cilynx
commented
1 year ago I have this working pretty well in a stand-alone script borrowing, correcting, and updating code from:
https://dontrepeatyourself.org/post/learn-opencv-by-building-a-document-scanner/ https://pyimagesearch.com/2014/08/25/4-point-opencv-getperspective-transform-example/
Why all these tutorials resize all images to the same aspect ratio regardless of source and then transform and dump out a squashed grayscale version, I'll never understand. Find the corners on a smaller version where you can control the blur kernel vs document resolution sure, but perform the transform on the original.
Still pondering how to implement this -- e.g. should this automatically run on every import? only jpgs? How do we handle reprocessing (#18)? What about multiple documents? Can we use the same edge detection routines to do #3 instead of sorting out whatever they're doing at the linked repo over on that one?
If I had a GUI grid/list working, I'd say this should be a one-shot option on a context menu.
There are a lot of times that I want to get a document into my library but all I have is my phone camera. Extracting a "good enough" document from crappy phone pictures is pretty much a solved problem these days, so we should support it.