I meant cleaning up the scanlines to make it easier to sample the pixel colors.
After running a Fourier transform you only need to paint over the obvious bright spots to remove these regular artefacts (see the linked article for a demonstration).
I mean, I'm not about to run any tests but I think you're underestimating how much work the human eyes do to iron out the little inconsistencies. I've done a lot of work with processing digitized images, and algorithmically converting a brand new 600dpi industrial flatbed scan of well-preserved black-and-white typed page into a 1-bit black-and-white tiff would often yield noisy, and sometimes unusable results for relatively clean looking images because our eyes see right through artifacts that most algorithms can't. I'm sure you could easily do the first 90% algorithmically– which is probably good enough for most developers— but for an artist to get the accuracy he was looking for, that last 10% would almost certainly take longer than the few hours he put into it.
