Abstract
Perceiving surface colors categorically may help in identifying and recognizing objects in a scene. Although often investigated with abstract colored patterns, it is unclear how well categorical color perception is maintained in more naturalistic environments, such as natural scenes, with varying natural illuminations, as in everyday life. To address this issue, a categorical color-naming experiment was performed with simulations of hyperspectral natural images presented on color display, under daylight of correlated color temperature either 6500 K or 25000 K, each of duration 1 s. A test surface in the image, a sphere physically inserted in the scene, was covered with a Munsell color drawn from the category subsets of brown, pink, and purple, each comprising approximately 60 colors. The test-surface color varied from trial to trial, and each category subset was tested in each experimental session. Observers had to name the color of the test surface by pressing one of nine computer keys labeled: red, green, blue, yellow, pink, purple, brown, orange and neutral (black, gray, or white). The positions of observers' focal colors were determined by the peaks of the smoothed distributions of their naming responses plotted in CIE 1976 (u′, v′) space. To quantify the effect of illuminant, a focal-color “constancy” index was computed by a comparison of the focal-color shift with the ideal reflected-color shift due to an illuminant change. As with the standard color-constancy index used with e.g. asymmetric color matching, perfect constancy corresponds to 1 and perfect inconstancy to 0. Focal-color constancy reached approximately 0.7 for pink and purple, and 0.2 for brown. Although not as good as traditional measures of constancy, categorical color perception appears to be more robust than expected under illuminant changes, and may provide a useful anchor for surface-color judgments in natural scenes.
Supported by EPSRC Grant No. EP/B000257/1.