Abstract

A compelling argument in Retinex experiments is that observers can discriminate papers with different spectral reflectances even when the spectral distributions of reflected light from the papers are adjusted to be identical. For example, Land¹ states that “… the two objects [give] identical readings on examination with any sort of physical light meter. In spite of these identical inputs of radiation to the eye, the objects are perceived to be completely different in color.” However, calculations based on a plausible realization of the Retinex experiments show that identical reflected lights, although constrained to match radiometrically through three filters, are still discriminable in chromaticity. The artifact is due mainly to the large bandwidths of the filters, which allow variations in the spectral distribution of the reflected light within their spectral extent, even though the integrated energies as measured by a radiometer are identical. Colorimetric variations between the Munsell test-color samples in the CIE Color-Rendering Index were ~22 times larger than the area of the nearest MacAdam ellipse using standard 50-nm bandwidth Retinex filters. Variations ~8 times larger were found using 10-nm bandwidth filters. Furthermore, the colorimetric variations turn out to be in just such directions that the object-color hues can be determined directly from the chromaticity space without Retinex processing.

© 1987 Optical Society of America