Urdu Alphabet (حروف تہجی) Handwritten Letters Recognizer

Recognition of Urdu Handwritten Letters Using Convolutional Neural Network

Urdu is one of the cursive languages that is widely spoken and written in the regions of South-East Asia including Pakistan, India, Bangladesh, Afghanistan, etc.

There are 39 alphabets in Urdu language, This Experiment is done using 175500 images, each letter contains 4500 images, The dimension of the image is 1x50x50x1.

Demo

Install

  git clone https://github.com/ranahaani/htlr.git
  cd htlr
  pip install -r requirements.txt

Download dataset

CSV dataset is available on Google Drive Link.

You can download images dataset on Kaggle

kaggle datasets download -d ranahaani/htlr-dataset

Requirements

python
numpy
matplotlib
sklearn
tensorflow
keras

Usage

# To Train
python htlr.py --save-model 1 --weights output/htlr.hdf5
python htlr.py --load-model 1 --weights output/htlr.hdf5

License

MIT

humsha/htlr