Sinhala handwritten character recognition using convolution neural networks

Abstract

Automating handwritten character recognition is still new, as Sri Lanka is the only country that uses Sinhala as the national language from all over the world. The alphabet of the Sinhala language includes 60 characters and they are somewhat complex than the other languages. There are nearly 25-30 researches have been done from 1990 towards Sinhala handwritten character recognition. But there is no accurate handwritten character recognizer for the Sinhala language. Therefore, a model using used the Convolutional Neural Networks to train and classify the Sinhala handwritten characters has been proposed. The training accuracy of the CNN method is 95 % and the testing accuracy is 85.71%. This is the highest accuracy obtained for 55 characters from 1990 when comparing with primitive methods.