Using Tesseract for Handwriting..

[Peter Geraghty]

Hi everyone!

Recently started a project attempting to use Tesseract for handwriting recognition. Anyone's thoughts or inputs would be greatly appreciated. 

Also, would contributions to this project aimed at recognizing handwriting be a welcome addition? 

Also, I've done some research on the components of tesseract and wanted to be sure that I understood which legacy components have been replaced:

- The Polygon approximation algorithm has been replaced by a recurrent Neural Network (character recognition)

-  A line selector has been replaced by a neural network of some kind.

For the Tesseract we're using, the NN has been replaced by a custom trained model. However, my understanding is that word and character recognition are still using the older algorithm.

Thanks for any input,

Peter G.

[Peter Geraghty]

sorry, by word recognition, I meant word and character localization.

[Shree Devi Kumar]

Please see https://github.com/tesseract-ocr/tesstrain/wiki for detailed examples of tesseract training for handwritten texts.

--
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/33561418-416a-44ca-a25e-639f3ea3427an%40googlegroups.com.

[Peter Geraghty]

Thank you!!! will do!

[Shree Devi Kumar]

Also see the Technical Information section in 

https://tesseract-ocr.github.io/tessdoc/

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/fb756dd5-251e-4367-bfa1-338970b474b9n%40googlegroups.com.

[Peter Geraghty]

Thank you again for your help this far. Does the RNN implemented in tesseract 4.0 use a connectionist temporal classification for outputs?

I seem to have difficulty navigating through the repo. Most of the documentation just refers to it as an RNN.

Thanks!