Hindi Handwritten Character Recognition using Deep Convolution Neural Network

Convolution Neural Network (CNN) is turning out to be a very powerful tool for solving Machine Learning (ML) problems, especially in multiclass image classification. With the availability of a huge handwritten dataset, it is possible to achieve a never thought machine accuracy in image classification. In this paper, we have proposed a Deep Convolution Neural Network (DCNN) for Hindi handwritten character recognition. The idea is basically an expansion of LeNet-5 architecture. We have trained our model using 96000 character sets, obtaining a validation set accuracy of 95.72 per cent using Adam optimizer and 93.68 per cent using RMSprop optimizer.