CNN based approach to Telugu OCR

Our paper is accepted at IEEE International Conference On Image Processing (ICIP) 2018, Greece.

Here the code for CNN based approach to Telugu OCR paper is avalilable.

Tested on 64 bit Linux.

Note: This package is old and not being maintained

Usage

git clone https://github.com/GayamTrishal/OCR_Telugu_code.git
cd OCR_Telugu_code/models/vattu_gunintam/ours/
unzip model_v_g_weights.hdf5.zip
cd ../../..
cd OCR_Telugu_code/code
python OCR.py

Training

Main Character

• Set file name training_testing/main_character/model_code.py, h5_file_location = 'YOURPATH/final_dataset.hdf5'

cd OCR_Telugu_code/training_testing/main_character
python model_code.py

Vattu Gunintam

• Set file name in training_testing/vattu_gunintam/model_code.py, h5_file_location = 'YOURPATH/final_dataset.hdf5'

cd OCR_Telugu_code/training_testing/vattu_gunintam
python model_code.py

Trained Models

Information on trained models can be found here.

Requirements

• Keras
• OpenCV 3.0.0 version (currently only supports this.)
• Skimage
• Numpy

Dataset

Dataset used can be downloaded here.

Testing on sample image

Replace img.jpg inside OCR_Telugu_code/code folder with your image and run the code. The result will be stored in OCR_Telugu_code/code/output folder.