Spatially coherent colour image reconstruction from a trichromatic mosaic with random arrangement of chromatic samples.
Résumé
Recent high resolution imaging of the human retina confirms that the trichromatic cone mosaic follows a random arrangement. Moreover, both the cones' arrangements and proportion widely differ from individual to individual. These findings provide new insights to our understanding of colour vision as most of the previous vision models ignored the mosaic sampling. Here, we propose a cone mosaic sampling simulation applied to colour images. From the simulation, we can infer the processing needs for retrieving spatial and chromatic information from the mosaic without spatial ambiguity. In particular, the focus is on the ability of the visual system to reconstruct coherent spatial information from a plurality of local neighbourhoods. We show that normalized linear processing allows the recovery of achromatic and chromatic information from a mosaic of trichromatic samples arranged randomly. Also, low frequency components of achromatic information can serve to coarsely estimate orientation, which in turn improves the interpolation of chromatic information. An implication for the visual system is the possibility that, in the cortex, the low frequency achromatic spatial information of the magnocellular pathway helps separate chromatic information from the mixed achromatic/chromatic information carried by the parvocellular pathway.