Digit Recognition Using CNN and MNIST Dataset

Abstract

Handwritten Digit Recognition (HDR) techniques are gaining traction in both academic and industrial domains. The complexity of recognizing handwritten digits arises from the diverse and intricate patterns involved. Identifying words in low-resource scripts presents significant challenges and is often time-intensive. Improving the performance of deep learning (DL) models, especially neural networks, can be achieved by expanding training datasets and incorporating sample randomization. Traditional HDR methods typically depend on manually extracted key point features. Variations in handwriting styles and input dimensions add to the difficulty of numerical classification and identification. This study addresses these challenges using a Convolutional Neural Network (CNN) model integrated into a robust HDR framework. The proposed HDR approach improves classification precision on the MNIST dataset. A Gaussian filter and Convolution algorithm are applied to enhance the digit images. The CNN-WOA model performance is benchmarked against state-of-the-art methodologies applying metrics like specificity, recall, AUC, F1-score, accuracy, and false positive rate (FPR).