Handling text scans and cleaning

[Ajinkya Bobade]

I have noticed that text cleaning is the most difficult part in OCR pipeline. I have struggled alot on this part, without properly cleaned text OCR simply fails in terms of accuracy. In order to handle text cleaning seperately I created  a GitHub repo that uses AI to clean up all text in a image. Once the text is cleaned we can choose our own custom OCR models on it. I have personally seen OCR accuracy shoot up to 99% on a properly preprocessed and cleaned image. 

Here is a Github: https://github.com/ajinkya933/ClearText link. 

Regards 

Ajinkya

[Zdenko Podobny]

Thank your tool - it is already listed in tesseract doc:
https://github.com/tesseract-ocr/tessdoc/blob/main/User-Projects-%E2%80%93-3rdParty.md#4-others-utilities-tools-command-line-interfaces-cli-etc

Zdenko

ut 8. 4. 2025 o 6:09 Ajinkya Bobade <ajinkya...@gmail.com> napísal(a):

--
You received this message because you are subscribed to the Google Groups "tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-oc...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/tesseract-ocr/CAHy6iNOjhs7ZY7r26fGzqJOUr2e%2BF3bY%3DeDCHjM-VD7XH5M%3DTA%40mail.gmail.com.

[Ajinkya Bobade]

Thank you, just saw from your link that it is posted !! 
I'm so glad to hear this news  

Ajinkya