This should be easier in software, reconstructing a 3D model of the relaxed open book from a stereo or multiple photos, then using AI to "upsample" to the scalable PDF document most likely to produce the modeled image.

I was part of a font consulting company during the Postscript / Truetype font wars, and we reconstructed fonts from scans or earlier digital formats. Most of the work was fixing bad data. This all should be easier now; think of Peter Jackson cleaning up the "Let It Be" sound, leading to the Beatles releasing one more track.

It baffles me that book images don't get this quality of attention. As a mathematician I spend a lot of time reading old journal articles that look terrible.