Abstract
Thresholds for discriminating hue have often been found to be better than those for discriminating purity, even when a common metric is used. This curious finding implies that color space is non-Euclidean. We have previously offered an explanation in terms of correlated noise in the midget (L/M) and small-bistratified (S/(L+M)) color channels (Danilova and Mollon, Proc. Roy. Soc. B, 2016). In the present study we consider a particularly instructive case, where discrimination of both hue and purity depends only on the L/M cone ratio. Consider a horizontal line running rightwards from the white point in the MacLeod-Boynton diagram. This is a line along which excitation purity increases and hue angle is constant. Consider now a line parallel to the first but close to the abscissa of the diagram. This is a line along which hue varies (as well as saturation). But along both lines, discrimination should depend only on the ratio of L:M excitation. The two lines differ in level of S-cone excitation but along a given line, the short-wave signal is constant. A four-alternative spatial forced-choice procedure was used to measure discrimination thresholds for foveal targets presented as brief (180 ms) increments on a steady, white, 10 cd.m2 field, metameric to D65. An advantage was still seen for hue discrimination, a result than cannot be explained by our correlated-noise theory in its original form.