Computational Visual Ceramicology: Matching Image Outlines to Catalog Sketches

Abstract

Field archeologists are called upon to identify potsherds, for which they rely on their professional experience and on reference works. We have developed a recognition method starting from images captured on site, which relies on the shape of the sherd's fracture outline. The method sets up a new target for deep-learning, integrating information from points along inner and outer surfaces to learn about shapes. Training the classifiers required tackling multiple challenges that arose on account of our working with real-world archeological data: paucity of labeled data; extreme imbalance between instances of different categories; and the need to avoid neglecting rare classes and to take note of minute distinguishing features of some classes. The scarcity of training data was overcome by using synthetically-produced virtual potsherds and by employing multiple data-augmentation techniques. A novel form of training loss allowed us to overcome classification problems caused by under-populated classes and inhomogeneous distribution of discriminative features.

Archaeological Practice and Digital Automation

“…For example, several systems have focused on the automation of pottery classification (e.g., Makridis and Daras 2012;Teddy et al 2015;Roman-Rangel et al 2016;Hein et al 2018;Rasheed and Nordin 2020;Wright and Gattiglia 2018). Others have taken related approaches to pottery, such as automating the creation of 3D models from 2D catalog drawings (Banterle et al 2017), an intelligent search engine for pottery retrieval (Benhabiles and Tabia 2016), the automated classification of mineral inclusions in pottery (Aprile et al 2014), automatic Munsell color characterization of sherds (Milotta et al 2018), and shape-and decoration-based identification of pottery (Itkin et al 2019), for example. Similar machine learning approaches have been applied to automated feature extraction from LiDAR and satellite imagery (see overviews in Bennett et al 2014;Optiz and Herrmann 2018;Davis 2019;.…”