Abstract
Natural illuminants — mixtures of skylight and sunlight - fall on an oblique line in the MacLeod-Boynton chromaticity diagram. This ‘caerulean line’ runs from approximately 476 nm to 576 nm. Chromaticities on one side of the line typically look reddish and chromaticities on the other typically look greenish. We have found that performance thresholds for color discrimination are minimal in the region of this line. We originally observed the effect in the parafovea, but have also found it in the fovea. For the latter measurements, the discriminanda consisted of the two halves of a circular 2-deg field divided vertically by a thin line. Each hemifield was independently jittered in luminance to ensure that discrimination was based on chromaticity. The background was metameric to Illuminant D65. Observers were asked to make a spatial forced-choice, indicating which hemifield was the greener, and feedback was given by auditory signals. To choose our stimuli, we rescaled the vertical axis of the MacLeod-Boynton diagram so that a line passing through 576 nm and the metamer of Illuminant D65 had a slope of -45 deg. We then selected a series of lines that intersected this line at 90 deg, and we measured discrimination along these lines. In a given block of trials, the chromaticities were always equidistant from a given point on the test line but their separation was increased or decreased according to the observer's performance. In different blocks, we probed different positions along the test line. Thresholds were expressed in terms of the factor by which L/(L+M) was changed. Discrimination was optimal near the caerulean line, and thus near the equilibrium hue where independent, phenomenological, measurements showed a transition from reddish to greenish.
These results reveal a performance optimum that approximately coincides with the phenomenological transition from red to green. Does the effect arise because the mapping of categories onto responses is simplest near the equilibrium hue, ensuring that observers make fewer response errors? To tackle this question (which will always arise where optimal discrimination is found at a phenomenological category boundary), we adopted a forced-choice version of a Same-Different judgment. On each trial, there are two presentations, 500 msec apart. In one presentation, the two target hemifields are physically identical and on the other they differ as in the earlier experiment. The observer reports merely which interval contains the differing stimuli. Enhanced performance is still found near the category boundary.