Performance Improvement in Handwritten Devanagari Character Classification

Abstract

The optical character recognition models have the capability to recognize the characters in real-time. Extensive research work in the field of optical character recognition has led to the development of robust recognition mechanisms for various languages. The concepts of Artificial Intelligence and Deep learning have played a significant role in technological advancements in this field. But there are still some languages that don't have efficient Optical Character Recognition (OCR) systems but have vast ancient literature in the form of scriptures and manuscripts which are still relevant in the present. In recent years, the conventional Convolutional Neural Network (CNN) has performed distinctly in image processing and pattern recognition applications. But the pooling operation in CNN ignores the important spatial information, which proves to be an essential attribute in many cases. The proposed Capsule Network extracts spatial information and improves the capabilities of traditional CNN. It uses capsules to describe features in multiple dimensions and dynamic routing to increase the performance of the network.