Closed howff closed 1 year ago
Partially 'solved' by downsampling (resizing) the image before OCR. Implemented via this commit https://github.com/SMI/dicompixelanon/commit/446f8dd0a87b6ec7bbb78724e193372a3fdc030d
Testing involved different reductions (by 2 or 3), and different methods such as nearest neighbour, bicubic, etc.
INTER_NEAREST
was best on the test data, but may have been a fluke, so chose NEAREST_AREA
.
Train the OCR to recognise dot matrix font, for example