A reader commented on the previous post:
The real tough part of posterization is when a gradient is colored and you get hue shifts as the color channels step across bins individually.
Can you redo this test with a color gradient, like maybe 0.6 0.8 1.0 relative values to the channels?
Pleased to do so. Here are the results of tests with various combinations of color plane weightings of 1.0, 0.8 and 0.6, with the read noise starting on the left at zero and finishing on the right at 1.5 LSBs, a 3 bit ADC, and the average levels running from an ADC count of 3 on the top to 5 on the bottom:
You can see the hue shifts that the reader predicted on the left hand side of the images. You can also see that they are gone by a third of the way across the image, which is a read noise of half an LSB.
Thus the ramps with different color plane weightings give the same result as those where all the color planes have equal weight.
Here’s the Matlab code that made the above images:
If anyone wants that code in a text file, I’ll be happy to oblige.
Huh, you went as far as including demosaicing in the process?
I did it simpler (my code was about 1/4 as long as yours) and came up with the same result.
I included demosaicing because I remember Iliah Borg saying at some time that posterization was affected by demosaicing. It seemed like a small effort to forestall a set of possible objections later on.