Abstract
Abstract: :
Purpose: Illuminant-independent visual discrimination of colored surfaces may be based on spatial ratios of cone excitations from distinct points (Nascimento & Foster, 1997, Proc. Roy. Soc. Lond. B, 264, 1395-1402). The aim here was to use information-theoretic methods to quantify the color information that might be available to the eye in representations by such ratios. Methods: Images of rural and urban scenes were obtained with a hyperspectral imaging system (e.g. Nascimento, et al., 2002, J. Opt. Soc. Amer. A, 19, 1484-1490), which provided estimates of the surface-reflectance function at 10-nm intervals at each point in a high-resolution digital representation. In computer simulations, scenes were illuminated by a 4300K daylight and by a 25000K daylight. Pairs of points were sampled randomly in each scene and labeled with their spatial ratios of long-, medium-, and short-wavelength cone excitations under the 4300K daylight. The change in this information under the 25000K daylight was quantified by the conditional entropy H, a measure of the uncertainty in one image given knowledge of the other. Results: Mean values of H based on 128 pairs of sample points in each scene were small, e.g. 2 bits, and varied little with the distance between the sample points in each pair unless they were close together. Conclusions: Spatial ratios of cone excitations are capable of representing most of the illuminant-invariant visual information available in natural scenes and, for the samples tested here, this information generally depends little on the distance between sample points.
Keywords: color appearance/constancy • scene perception • color vision