Digits only recognized when mixed with letters

[Iman Firouzian]

Dear Friends,

Farsi Numbers are not recognized when they appear alone and not next to letters.

For example 

is recognized as: ۷۵ ۱ب

which is not correct.

But when it is mixed with letters as:

it is recognized as: ماه ۸,۱۷۵,۰۰۰

which is correct

Please help me with this

[Tom Morris]

On Wednesday, February 28, 2024 at 3:28:51 AM UTC-5 Iman Firouzian wrote:

Please help me with this

Please include more details about what version of the software you are using and which language (or script) model(s).

Tom

[Iman Firouzian]

I've installed it using:

!sudo apt install tesseract-ocr

in Google Colab.

it says it's the latest version: tesseract-ocr is already the newest version (4.1.1-2.1build1).

and the language model is "fas" and is installed by:

!sudo apt install tesseract-ocr-fas

thanks for helping

[Iman Firouzian]

Hi again,

I've tested it on windows and pycharm.

the tesseract version is tesseract v5.0.0-alpha.20200328

the result is roughly the same.

it would recognize correctly when numbers are mixed with letters. 
Any specific confugurations needed?

thanks for helping

[Philippe Argouarch]

I have a similar problem with the breton language, the lib does not recognize the verbal particle o and replace it by a zero 0 . oa which mean "was' in english becomes 0a 

philippe

[Tom Morris]

Thanks for the version and model information. That'll be useful for anyone trying to help.

My best guess is that there's something about the Farsi training data which is causing this, but I don't know what (and I don't speak Farsi). One thing you might try is using the Arabic script model and see if that's any better. Other than that, I'm afraid I don't have any good suggestions.

Tom

[Tom Morris]

p.s. You can find the Farsi training data here: https://github.com/tesseract-ocr/langdata_lstm/blob/main/fas/fas.training_text