Decoding Nagri: Handwritten Digit Recognition Using Deep Convolutional Neural Networks

Handwritten digit recognition plays a critical role in applications ranging from optical character recognition (OCR) to banking and postal code identification. While deep learning techniques, particularly Convolutional Neural Networks (CNNs), have achieved remarkable success in recognizing digits from languages like English and Bengali, lesser-known scripts remain underexplored. One such script is Sylheti Nagri, used by speakers in the Sylhet region of Bangladesh and the Barak Valley of Assam, India. Despite its cultural significance and similarities to Bengali, Sylheti Nagri has seen limited research in the field of digit recognition. This study introduces a pioneering approach to recognizing Sylheti Nagri handwritten digits using a deep CNN model, along with a Support Vector Machine (SVM) model for comparative evaluation. To support this effort, this research developed NagriDigitDB, a dataset containing 9,376 handwritten digit images, making it a unique resource for future research. The deep CNN model achieved an impressive accuracy of 94.35%, underscoring its effectiveness for digit recognition in this lesser-studied language whereas in SVM the accuracy was 84.5%. This work not only fills a gap in the research on Sylheti Nagri but also paves the way for broader applications in linguistic preservation and technological advancement.