Abstract
In order to achieve color constancy, the visual system needs to estimate an illuminant and discount it to retrieve surface colors in a scene. Many proposed algorithms only succeeded in part in accounting for this color constancy process in the visual system. Therefore we need a more comprehensive model to be applied for general scenes. Uchikawa et al. (2012) proposed the optimal color hypothesis, in which the visual system would choose the best-fit optimal color shell for a given chromaticity-luminance distribution of a scene to estimate an illuminant. The optimal color shell is uniquely determined in a chromaticity-luminance color space for a given illuminant. Kusuyama et al. (ICVS, 2015) formularized this hypothesis based on the least square method, and then verified this model by comparing illuminants calculated by this model with those obtained by observers in psychophysical experiments. The present study aimed at validating this model using other scenes. We compared performance of this model with those of other models based on mean chromaticity, mean L, M, S responses and chromaticity-luminance correlation. In experiments, observers adjusted chromaticity and luminance of a center test stimulus surrounded by 60 stimuli with no spatial gap so that it appeared as a full-white paper. We manipulated the chromaticity-luminance distribution of surrounding stimuli. Three black-body illuminants, 3000K, 6500K and 20000K were used as test illuminants. It was found that our optimal color model predicted observer's settings equally well for redness and blueness directions in MacLeod-Boynton chromaticity diagram in most conditions better than those of the other three models. These results confirmed that our optimal color model works well at least for conditions tested, suggesting that the visual system could estimate an illuminant utilizing the optimal color shell.
Meeting abstract presented at VSS 2016