My first attempt was to use OCR,
but I was unable to get satisfying results quickly.

However, I keep the code here for future reference,
in case someone (even myself) might find it useful.