Deciphering Handwritten Text: A Convolutional Neural Network Framework for Handwritten Character Recognition

Abstract

HCR (Handwritten Character Recognition) is considered one of the most challenging research areas, given the vast array of potential applications. Character recognition has been the focus of research since the beginning of Artificial Intelligence. Numerous studies, including HCR, have been conducted in this sector. A typical procedure requires two steps: feature extraction and Classification. Many forms of neural networks have been used in this cause over the years, with notable results. CNN has altered the scenario in recent years. It has had remarkable success in this industry due to its cutting-edge extraction of features and Classification. To produce recognized characters, CNN uses images for input and sends them through a sequence of layers, including a convolutional layer, a nonlinear function, a pooling layer, and interconnected layers. We utilized a dataset containing 372,450 handwritten character images covering the entire alphabet in English. We created a model using the CNN model and achieved 99% test accuracy. CNN is an efficient and powerful approach for HCR. Our model's high accuracy suggests that it has the potential to be applied in various practical scenarios such as postal address reading, digital libraries, and traffic sign detection.