Best Traineddata - Kannada

tesseract-ocr / tessdata

Trained models with fast variant of the "best" LSTM models + legacy models

Apache License 2.0

6.38k stars 2.18k forks source link

Best Traineddata - Kannada #81

Open Shreeshrii opened 7 years ago

Shreeshrii commented 7 years ago

Forwarding feedback regarding Kannada

Main obsevation of new version is,

ದ that there is improvement but still there are many problems with kan.traineddata file.

recognition of compound words of ನ and ಯ are very often bad.
ಂ anuswara very many times appears as ರಿ, ಲ, 0, ಠ, ರ.
arkavattu as ೯.
mis recognition of
- ಚ as ಜ;
- ಷ as ಪ
- ದ್ಧ ದ್ದ and viceversa,

ಯ as various alternatives make a cogent Postprocessing also difficult.

belurusudarshana commented 5 years ago

Hi I am interested in knowing people who are working for Kannada OCR development. Hence I joined this forum. Kindly let me know how I can discuss these things with you. I am now working as consultant at CIIL, Mysuru. We are interested in supporting the Tesseract development activity for all the Indian Languages.

Shreeshrii commented 5 years ago

@belurusudarshana You can post your queries in the tesseract-dev and tesseract-ocr forums/mailing lists.