Color constancy compared: Experiments with real images and color monitors

Purpose. Color constancy is often studied with stimuli displayed on color monitors. We seek to understand whether performance measured with such stimuli agrees with performance for more natural images. Methods. In our first experiment, observers viewed stimuli in a laboratory room where the illumination was under computer control. Observers used a projection colorimeter to adjust a test patch until it appeared achromatic. Settings were made under two different illuminants. In our second experiment, the same observers viewed stimuli displayed on a color monitor. These consisted of images of the central 46° h by 35° v of the stimuli used in the first experiment. We acquired the images with a hyper-spectral camera that provided spectral data (400 nm to 700 nm at 10 nm increments) at every pixel. We used the images to produce monitor stimuli that were point-by-point metamers of the stimuli used in the first experiment. We placed observers 13.4 from the monitor so that the test patch subtended the same visual angle (4.3° h by 5.7° v) in the two experiments. We again had observers adjust the test patch so that it appeared achromatic. Our two experiments were thus well-matched in terms of stimuli, psychophysical method, and observers. R suit. In both experiments, the achromatic point shifted with the illuminant The shift was larger for the real image experiment. In terms of a simple constancy index, three observers showed 83% constancy in the real image experiment but only 58% constancy in the monitor experiment. Conclusions. Experiments that study human color constancy using monitors are likely to underestimate its degree. Supported by EY10016, EY06745, and a grant from the Hewlett-Packard CcHporatkm. C5.