Predictability of melanopsin signals from luminance signals in natural scenes

The response of the human eye to incident light has two components, one image-forming and one non-image-forming. The image-forming response follows luminance spectral sensitivity, peaking at approximately 550 nm, and the non-image-forming response follows melanopsin spectral sensitivity, peaking at approximately 480 nm. But the two signals may be interdependent, especially in natural scenes with large illumination variations. To test this hypothesis, luminance and melanopsin signals were computed in over 32 hyperspectral radiance images of vegetated and non-vegetated outdoor scenes under natural lighting, with correlated colour temperature 3000 K to 20,000 K within scenes, and in time-lapse images with daylight correlated colour temperature 3000 K to 7500 K. With allowance for prereceptoral lens absorption, luminance and melanopsin signals were computed at each pixel and grouped into 64 patches of 155  125 pixels within each 1240  1000 pixel image, and their correlation quantified. Across scenes, mean R2 over patches ranged from 76% to 99%. Within scenes, R2 varied more widely, with minimum less than 10%. These results suggest that although the two signals are broadly correlated, the predictability of melanopsin signals from luminance signals in any given scene is limited, presumably owing to the complexity of natural scenes.