Visual and Material Identity in Natural Scenes: Predicting How Often Indistinguishable Surfaces Become Distinguishable

If surfaces from a scene are visually indistinguishable under one light, they may become distinguishable under another. The aim here was to test whether the frequency of such metamerism can be predicted by a statistical property of the colours of a scene, namely their conditional entropy. Simulations were based on 50 hyperspectral images of natural scenes under sunlight and north skylight. The correlation between the logarithm of the conditional frequency of metamerism and the conditional entropy of colours was strong, with r = 0.80–0.87. Additionally, the more likely that indistinguishable surfaces were distinguishable under a different daylight, the more reliable the prediction by conditional entropy.