Handwritten Gujarati Character Recognition

Abstract

Character recognition is a task of classifying character images into one of the many predefined classes. The focus of this thesis is on handwritten Gujarati character recognition. Specifically, thesis focuses on applying deep learning techniques for the task of handwrit- ten Gujarati character recognition. Experiments are carried out on three datasets, out of which two are Gujarati numeral datasets while one is a Gujarati character dataset. LeNet - a well known deep neural network is used for the task. A significant contribution of the thesis is ILeNet which is inspired from LeNet and ne-tuned for the requirement of handwritten Gujarati character recognition. Experimental results demonstrate that classification accuracy of LeNet on all three datasets is significant. ILeNet improves the accuracy further and establishes the importance of ILeNet.