Offline handwritten Malayalam character recognition using stacked LSTM

In this paper we propose a model for isolated Malayalam handwritten character recognition using stacked LSTM. Ninty symbols from the Malayalam character set is considered for the recognition and total samples used are 18000. Network consists of two LSTM layers and final output layer for prediction. Accuracy achieved is more than 90 %. Top-2 results shows that it can be improved by addition of more samples in the dataset.