[fallmonkey]

This is indeed very interesting to see and made me want to share my stories. I've also scanned over thousands of books (mostly Japanese manga, in a way to digitalize them and make them persist) myself and ever wondered where tech could bring this digitalization to. I've spent more than 10,000 hours on this for more than a decade and will probably do it until I couldn't.

Due to the requirement for maximum restoration of manga contents, especially for those cross-spine scenes, the high speed camera-based solution like above or other places which focus on formal prints won't work quite well and that leads to two major ways in the trade for decades since some Japanese started this self-scanning thing around 80s:

1) Destructively break the book into unglued pages (which also invovle cutting the glue with heavy-duty paper cutter) and send them to scanner with auto feeder, (then try to glue it back if keeping the book is vital, at best effort). 2) Manually press each page on flatbed scanner (thus A3 scanner works better as it allows vertically scanning spine area usually with the greatest shadow) and scan each page using predefined rough area which would therefore include non-content area as shown in : https://imgur.com/a/2ITBlJg (left/right edge and spine).

Solution 1) works great in terms of efficiency but could be a pain for those who love to keep book still in the collection (though malformed anyway, and I myself might consider this when I get old and couldn't bear the effort). 2) would cost a lot more time (on average, #1 takes less than 30 mins per book and #2 takes at least an hour) especially for manually removing noises as shown above. I've tried some basic CV tools/scripts to auto cut at least the edges (middle spine could be troublesome and I'm fine with manually working on that), which work really poorly since content itself could also be largely variant if we try to determine "edge" by checking pixel distribution.

Hopefully before 2nd decade finishes, I either give in to destruction or find a perfect way to automate the noise-clearing process.