Abstract

Natural visual stimuli are multidimensional signals encompassing parameters of space, time, and color. Information in these signal dimensions is redundant because natural images exhibit some spatial, temporal, and chromatic regularities. Further, these signal dimensions are not independent. The incoming visual stimulus is processed and coded in the retina by spatially organized center-surround chromatically antagonistic receptive fields with complex temporal characteristics. The basic hypothesis is that the purpose of retinal signal processing is to reduce signal redundancy and to efficiently code visual information before transmission to higher stages of the visual system. Under this hypothesis it is shown that retinal coding is designed to obtain optimality in coding, considering the relative amount of information in each dimension, corresponding to a high bit rate in the spatial dimension and a low bit rate in the chromatic dimension. The coding in these dimensions achieves favorable image degradation minimizing sensitivity to picture-to-picture variations, considering the computational load required from the retina.

© 1986 Optical Society of America