Abstract
Color helps inform us about the contents of the natural world. In principle, the more colors there are in a scene, the more individual surfaces that can be perceptually identified. But an analysis based on the size of the gamut of colors ignores how identification is affected by the distribution of colors within the gamut; for example, surfaces with rare colors are easier to find than surfaces with common colors. To take into account the effects of color distribution on identification, information-theoretic methods were used to estimate the number of perceptually identifiable surface colors, N say, in each of 50 natural scenes. Surface colors were specified numerically in the approximately uniform color space CIECAM02, and a nominal color-discrimination threshold was set to 0.6 units, although the exact value was not critical. When the gamut of each scene was given the uniform distribution, allowing comparison with estimates based on gamut volume, it was found that N ranged from 7.4x10^4 to 1.0x10^6. When, instead, the gamut of each scene was given its true nonuniform distribution, it was found that N was much reduced but more wide-ranging, from 4.0x10^3 to 3.0x10^5. There was little correlation over scenes between the reduction in N and gamut volume. In practice, therefore, it is the gamut distribution, not the gamut volume per se, that sets the bounds on surface-color identification in natural scenes.
Meeting abstract presented at OSA Fall Vision 2012