ambuda
ambuda copied to clipboard
OCR doesn't work well on two-column text
See https://ambuda.org/proofing/isavasyopanishad/1/
This:

Becomes:

It works well on some two-column text. Perhaps here the columns are too close for Google OCR to tease these apart.
If better detection is infeasible, one path forward might be to ask the user to select the region in which OCR should be applied.