LSTM based Odia Handwritten Numeral Recognition

Recent works on character recognition involves in different applications like security, biometrics etc. In most of the works Convolutional neural networks (CNN) are used for image recognition with large dataset where convolutional filters with pooling layers are used for feature extraction. In this work we have utilized the concept of Long Short Term Memory (LSTM) for recognizing the Odia handwritten numerals. The loss in recognizing the digits is evaluated using categorical_crossentropy loss function and for optimization we have applied Adam optimization to minimize the error. The network is trained with supervised learning method. The accuracy found in this work is 97.93% which is a remarkable achievement in Odia numeral recognition.