Skip to Main content Skip to Navigation
New interface

Comparaison de motifs sur des nuages de points 3D et application sur des monnaies et objets celtiques

Abstract : Clustering coins according to their die is a problem that has many applications in numismatics. This clustering is crucial for understanding the economic history of tribes (especially for tribes for whom few written records exist, such as the Celts). It is a difficult task, requiring a lot of times and expertises. However, there is very little work that has been done on coin die identification.This thesis project aims at proposing an automatic tool to know if two patterns have been impressed with the same tool, especially to know if two coins have been struck with the same die. Based on deep learning-based registration algorithms, the proposed method has allowed us to classify a hoard of a thousand Riedone coins dating from the 2nd century BC. This treasure allowed us to build an annotated dataset of 3D acquisitions called Riedones3D. Riedones3D is useful for Celtic coin specialists, but also for the computer vision community to develop new coin die recognition algorithms. Rigorous evaluations on Riedones3D and on other Celtic works show the interest of the proposed method. Indeed, it can be adapted to unknown patterns. Finally, we propose a new registration algorithm that can be adapted to any type of sensor. Thanks to this algorithm, it is potentially possible for a specialist to use faster or less expensive sensors to acquire coins or engraved patterns.
Document type :
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Tuesday, September 27, 2022 - 3:32:12 PM
Last modification on : Saturday, October 22, 2022 - 5:11:27 AM


Version validated by the jury (STAR)


  • HAL Id : tel-03789632, version 1



Sofiane Horache. Comparaison de motifs sur des nuages de points 3D et application sur des monnaies et objets celtiques. Robotique [cs.RO]. Université Paris sciences et lettres, 2022. Français. ⟨NNT : 2022UPSLM019⟩. ⟨tel-03789632⟩



Record views


Files downloads