Text Documents Optical Character Recognition

1. Required Packages and Credits

LSTM part are developed from https://github.com/weinman/cnn_lstm_ctc_ocr
You need following libraries to run the code:

• Tkinter
• Tensorflow
• Opencv
• Numpy

2. word_detect:

Word Detection from Sentence + Word Recognition with CNN Network

cnn_detect:

• python cnn_detect.py

lstm_detect:

• Download the lstm model from google drive: "https://drive.google.com/file/d/1fvNFbq1PWx_7PE0B9MnBmWHPVAUy8zKI/view?usp=sharing"
• place the downloaded model folder at word_detect/lstm/
• python lstm_detect.py

3. model_training:

Training Code of Neural Network for Recognition purpose

cnn_train: Train CNN network for recognition

• Download Emnist Dataset from http://www.itl.nist.gov/iaui/vip/cs_links/EMNIST/gzip.zip
• Move Downloaded dataset in "model_training/cnn_train/dataset/emnist"
• Use "gunzip" command to unzip all .gz files
• cd "model_training/cnn_train", then "python main.py"
• Use user interface for further operations

lstm_train: Train LSTM network for recognition

• Follow the README.md document in lstm_train folder