Optimal nonlinear codes for the perception of natural colours

We discuss how visual nonlinearity can be optimized for the precise represe ntation of environmental inputs. Such optimization leads to neural signals with a compressively nonlinear input-output function the gradient of which is matched to the cube root of the probability density function (PDF) of th e environmental input values (and not to the PDF directly as in histogram e qualization). Comparisons between theory and psychophysical and electrophys iological data are roughly consistent with the idea that parvocellular (P) cells are optimized for precision representation of colour: their contrast- response functions span a range appropriately matched to the environmental distribution of natural colours; along each dimension of colour space. Thus P cell codes for colour may have been selected to minimize error in the pe rceptual estimation of stimulus parameters for natural colours. But magnoce llular (M) cells have a much stronger than expected saturating nonlinearity ; this supports the view that the function of M cells is mainly to detect b oundaries rather than to specify contrast or lightness.