Abstract
The visual system coarsely samples the light spectrum with only three classes of cones to represent color with the perceptual attributes of hue, saturation, and lightness. We have examined potential inferences these attributes might embody about the underlying color signal, by asking how spectra that share the same hue but differ in saturation are related. The hues of many wavelengths change when white light is added, an interaction between hue and saturation known as the Abney effect. However, when the spectrum is instead diluted by broadening the bandwidth (e.g. in spectra with Gaussian profiles), hues tend to remain more invariant. This suggests that the visual system may attempt to represent the spectrum in terms of its predicted centroid and bandwidth, and thus is tying constant hues to constant inferred properties of the environment (e.g. the spectral peak), rather than to constant physiological responses (e.g. the relative responses of color-opponent mechanisms). By this account, the conventional Abney effect occurs because the visual system is tricked into making the wrong guess because the spectrum is unnatural. A Gaussian inference provides a good approximation to hue percepts at shorter wavelengths and predicts some differences between hues in the fovea and periphery, but fails to explain color appearance at longer wavelengths, pointing to a potential role of S cones or screening pigments in shaping color inferences. A Gaussian model is also a reasonable inference for a trichromatic visual system to adopt because it approximates natural spectra roughly as well as comparable linear models.
Meeting abstract presented at OSA Fall Vision 2012