Using computer vision and digital image processing to define ceramic fabric groups

Summary

This is an abstract from the "Ceramic Petrographers in the Americas, Production Practices and Social Networks from Multilevel Angles" session, at the 90th annual meeting of the Society for American Archaeology.

Image thresholding algorithms can help extract useful data, such as particle counts, from ceramic petrographic slide images. These metrics can assist archaeologists in identifying ceramic fabric groups, which in turn helps answer broad questions about pottery provenance, exchange networks, potter decision-making, and communities of practice. Computer vision models like Segment Anything (SAM) can segment ceramic petrographic slides with results comparable to widely used image thresholding algorithms. In this poster, we implement and evaluate a Python workflow that uses computer vision and digital image processing to extract ceramic fabric data from petrographic slide images. SAM first segments each slide image into discrete objects, and the resulting mask is vectorized. Segmented objects are then divided into particles and voids based on color thresholding, and individual particles are analyzed for their size and sphericity. Finally, we perform a cluster analysis on a set of derived metrics (% particle area; % void area; particle size, sortedness, and sphericity; and void size and sphericity) to determine fabric groups. We compare these fabric group results to groupings assigned by a ceramic petrographer to determine if these metrics alone can be used to define valid groups.

This Resource is Part of the Following Collections

• Society for American Archaeology 90th Annual Meeting, Denver, CO (2025)   •
• Ceramic Petrographers in the Americas, Production Practices and Social Networks from Multilevel Angles

Keywords

Record Identifiers

Abstract Id(s): 52592