Fixational eye movements and edge integration in lightness perception.

A neural theory of human lightness computation is described and computer-simulated. The theory proposes that lightness is derived from transient ON and OFF cell responses in the early visual pathways that have different characteristic neural gains and that are generated by fixational eye movements (FEMs) as the eyes transit luminance edges in the image. The ON and OFF responses are combined with corollary discharge signals that encode the eye movement direction to create directionally selective ON and OFF responses. Cortical neurons with large-scale receptive fields independently integrate the outputs of all of the directional ON or OFF responses whose associated eye movement directions point towards their receptive field centers, with a spatial weighting determined by the receptive field profile. Lightness is computed by subtracting the spatially integrated OFF activity from spatially integrated ON activity and normalizing the difference signal so that the maximum response in the spatial lightness map at any given time equals a fixed activation level corresponding to the percept of white. Two different mechanisms for ON and OFF cells responses are considered and simulated, and both are shown to produce an overall lightness model that explains a host of quantitative and qualitative lightness phenomena, including the Staircase Gelb and related illusions, failures of lightness constancy in the simultaneous contrast illusion, Chevreul's illusion, lightness filling-in, and perceptual fading of stabilized images. The neural plausibility of the two variants of the theory, as well as its implication for lightness constancy and failures of lightness constancy are discussed.