Unified Multilingual Numeral Classification with Deep Neural Architectures

This study uses deep learning models to classify handwritten numerals across multiple languages. We curated a diverse dataset of Urdu numerals and extended it to classify them in Urdu, Persian, and English, leveraging their similarities and unique features. By employing a unified classifier, we effectively distinguished numerals across these languages. Our approach compared the custom numeral-CNN with established architectures like VGGNet, ResNet, GoogLeNet, and Xception. Numeral-CNN and Xception delivered superior performance, achieving accuracies of 98.91 % and 99.01 %, respectively, surpassing other models. We utilized data augmentation techniques to enhance dataset diversity and improve model robustness. These findings highlight the effectiveness of our approach and emphasize the role of deep learning in crosslinguistic numeral recognition.