1. 8

  2. 1

    Tesseract seems mostly optimised for parsing blocks of text like pages from a book or receipts.

    I can say from secondhand knowledge that Tesseract was explicitly designed for magazine pages and that sort of thing. The colors will be fine (I think it just converts everything to luminance, but don’t quote me on that), but the layout/text detection will definitely suffer with this sort of thing.

    1. 1

      The output feeding it the raw screenshot was complete garbage - almost completely unusable. I almost want to investigate using some OCR lib that doesn’t do bounding box detection to see if I can do a better job, but I haven’t read too much into it yet.

    2. 1

      And jumping back in the other direction: Tesseract looks like a pretty neat engine. They claim to be octree-based, but I thought combining and deleting nodes was hard with octrees - how do they manage that?