the fovea) is more effective at crowding a target than is an inward flanker (Bouma, 1973). Our model assumes a columnar architecture of the cortex, with columns packed hexagonally in cortical space. We assume that the receptive field sizes of V1 columns increase with eccentricity with a slope of 0.1 (Motter, 2002). We further assume that the initial receptive fields of higher cortical areas (V2, V3, V4) are constructed by recursive isotropic axonal projections in the cortical space, and that such connections can be both integrative and competitive (Reynolds et. al., 1999). A Hebbian learning scheme, assumed to be active during development, modifies and sharpens the initially homogeneous synaptic weights. We noticed that attention-mediated saccades, which bring a peripheral target to the fovea with a radial eye movement, can lead to anisotropy in the weight adjustments. Our simulations showed that connection weights are more sharply pruned along the tangential direction than the radial direction. The end result is that the zone of feature competition for a V4 neuron has the size and shape that agree with Bouma's Law and exhibit the radial-tangential anisotropy. Considerations of V1 receptive field layout (denser sampling toward the fovea) and the resulting precision in position coding (more precise toward the fovea) reveal the inward-outward asymmetry.