require an update of japanese trained data

[horohoro]

I tried to run the OCR on an image with ¥ symbols and the engine was totally unable to match any of them. It usually translated them into "\ く".
The ¥ were "" for a UNC path (because on Japanese Windows, all the \ are replaced by ¥)

Also, all the number (except 0) were translated into their circled version.
For example:
1 -> ①
2 -> ②
3 -> ③
...
Japanese uses circled number some times (often compare to the rest of the world) but not that often. Number should still be translated into their normal form.

I think that these issues come from the training data that did not include ¥ and has as input number that were circled in the expected result.

I am quite a noob in ML and I do not know if I can and how to extract source data from traineddata file.

[amitdo]

You can try the Japanese.traineddata from the best or fast repo.

[gitvophu]

You can try the Japanese.traineddata from the best or fast repo.

Do you mean jpn.traineddata? isn't it?

[amitdo]

https://github.com/tesseract-ocr/tessdata_fast/tree/master/script

https://github.com/tesseract-ocr/tessdata_best/tree/master/script