An Outline of a Neural Architecture for Unified Visual Contrast and Brightness Perception
In this contribution a neural architecture is proposed that serves as a framework for further empirical as well as modeling investigations into a unified theory for contrast, contour and lnightncss perception. The computational mechanisms utilize a center-surround antagonism based on shunting interactions which allow to multiples contrast. as well as luminance data. As a key new feature, this data is demultiplexed at a later stage into segrcgated processing streams that signal both local contrast information of each polarity, and a scaled, low-pass filted and compressed version of the luminance information respectively. In correspondence with recent findings about the major processing channels in the primary visual system, the ON and OFF contrast channels feed into a subsystem for contrast processing, perceptual organization, and grouping (boundary contour system, BCS). The activity in the Segregated luminance path, however, is hypothesized to he contrast enhanced via shunting interaction, utilized hy the coc~xisLing contrast. channels. Following Grossbergs FACADE architecture, it is suggested that activity generated in the BCS acts as a modulation mechanism that controls the local diffusion coefficients for lateral activity spreading within the segregated brightness&darkness (B&D) channel. A three stage process is suggested for brightness reconstruction and filling-in. Based on the segregation of ON and OFF contrast information and basic neural principles such as divergence, convergence, and pooling, the nrodel accounts for the linear response properties of odd and even symetric simple and complex cells in VI. Theoretical analysis of the network's function at various stages of processing, provides a framework for quantitative studies referring to available data on visual perception.