Variation of melanopsin signals in natural scenes

Light incident on the eye generates both image-forming and non-image-forming responses. The image-forming response is mediated by three classes of cone photoreceptors, whose overall light sensitivity is described by the photopic luminous efficiency function, peaking at approximately 555 nm. The non-image-forming response is mediated by intrinsically photosensitive retinal ganglion cells, whose sensitivity is derived from melanopsin, peaking at approximately 480 nm. In natural scenes, the size of the two responses vary with the light incident on the scene and its reflecting properties. But can melanopsin signals in scenes be predicted from luminance signals? This question was addressed by computing luminance and melanopsin signals in over 30 hyperspectral radiance images of outdoor scenes under natural lighting, whose correlated colour temperature ranged from 3000 K to 20,000 K within scenes. The scenes contained mixtures of herbaceous vegetation, woodland, barren land, rock, and artificial objects, such as rural and urban buildings. For sampling purposes, the images were cropped to 1000  1000 pixels. Each was divided into 64 patches of 125  125 pixels to represent the effects of local scene properties. Luminance and melanopsin signals were calculated at each pixel, with allowance for prereceptoral lens absorbance. The relationship between the two signals in each local area was quantified by the correlation coefficient R2. Across scenes, the value of R2 averaged over local areas within each scene varied from 76% to 99%. But within scenes, the value of R2 varied much more from local area to local area. In one scene containing yellow flowers it ranged from 21% to 99%. These results suggest that luminance alone may be an imperfect predictor of melanopsin responses. The complex spectral structure of natural scenes and their interaction with illumination need both to be considered for reliable predictions of activity in intrinsically photosensitive retinal ganglion cells.