I found this thread at Magic Lantern forum DIGIC 8 'PowerShot' development (M50, SX70, SX740) and I wonder how a banding correction like that works and how to do it myself?
Edit: added underexposed raw file with bandning and a dark frame.
CC0
IMG_1094.CR2 (22.5 MB) IMG_1093.CR2 (23.3 MB)
You can add dithering. Dither - Wikipedia
It’s amethod to add noise to the image signal to randomly distribute quantization noise instead of having it uniformly distributed in bands.
Some of the more complex dithering techniques, like Floyd-Steinberg work really well, but even a random dither will work OK.
There’s also a banding denoise filter in G’MIC (Repair>Banding Denoise)
I can’t see the images in the link currently (they all come up as blank here, I’ll try on another machine tonight), but it sounds like he’s not talking about “banding” due to quantization error, but is instead talking about banding due to rolling shutter artifacts in artificial light.
Edit: I finally can see those images. It’s definitely not quantization error, and it’s unlike any rolling shutter artifact I’ve ever seen before. Usually RS artifacts are a repeating pattern with the size of the bands dependent on readout rate and light modulation frequency. (many artifacts at line frequency or 2x line frequency, so 50/60 or 100/120 Hz, but with modern RGB lighting, often you get banding with varying colors depending on the relative PWM frequencies. Philips Hue bulbs are especially nasty in this regard, PWMing at around 600 Hz)
1 Like
And this?
Sorry, @Peter, I don’t want to hijack your thread, just add more (related?) info and maybe help find a solution.
EDIT: Now addressing your question, I followed @sguyader’s method on dpreview
Before:
After:
I pulled down the H Cutoff slider to 0.2 and raised Space and Value till the point where the bright areas aren’t affected. If you increase them too much, you begin getting distortions on the bright areas:
(see the bright strip at top left?)
Maybe we could use it this aggressive way and then mask the bright areas in Gimp.
Tried dithering in darktable. Only thing that came close to remove the banding was random and -20dB. Not close the the banding removal at Magic Lantern.
I have uploaded raw sample in topic. I understand underexposing will give me banding but Greg at ML was able to remove it.
Thanks for sharing the raw files.
This is the best I could achieve with gmic banding denoise.
Before:
After:
I got the impression that the algorithm works fine, but the the sliders scale should be broader so that we can treat the larger bands (I’m thinking of the red large bands showing between the box and the camera strip.)