Sinhala Handwriting Character Recognition System Via a Deep Convolutional Neural Network

Abstract

Sinhalese is a language mostly used in the country of Sri Lanka and it consists of a spoken and written form. As many Sri Lankans write in Sinhala, digitization of handwritten text will add value to many industries and domains that use the Sinhala language. Many attempts at Sinhala handwriting recognition has yielded average results and therefore the project scope is aimed at finding a solution. Handwritten digit and character recognition for other languages such as English has seen a marked improvement in recent years since the modern advent of deep learning artificial neural networks such as Convolutional Neural Networks. The research aims to combine and apply the technological advancements in the deep learning field of Convolutional Neural Networks to the problem domain in order to explore the possibility of increasing the recognition accuracy for Sinhala handwriting recognition. The research was able to create a deep learning model and a Sinhala handwritten character dataset that performs well at an 83% accuracy. The applications and implications of the convolutional neural network used is discussed and evaluated in the project report.