A Comparative Analysis of CNN and Tesseract OCR on Handwritten Recognition System

Abstract

This study introduces a handwriting recognition system that utilises Convolutional Neural Networks (CNNs) to accurately identify handwritten characters. The methodology involves comprehensive data preprocessing, including normalisation, binarisation, and noise reduction, to enhance the quality of input images. A CNN model is trained on a diverse dataset to extract complex handwriting patterns. The system is deployed as a Flask-based web application for real-time predictions, ensuring accessibility and scalability. Results demonstrate that the CNN model achieves an accuracy of 95.8%, precision of 94.5%, recall of 93.7%, and F1-score of 94.1%. Compared to traditional Optical Character Recognition (OCR) systems, the CNN-based approach excels in handling varied handwriting styles and noisy inputs. Limitations in recognising non-Latin scripts indicate future research directions. This work highlights the effectiveness of CNNS for applications in digitising historical documents, banking, and education.