I’ve been fiddling around with how kwakk.info deals with vertical pages. The vast majority of magazine and zine pages are oriented “correctly”, of course, so this is a minor problem — but sometimes pages are printed vertically, and then you have to turn your laptop sideways to read those pages, which isn’t er ideal.
The OCR I’m using now doesn’t actually report page orientation, but it can be computed by looking at the bounding boxes for the lines — if they tend to be mostly vertical, one can make an educated guess. I mean scientific estimation.
So now you get automatic rotations:
It seems to be working OK… This is a pretty marginal problem, though — far fewer than 1% of the pages seem to be vertical. But there you go.
Hm, I wonder whether there are any pages that are upside down!
Yeah, but only a handful.
Anyway.