Dithering is only effective to reduce fixed pattern noise; his color mottle is simply the result of transformed (by the debayering) random shot noise; the 'Color Mottle' is never in the same place, so dithering isn't required to fix that. Tony's solution to use dithering is just a very roundabout way of fixing up his initial error of debayering his frames with Adobe Camera RAW. It's really a less effective way of applying Bayer Drizzle (scroll down the full page) of the RAW frames. The latter procedure doesn't introduce the mottling in the first place.
Hi Ivo, great post, thanks for that.
I don't know if it's entirely accurate though to say that color mottle isn't in the same place. As I understand it, color mottle is largely the result of dark thermal signal, which is a property of the sensor. Sure, additional color mottle can be created by debayering prior to stacking, but most people avoid this.
Essentially, Tony was right to advocate for dithering, but it was overblown in his examples because of the debayering problem that he introduced in his workflow (using Adobe Camera RAW). Dithering is still very useful in getting rid of color mottle regardless.
I'd happily stand corrected on this, but I don't think there is any reason why warm and/or hot pixels would correlated; they should not result in mottling by themselves (mottling being defined as spots and blotches made up of multiple co-located pixels with roughly the same color). This 'mottling' only comes into existence after applying some types of filtering and debayering.
These are two GIFs demonstrating the mottling is *not* a fixed pattern;
First GIF 1; notice that the hot pixels stay in the same place.
Now GIF 2; a median filter was applied, removing high frequency detail/noise and leaving low frequency detail and noise ('blotches' aka mottling). It is now abundantly clear that any perceived blotches appear in random locations. Dithering is not going to help one bit with those.
The source are two CR2s (300s each at ISO 800) taken in sequence from a Canon T3i converted with dcraw with parameters -T -6 (TIFF and 16-bit).
Also notice that in GIF 1, the hot pixels are never single pixels (particularly not the blue and red hot pixels). This is due to dcraw's AHD debayering algorithm which (necessarily) bleeds noise into neighbouring pixels through the interpolation.
Edited by Ivo Jager, 17 November 2014 - 09:10 PM.