Neural Mechanisms in Boundary Grouping, Illusory Contour Generation, and Spatial Tuning of Receptive Field Selectivity

Abstract

A model for cortical boundary detection and contour grouping at visual cortical areas V1 and V2 is presented. The functional organisation of the model architecture is based on principles of adaptive resonance theory (Grossberg, 1980 Psychological Review87 1 – 51). The primary visual areas are connected in a bidirectional scheme of feedforward and feedback projections. Signal features are initially measured by orientation-sensitive neurons at the simple and complex cell level. These measurements undergo local competitive processing and subsequent nonlinear bipole integration of coherent activity (von der Heydt and Peterhans, 1988 Journal of Neuroscience9 1731 – 1748). The integration is based on an antagonistic interaction in position and orientation space utilising spatial coupling and compatibility functions for the evaluation of coherent activity distributions. This step can therefore be interpreted as implementing a process of matching neural codes of most-likely contour outlines against the noisy input distribution of local contrast measurements. The result of the integration stage is fed back to the first competitive stages to enhance compatible activations via multiplicative gating. The net effect produces grouping and illusory contour completion. In particular, the results of model simulations demonstrate that (i) the orientation selectivity of V1 cells is sharpened depending on the spatial arrangement and visual context, (ii) orientation-selective contrast measurements are enhanced at boundaries of spatially homogeneous stimulus arrangements and suppressed in the interiors, (iii) oriented cells at model V2 stage only respond to curvilinear input arrangements of activity at both branches of bipole cells, and (iv) model V2 cells show fine tuning for orientation selectivity and generate subjective contours to bridge gaps in arrangements of oriented contrast.