Word Transcription of MODI Script to Devanagari Using Deep Neural Network

Many historical documents and letters are written in MODI script. Study of “Shivakalin” and “Peshvekalin” era documents is almost impossible without the knowledge of MODI script. This work aims to bridge the gap between Devanagari and MODI Script by developing a system to map the recognized MODI characters to its Devanagari equivalent. Our dataset would comprise 57 different classes of MODI Script characters. We has solved the problem of loss of information in CNN as the model goes deeper using Dense Net-121. The various approaches for feature extraction usually used include moment invariant, affine moment invariant, chain code histogram, intersection junction and for character classification. Deep Neural Networks on the other hand do not require any feature to be explicitly defined, instead they work on the raw pixel data to generate the best features and use them to classify the inputs into different classes. Hence, we propose a deep learning architecture for character recognition. Dense Net-121 uses little pre-processing compared to other image classification algorithms. This means that the network learns the filters which in traditional algorithms were hand-engineered.  The system aims to provide a good recognition rate by implementing DCNN.