Illuminants estimated by human observers for natural objects under daylights compared with those predicted by the optimal color hypothesis

Color constancy is a visual function to perceive the world to be constant in color under various illuminants. The visual system must estimate an illuminant in a scene to achieve color constancy. It was reported that the optimal color hypothesis (Uchikawa et al., JOSA, 2012) computationally predicted physical illuminants better than other candidate models based on the mean chromaticity or on the mean cone signals, for the distribution of 400 colors in a color space calculated with the spectral surveys of natural objects and daylights (Uchikawa and Fukuda, ICVS 2017). In the present study, we carried out a psychophysical experiment to compare the predictions of the optimal color hypothesis and human illuminant estimations in the same color distributions. In the experiment, stimuli were presented on a CRT, consisting of 217 hexagonal elements: a center test and 216 surrounding elements. Each element subtended 1-deg in diagonal. The colors of the surrounding elements were calculated using spectral reflectances of 215 natural objects and spectral power distributions of five test daylights, which were both measured in natural environments in Japan. Five male observers with normal color vision participated in the experiment. Observers adjusted the chromaticity and luminance of the center test element so that it appeared as a full white paper (paper match) under a test daylight. The results showed that mean illuminant estimations obtained for each daylight condition are fairly close to the predictions of the optimal color hypothesis although individual differences in illuminant estimations were found large.