Training Tesseract for new fonts

[Umanda Dikwatta]

Hello,

I've been using Tesseract 4.1 for some time. I am using Tesseract with Sinhala language. I got good results for most of the images I tried. I trained Tesseract with different fonts. But as the documentation says, I had to preprocess my images to obtain good results. 

Then I tried Tesseract 5 with line images as .tif and the labels as .gt.txt. Then I used the generated .traineddata file to extract the text. But that didn't give me good results. I used image processing segmentation to obtain line images. Is it wrong to obtain line images using python segmentation? 

Could someone please explain me the possible reason?

Thank you very much

[Saman Kurdi]

Hello,

This might help.

https://www.mdpi.com/2076-3417/11/20/9752

Refards.

--
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 on the web visit https://groups.google.com/d/msgid/tesseract-ocr/40a95c6f-b459-4937-930f-1eb103bc4f82n%40googlegroups.com.

[Umanda Dikwatta]

Thank you very much for the link. Can we use non-unicode fonts as well? I have attached a sinhala font that I'm struggling to train. 

Thank you very much

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAH4VOMLc9f9choNcjUkJVNSt%3DHJazzxBNb-MfDeLvwVUqDMO7Q%40mail.gmail.com.