Categorization of Aztec potsherds using 3D local descriptors

Abstract

We introduce the Tepalcatl project, an ongoing bi-disciplinary effort conducted by archaeologists and computer vision researchers, which focuses on developing statistical methods for the automatic categorization of potsherds; more precisely, potsherds from ancient Mexico including the Teotihuacan and Aztec civilizations. We captured 3D models of several potsherds, and annotated them using seven taxonomic criteria appropriate for categorization. Our first task consisted in exploiting the descriptive power of two state-of-the-art 3D descriptors. Then, we evaluated their retrieval and classification performance. Finally, we investigated the effects of dimensionality reduction for categorization of our data. Our results are promising and demonstrate the potential of computer vision techniques for archaeological classification of potsherds.