Abstract
Color constancy supports object recognition by allowing the identification of a particular surface independently of illumination. We studied the stability of the color appearance of rendered and real Munsell surfaces under changing illumination with a color naming method.
In the first experiment, subjects named the colors of surfaces rendered under different artificial illuminants and displayed on a CRT monitor embedded in a lighting chamber. In the full-field viewing condition, the monitor background and the surrounding walls had the chromaticity of the illuminant. In the reduced cue viewing condition, the monitor background was set to black to remove local contrast between the stimulus and the illuminant. In the second experiment, subjects named the colors of real Munsell chips placed on a gray cloth in a room illuminated by natural or filtered daylight. In the reduced cue viewing condition, the cloth under the chips was changed to black.
Color constancy was quantified as the consistency of color naming across illuminants within each observer. We also analyzed naming consistency between observers for any given illuminant by counting the occurrence of same color names across observers.
Naming consistency for real surfaces was overall better and less affected by manipulations of illuminant cues than consistency for rendered surfaces. However, the pattern of naming consistency over stimulus hue, saturation and lightness was very similar for both stimulus types. Consistency was highest around most of the prototypical hues and increased with stimulus saturation. Naming consistency across observers followed the same pattern, indicating that observers might be more likely to agree on colors that remain perceptually more constant under illuminant changes.
We conclude that measurements with real and rendered stimuli under natural and artificial illuminants, respectively, give qualitatively similar results: categorical color constancy is not uniform across color space, but generally reaches its maxima around prototypical hues.
This work was supported by the German Research Council grant Ge 879/5-3.