Computing chromatic adaptation

Most of today’s chromatic adaptation transforms (CATs) are based on a modified form of the von Kries chromatic adaptation model, which states that chromatic adaptation is an independent gain regulation of the three photoreceptors in the human visual system. However, modern CATs apply the scaling not in cone space, but use “sharper” sensors, i.e. sensors that have a narrower shape than cones. The recommended transforms currently in use are derived by minimizing perceptual error over experimentally obtained corresponding color data sets. We show that these sensors are still not optimally sharp. Using different computational approaches, we obtain sensors that are even more narrowband. In a first experiment, we derive a CAT by using spectral sharpening on Lam’s corresponding color data set. The resulting Sharp CAT, which minimizes XYZ errors, performs as well as the current most popular CATs when tested on several corresponding color data sets and evaluating perceptual error. Designing a spherical sampling technique, we can indeed show that these CAT sensors are not unique, and that there exist a large number of sensors that perform just as well as CAT02, the chromatic adaptation transform used in CIECAM02 and the ICC color management framework. We speculate that in order to make a final decision on a single CAT, we should consider secondary factors, such as their applicability in a color imaging workflow. We show that sharp sensors are very appropriate for color encodings, as they provide excellent gamut coverage and hue constancy. Finally, we derive sensors for a CAT that provide stable color ratios over different illuminants, i.e. that only model physical responses, which still can predict experimentally obtained appearance data. The resulting sensors are sharp.