HINDI HANDWRITTEN CHARACTER RECOGNITION WITH DEEP LEARNING

Abstract

Handwritten character recognition plays a crucial role in various applications, including document digitization, language translation, and text analysis. This thesis presents a comprehensive investigation into the application of deep learning techniques for the recognition of handwritten characters, with a specific focus on Hindi characters. Leveraging insights from existing literature and building upon recent advancements in deep learning methodologies, the research aims to develop and evaluate novel approaches for improving the accuracy and efficiency of character recognition systems. The study begins with a thorough review of traditional methods and contemporary deep learning architectures used in handwritten character recognition. Emphasis is placed on understanding the evolution of techniques and identifying key challenges in the field. Subsequently, a detailed methodology is proposed, encompassing data collection, preprocessing, feature extraction, and the implementation of deep neural networks optimized with advanced algorithms such as RMSprop and Adam. Experimental evaluations are conducted on a substantial collection of handwritten Hindi character datasets, employing rigorous training and testing procedures. The results demonstrate the efficacy of the proposed methodology, with significant improvements in recognition accuracy compared to baseline models. Comparative analyses are presented, highlighting the strengths and limitations of different deep learning approaches.