A novel dataset of North-Eastern Indian coins for machine learning-based classification.

Old coins made by hand in ancient times and the Middle Ages are difficult to categorize because they are important for understanding culture and history, and they have complicated designs. The recent breakthroughs in machine learning have demonstrated strong potential for solving such classification tasks, but these approaches require well-structured datasets. This paper presents a new dataset of Indian coin images from the North-Eastern region, including Assam, Tripura, Koch Bihar, and Jaintiapur, which were historically part of the Bengali kingdoms. This dataset contains 51 classes and 2147 images. The images were collected from verified private collections and five well-established auction houses with permission. Each image showcases the ideal examples of these coins. They are usually difficult to recognize because they are very scarce and feature intricate designs and inscriptions. Each coin was carefully checked and confirmed using reliable numismatic sources, including the S. K. Bose numismatic series, verified auction house records, and authenticated private collections. The dataset includes images of extremely rare coins from the early 9th century AD to 1947, just before the partition of the Indian subcontinent. The historical, cultural, and antiquarian significance of these coins makes the dataset a valuable resource for study. It is anticipated that the methodologies will advance coin classification and make significant contributions to the archaeological research of ancient North-Eastern Indian history and culture.