Briefly, we decompose the image via the fast Fourier transformation (FFT) and mitigate the repeating pattern using the magnitude component. For this image, it looks like this:
The star, lines and haze propagating from the middle are good. Attempt to remove the other ones; for example, with the clone or heal tools in GIMP. Be careful though, this method can introduce new unwanted patterns.
If that’s the case, then I suppose a interactive FFT Painting filter in G’MIC would solve this problem? That’s something I wanted to make, but avoided b cause it’s so time-consuming making a interactive filter.
@Lee1 You may clone/heal beforehand; however, it is best to remove periodic artifacts (repetitive patterns) using Fast Fourier. The disadvantage is that you may potentially weaken real structures such as the fence, roof, pant legs and text; and introduce new distortions.
People mostly cover the stars and lines using solid or feathered black circles but I use clone/heal tools because, in attempt to reduce the possibility of introducing new distortions, I try to make the magnitude portion of the decomposition as close to its neighbourhood as possible.
Workflow FFT (polar) decomposition → remove all stars and lines except for central ones → re-composition → observe effect and adjust if necessary
@Reptorian Let’s try a trimmed mean instead. The trimmed median method is a riff off that and may not be as practical. (The earlier sample looks weird probably because I normalized it.)
@Reptorian This version uses the trimmed median’s second finest details scale to guide the trimmed mean’s finest. For a much higher cost, the returns are small.
You misread my post. None of the sample images are FFT-based restoration but rather trimmed median and/or trimmed mean.
This is from FFT restoration. I did this without clone/heal, only covering off-centre stars/lines. As you can see, doing so unfortunately weakens the subjects’ edges as well.
I’m working on the same problem. The best result was from the G’MIC “descreen” by Andreas Påhlsson. The preview showed an almost perfect result, but running the filter left vertical-line remnants not visible in the preview?
@Lee1 Please give FFT restoration a try and report back. If you are stuck along the way, we can give you pointers.
@okieman There is no way around the inaccurate preview. The only workaround I can think of is doing apply and checking to see if it looks okay in the host app; if not, undo, adjust parameters, then repeat.
BTW, if you want to explore, @Iain has a few restoration filters you may want to try.