This questionnaire in which people were asked to name colours takes no account of eg the adjacent hues or edge conditions, but it’s fascinating, nonetheless.
The map at the bottom was surprising, since you might expect a more uniform, rays-of-the-sun distribution within RGB space. Instead, it seems many more shades are labeled red than blue, for example.
If it’s true that we can discriminate many fewer blues than reds, then this immediately offers a new colour image compression algorithm.
At its simplest, each pixel in an image would be assigned a certain colour ‘depth’ in terms of bits. Fewer bits could be assigned to local shades of blue than to shades of red, for example, in proportion to the areas indicated on the XKCD map.
This could be extended to optimise discrimination at the boundaries between adjacent regions of the map (so that more bits could be allocated to emphasise the difference between eg a yellow and an orange).
