Devanagari and Gurmukhi Handwritten Character Generation using Generative Adversarial Networks

Abstract

Today, computers have influenced the life of human beings to a great extent. The theory that computers can learn without being programmed to perform a specific task, have attracted the researchers to see if computers could learn from data. As deep learning became popular, the need for huge amounts of data has risen. The major problem faced in the area of deep learning is the data availability. In this dissertation, a generative technique is used to generate the handwritten Gurmukhi and Devanagari characters. This dissertation describes the implementation of handwritten Gurmukhi and Devanagari characters using deep learning technique named as Generative Adversarial Networks. Gurmukhi and Devanagari script is chosen for this research work as it is used directly or indirectly by more than 500 million people in the Indian subcontinent and less work is done on Devanagari and Gurmukhi script as compared to work done on other scripts such as English and Chinese. GANs are chosen for the implementation as it is proven to be the very powerful generative method. Here, we use 3-layer CNN having stride value of 2 for the feature extraction of handwritten character. The characters generated look like the character in the original dataset. Also by increasing the number of epochs the images are more recognizable and clear as well as the loss starts decreasing.