Improving of very old jpg-photos with low resolution using the command line

[linuxuser]

Improving of very old jpg-photos with low resolution using the command line

I discussed this topic already at other places and assuming other circumstances. It is all, what was said correct, but at the end there is no solution.

A lot of people say, darktable is not the best option to do upscaling and I agree.

@afre

As for processing, follow snibgo’s implicit advice and deal with the JPEG/camera-baked-in artifacts first before doing anything else. Otherwise, you are making the base image worse by enlarging and sharpening it further

I fully agree, that bad things get more worse, but I have no idea what could help. Maybe there are AI tools for the 1st step.

My idea is to create different versions of 1 photo (of thousands) with a script / command and choose the best then. So it doesn’t help to load images manually, this will take too long for thousands of old photos.

I have 2 different challenges:

All my files, which I post here, are licensed Creative Commons, By-Attribution, Share-Alike.

https://drive.google.com/drive/folders/1sYgptFeKkcTgPvxwGjwLfGHiLIpXkDXW?usp=sharing

The problem here is, that they are very unsharp.

From another camera there is:

https://drive.google.com/drive/folders/1OxaV43m9dkh4uELY8GvjidETvavC8adu?usp=sharing

Here is the problem, that the photos are oversharpened and have a lot of artifacts.

These photos are examples of the problems and I am not interested in advices to use a mask or whatever to improve some things of the examples. The photos are examples how the quality is generally. As already said, I want commands in a shell to create different versions of a lot of photos.

I played already with gmic and scale2x_cnn and / or deblur_richardsonlucy. But this is a 2nd step.

Is there a chance to prepare my sources before upscaling 2x?

[afre (Alan)]

I have been busy, so I cannot help you directly. However, I can provide you with some tips for efficient processing.

1. Process representative regions rather than entire images. It could be a crop to keep the centre or a sophisticated one that considers salience (points of interest).
2. If you are using G’MIC commands, I recommend batching within G’MIC. I have written posts about that. Also refer to G’MIC Reference. Folks on the forum could help too.
3. There is research on reversing image filtering. I have not had the time to explore such algorithms, but sample code is available for you to try. I suspect that they are not magic, but something we could try.

[linuxuser]

@afre

Process representative regions rather than entire images

Is this possible automatically? I fear it is not.

@linuxuser

My idea is to create different versions of 1 photo (of thousands) with a script / command and choose the best then.

It are too much photos to edit, so I am happy if they get somehow better. It is impossible to spend a lot time with every photo.

@afre

I recommend batching within G’MIC

You mean doing 1 gmic command and then the next? This would be easy with a script. I have to search for these posts.

@afre

There is research on reversing image filtering. I have not had the time to explore such algorithms, but sample code is available for you to try. I suspect that they are not magic, but something we could try.

Any help is appreciated!