Part of Advances in Neural Information Processing Systems 6 (NIPS 1993)
Daniel Ruderman, William Bialek
In order to best understand a visual system one should attempt to characterize the natural images it processes. We gather images from the woods and find that these scenes possess an ensemble scale invariance. Further, they are highly non-Gaussian, and this non(cid:173) Gaussian character cannot be removed through local linear filter(cid:173) ing. We find that including a simple "gain control" nonlinearity in the filtering process makes the filter output quite Gaussian, mean(cid:173) ing information is maximized at fixed channel variance. Finally, we use the measured power spectrum to place an upper bound on the information conveyed about natural scenes by an array of receptors.