Dither loses information, so perfectly reversible might be an unreasonable goal. But I bet you could train or fine-tune a denoiser to do a really good job. This might make a cool paper, if there’s anything at all non-standard or surprising you discover along the way.

No, although AI gets really close. Dithering represents adding noise to an image and also represents doing a lossy type of compression.

It's a somewhat active inverse problems research subfield https://github.com/MenghanXia/ReversibleHalftoning

Do you want to undither, or un-half-tone, like the images in TFA?

Unhalftoning is called descreening.

I use this:

https://descreen.net/