Can you use Machine Learning to "enhance" scans of old album cover art?

Hi,

I hope I’m in the correct category for this topic - I’m an old-fashioned music collector that likes to sometimes get older audio CDs on the second-hand market.

While there is a good chance to find album cover art of recent releases (year 2000 and younger) on the internet, when diving further back you often have to scan the covers yourself if you want to have your digital library to look pretty.

Of course, this way you don’t have a real high-resolution source where the scanned images look nice on a computer display but printing artefacts get subjectively worse when seeing the scanned image of a printed cover on a screen.

I have absolutely no experience in this field but as a layman am quite amazed what is possible with Deep Fakes or the ML upscaling of old videos.

Would it be possible to use ML to “enhance” scans of printed images like album covers?

Thanks for any advice!

Regards,
aBavarian Normie-Pleb


There’s a web based version and local windows version

2 Likes

http://waifu2x.udp.jp/

Have used waifu2x for this exact purpose. Works great as long as you are not starting off with an extremely low quality image.

1 Like

I do not remember the software, but I recall a friend using a thing that did its best to convert pixel images into vector art. As a side effect it really improved the image quality if the image was a line art design. Photographs did not really work.

Thanks for the hint with waifu2x, will check it out.

@zlynx

Could you mean Vector Magic?

https://vectormagic.com/

If it wasn’t Vector Magic then it was very similar. That does look like the program I remember.

Adobe is playing around with ML in this space on and off for the last few years and even longer before ML was feasible.