Abstract
Luminance is at the foundation of vision science and lighting technology. Its standardization marked an important step by including the human visual system in the characterization of lights. Thus, the candela as an S.I. unit refers to the human luminous efficiency curve for the comparison of the intensity of lights of different spectral distributions. Luminance has the property that it is additive. However, it is well-known that it is severely deficient for assessing the perceived intensity of the steady lights that are predominant in our environment. Observers had to rank order of 12 color patches varying in hue, saturation, and intensity according to their perceived brightness in each of 66 trials. In addition, observers performed heterochromatic flicker photometry and unique yellow settings to gain information about their individual display environment. We collected one data set in a well-controlled lab environment and two data sets online using Prolific. We used the first group of online observers (N=99) to estimate the optimal group weights of R, G and B for predicting the correct rank orderings of the second group (N=110). Observers were fairly consistent in 90.3% of unambiguous rankings. Luminance predicted 75.5% of the rankings correctly, but linear (sum) and non-linear (max) models with different weights for RGB performed significantly better at 80% percent correct or above, with the max model performing the best at 82%. Using weights based on the average radiance of several displays that we calibrated performed better (78.7%) than luminance. The best fitting models gave equal weights to L- and M-cones and a significant 17% weight to S-cones. The results agreed well with the lab-based experiment (N=43). Our results strongly indicate that luminance should not be used to specify the effectiveness of lights of different spectral distributions.