it also doesn't require new 5-channel image formats. Oversimplifying (neglecting various weighting factors and gamma curves and the like), Y is min(R,G), C is min(G,B). If C+Y > G then they're both decremented by G-(Y+C). Then R'=R-Y, G'=G-Y-C, and B'=B-C, and you display the colors as R',Y,G',C,B.
The end result is that you get a yellow which doesn't also partially trigger blue receptors (which real-world yellows don't, as they are NOT red+green), and a cyan which doesn't also partially trigger red (same reason).
Register
And, to add to Francis Vaughn's excellent post,
it also doesn't require new 5-channel image formats. Oversimplifying (neglecting various weighting factors and gamma curves and the like), Y is min(R,G), C is min(G,B). If C+Y > G then they're both decremented by G-(Y+C). Then R'=R-Y, G'=G-Y-C, and B'=B-C, and you display the colors as R',Y,G',C,B.
The end result is that you get a yellow which doesn't also partially trigger blue receptors (which real-world yellows don't, as they are NOT red+green), and a cyan which doesn't also partially trigger red (same reason).