Abstract
The subjective colors of visual figures can be strongly influenced by color and luminance contrast at the figural borders or edges. Edges can even generate false color that perceptually fills in the figure (Craik-O'Brien-Cornsweet, Watercolor illusions). Here, I discuss the relationship between such edge (i.e. oriented contrast) based filling-in phenomena and the neural mechanisms supporting color and lightness constancy. I propose a model of color computation within the ventral pathway in which high-level cortical computations seek to differentiate between reflectance-derived and illumination-derived contrast spanning a range of spatial scales in the retinal image. The outputs of these computations feed back to suppress the responses of neurons in early visual cortex to illumination-derived contrast. The responses of non-suppressed (active) neurons undergo contrast normalization prior to being spatially integrated to form a neural representation of surface color (likely in area V4). The model accounts for the results of achromatic color matching experiments performed with perceptually ambiguous displays in which instructions bias the observer to interpret local contrast as being due to either reflectance or illumination variation. It gives a precise quantitative account of the data, explaining instances of lightness contrast and assimilation, lightness constancy and its failures, and their dependence on instructions (attention). Contrast normalization plays a critical role in the theory by reducing response saturation in the neurons comprising the cortical color map; and it does so in a principled manner that takes into account the neural classification of local luminance and wavelength variation in the retinal image.
Meeting abstract presented at OSA Fall Vision 2012