“ I can’t help wonder if one could use machine learning to identify title text from the image, thus eliminating the manual input of any information, except for price maybe.”
There are far, far too many separate editions of most books (typically at least two — hardback and paperback — though often several more, particularly with classics, or titles that have been regularly reprinted across decades) for this to work, I’m afraid. Title text is not nearly enough information to identify the edition (or even imprint) of any book. Even cover images (particularly, again, for classics) are reused, so that’s not going to work, either.
My point is, I guess, having worked with collectible/antiquarian for some time now, identifying editions and versions of books is more complex than can currently be automated. Plus, it seems to me, any time people try to introduce automation to processes, they always underestimate how much work is done originally by the human element, and tons of mistakes are made as a result.