Optical Character Recognition Using DeepLearning

Text is everywhere! It is present in PDFs, docs as well as images. There are lots of applications where text data is useful for doing analytics. Such applications include receipts recognition, number plate detection, extracting the latex formulas from the images etc. General Computer Vision can be used for such task but it lacks in accuracy. In order to solve the low accuracy and variance problem, we use the state of the art deep neural networks.

This repository includes:

1. A TensorFlow implementation of the CNN+LSTM+CTC model for OCR.
2. supporting scripts to apply the RCNN appraoch for OCR.

Architecture

Instructions on How to run

Get the repository

git clone https://github.com/harshul1610/OCR.git

Get the NIST19 dataset

mkdir data
wget https://s3.amazonaws.com/nist-srd/SD19/by_class.zip
unzip by_class.zip
mv by_class NIST19

Get the Captcha data

cd OCR
python2 generate_captcha.py

Run the final notebook for training and testing

CNN_LSTM_CTC_OCR-captcha.ipynb

LICENSE

MIT