Unsupervised statistical learning for die analysis in ancient numismatics
Die analysis is an essential numismatic method and an important tool of ancient economic history. Yet, manual die studies are too labor-intensive to comprehensively study large coinages. We propose an unsupervised computational framework that reduces the required time investment by orders of magnitude. To cluster poorly preserved and imbalanced coin data, involving an unknown and large number of highly similar dies, our method derives dissimilarities between coin faces using specially devised Gaussian process-based keypoint features within a Bayesian distance clustering framework. We demonstrate the efficacy of our method on a semantically demanding collection of 1135 Roman silver coins struck between 64–66 C.E., many of which exhibit substantial wear and poor preservation. Our framework remains robust under such conditions, indicating its suitability for real-world historical datasets where preservation quality and complexity limit the applicability of neural-network-based approaches.
Authors
- Abhinav Natarajan (ORCID: https://orcid.org/0000-0002-2563-6721)
- Emanuel Mayer (ORCID: https://orcid.org/0000-0002-6471-952X)
Institutions
- National University of Singapore (SG)
- Yale-NUS College (SG)
- University of Oxford (GB)
- University of Newcastle Australia (AU)
Publication Details
- Journal
- npj Heritage Science
- Published
- 2026-11-17
- DOI
- https://doi.org/10.1038/s40494-026-03087-1
- Citations
- 15
- Primary Topic
- Image Processing and 3D Reconstruction
- Type
- article
- Field-Weighted Citation Impact
- 6.81