In an RGB scene, why can we assume color difference images are band-limited?

Question

A common assumption in image processing is that the difference between two color channels of the same scene are band-limited.

For instance, the interpolation method used in Hirakawa's paper introducing Adaptive Homogeneity-directed Demosaicing fundamentally uses the assumption that (G-R) and (G-B) images of the same scene will be bandlimited relative to R, G and B images.

This assumption is also used to motivate decomposing video signals into Luminance/Chrominance components to save on data rate.

I feel like I am missing something very simple here. What justifies this assumption, and can it also be made for (R-B)?

Answer 1

It isn't the scene that is bandlimited in the color channels, it is the human visual system. The color-sensitive cones are not nearly as dense as the intensity-sensitive rods in the retina.

Therefore, the argument is that it is not necessary to preserve as much color bandwidth when transmitting or storing images or video, because the human visual system can't take advantage of it anyway.