Abstract
Color constancy is often measured using asymmetric matching or achromatic adjustments. These methods are accurate but have limitations: achromatic adjustments allow the characterization of only one point in color space, and while asymmetric matching overcomes this limitation, measuring many points in color space is time consuming. Color naming allows quick sampling of a large portion of color space and the direct measuring of color appearance, making it a powerful tool for studying color constancy when coupled with appropriate data analysis methods. We asked observers to classify Munsell chips simulated on a monitor varying in hue, chroma and value to one of nine color categories under a neutral and four chromatic illuminants. Category boundaries were fitted to the naming data in order to determine the effect of illuminant change on color categories and on the convergence point of the boundaries. Observers were fairly consistent in their classification performance: 73% of the chips were classified in the same color category under four of the five illuminants and 43% of the chips under all illuminants. The chips that were categorized most consistently fell closest to category centroids and formed clusters in color space that were close to the unique hue loci. The fitted category boundaries and convergence points were highly similar under all illuminants, and color constancy calculated from the achromatic points was close to 90%. The convergence points were slightly shifted away from the direction of illuminant change, revealing incomplete adaptation to the illuminant. This undershoot was small, however, when compared to the change in the color signal from the surfaces under illumination changes. In conclusion, the classification of Munsell chips was based to a large extent on their reflectance rather than on the overall color signal, this being particularly clear for chips close to the loci of unique hues.
Supported by DFG grant GE 879/5.