How you can train tesseract 4.0 LSTM for receipts

[Ahmad Moawad]

Hello All,

I want to train tesseract 4.0 LSTM for receipt, So what I am asking related to:

1. Training based on image
2. Image processing
3. Add new words to the dictionary
   

• I have read the documentation and I think the good option is: Finetune. So I need to provide box/tiff before training.
  

• I know this command will create box file in under directory in /tmp, So should I edit the box file here or edit and provide it to this command in this case how can I provide it to this command.

training/tesstrain.sh \
 --fonts_dir /usr/share/fonts \
 --training_text ../langdata/ara/ara.training_text \
 --langdata_dir ../langdata \
 --tessdata_dir ./tessdata \
 --lang ara \
 --linedata_only \
 --noextract_font_properties \
 --exposures "0" \
 --fontlist "Arial" \
 --output_dir ~/tesstutorial/aratest

• for the image processing I am using the libraries that provided in documentation:https://github.com/tesseract-ocr/tesseract/wiki/ImproveQuality , if there are another options for image processing please tell me.

• for Adding new words to the dictionary, should I add them directly to ara.wordlist

Any Help, thank you.