OCR and Image Load Questions

[nark31]

It may be awkward because it is a sentence using a translator.

I'm converting it to a Korean server.
There is a phenomenon that some character recognition is grouped into two or more and printed out, perhaps due to a font problem.

Characters that should be divided into 'li', '0', '0', '1', and '2' are combined and recognized like a picture.
Is there a solution to the above problem?
Also, I want to change the image to match the Korean server, so is there a function to change the file in real time and load it?
Sometimes the recognition rate is low even though it is the same image/coordinate.

Thank you.

[NateScarlet]

text segementation is hand written by my self

enable verbose log by set env var DEBUG to auto_derby.ocr

you can try do the same or use a third party ocr engine

[NateScarlet]

i don't undersand second question well.

if you want reload template without restart, call auto_derby.template._LOADED_TEMPLATES.clear()