Abstract
Computer simulations are presented (including a video) of a dynamic model of lightness computation by visual cortex. Input images consist of: 1) a staircase Gelb display (grayscale papers arranged from darkest to lightest) and 2) two related displays in which the papers are spatially reordered. Fixational eye movements are modeled as 2D brownian motion. As the image moves relative to the retina, ON and OFF cells are transiently stimulated by the paper edges. The two cell types have different neural gains (0.27 for ON and 1.0 for OFF), consistent with physiological data from macaque LGN. At a subsequent cortical stage, ON cell responses are spatially integrated to form a cortical lightness map, and OFF cell responses are integrated to form a cortical darkness map. The spatial integration is performed by large receptive fields and occurs in the direction of luminance change at edges with an exponential falloff (space constant = 1.8 deg). The directionality of the spatial integration is opposite that of the eye movement direction at the time of ON or OFF cell activation. The difference between the lightness and darkness maps is used to compute an achromatic color map, which models the conscious percept (model output). Activation of the color map occurs with an exponential time constant of a 100 ms and decays with a relaxation time of 3 seconds. When fixational eye movements are artificially halted, the color map decays, resulting in a gradual fading of the conscious percept. The model properties are consistent with both known cortical physiology and the phenomenology of achromatic color perception. Furthermore, the model reproduces both the exact quantitative properties of lightness matches obtained for papers in the input displays and the approximate temporal properties of both the onset of conscious visual percepts and their decay in the absence of eye movements.