Abstract

The limits to lightness constancy can be measured by the accuracy with which human observers estimate the reflectance of an object as a function of various illumination conditions. Our paradigm is analogous to traditional psychophysical methods used for signal identification, with the reflectance representing the signal and the illumination serving as noise. We used computer graphic techniques to simulate monochrome scenes consisting of a plane grid of square pieces of paper with various initial reflectances. The squares were then non-uniformly illuminated. We measured the ability of observers to correctly identify the reflectance of a randomly chosen patch by selecting a matching piece of paper from a standard palette, also simulated. Performance was measured as a function of the number of patches, number of possible reflectance values, illumination contrast, and illumination spatial spectrum. With uniform illumination, observers could accurately identify only six reflectance values for the three conditions varying in number of patches. However, within this range of reflectance patches, reflectance identification was robust over the range of illumination contrasts and spatial frequencies tested. These results suggest interesting constraints to be considered in the formulation of algorithms for lightness constancy.

© 1986 Optical Society of America