Arabic-Handwritten-Text-Image-Recognition

Convert Arabic Handwritten Images to Text

Problem Statement:

Given a Arabic handwritten word in image form. Convert it into text form or recognise the word and get the word in text form

Solution Implemented:

Used 3 layer CNN to learn the features of Arabic Text. Fed that leaning into dynamic_rnn module with LSTM cell to predict the output.

Results:

The model is trainable and able to converge on small dataset.

File Description:

train.py:

to train the model

cnn_lstm_model.py :

Implemntation of CNN-LSTM model using tensorflow.

configuration.py:

Configuration file for all the hyperparameters and variable

helpers.py:

contains helpers function to create batch to feed into model, prepare arabic character set and dictionary

Run.ipynb:

contains the results of training along with predicted result

Result:

screenshot from 2017-12-26 14 13 42

Further Work:

Need good amount of GPU resource to train the model on full dataset to get the Results.