We examine the implications of the hypothesis that color information in the cortex is adaptively coded into a factorial (statistically independent) and gain-controlled representation. We show that this hypothesis explains the results of the recent experiments by Webster and Mollon on changes in color appearance following post-receptoral adaptation. We also give a neural network with a deterministically convergent, unsupervised learning algorithm that reproduces the adaptation observed.
Published in Vision Research vol. 33, no.1, p123-129, 1993