Ornamental metal artefacts are central to many archaeological classifications and typologies. Inconsistent classification systems and large numbers of finds, meanwhile, challenges identification. As 3D digitization of cultural heritage data becomes increasingly widespread, automated feature recognition offers potential to augment research. However, existing algorithms used for computer vision are ineffective for similarity search in 3D data. Here we analyse a sample set of 12 objects comprising complete artefacts and fragments pertaining to clearly distinct typological groups. In this paper, wevoxelize the 3D meshes and perform a pairwise point cloud registration to cluster the artefacts into groups using a distance measure. This method is found to correctly identify positive matches but struggles to compare fragments or pieces of very different size. We conclude that the method is viable, but distance measure must be further elaborated to take into consideration the number of points in point clouds as source and target. • Similarity search correctly identifies typology in 3D scanned artefacts. • Pairwise point cloud registration to cluster the artefacts into groups identifies positive matches, and highlights limitations in comparing fragments or pieces of very different size. • Pairwise point cloud registration to cluster the artefacts into groups identifies positive matches, and highlights limitations in comparing fragments or pieces of very different size. • The potential and challenges for the use of AI computer vision in 3D data is outlined.
Boesen et al. (Tue,) studied this question.