Computational Flint Refitting

Lithic refitting is a fundamental method in prehistoric archaeology for reconstructing past technologies, knapping sequences, and site formation processes. Performed manually, however, this task is highly labor-intensive and requires significant expertise and patience. This project presents a novel computational pipeline for the fully automatic refitting of 3D lithic artifacts, treating the task as a complex 3D puzzle-solving problem.

The proposed method begins with 3D-scanned lithic fragments, which contain both geometric and pictorial information, and are first segmented into their faces based on surface curvedness. These faces are then processed for three types of features: boundary curves approximated by cubic Bézier curves, internal geometric anomalies, and pictorial data derived from surface texture and contrast. Features from different faces are then matched computationally to find candidate pairs of fragments that should be refitted. Finally, global assembly is obtained by implementing these matches as attraction and repulsion forces via springs in a physical spring-mass system. Simulating the dynamics of this system, as it seeks to minimize its potential energy, effectively refits the set of virtual flint fragments into their optimal spatial configuration.

We evaluated the approach using a dataset of several refits, comprising experimentally knapped flint and archaeological refits from the Late Middle Paleolithic site of Far’ah II (49ka). With qualitative results comparable to the ground truth, we also assessed performance quantitatively using various metrics. Further analysis also indicates that, unlike facial geometric cues, boundary and pictorial features are the most significant drivers for successful refitting. With fully automatic refitting in sight, this work can thus assist archaeologists in deciphering lithic assemblages and potentially revolutionize their work.

Code / Data

Code and scanned fragments are currently available here.

Acknowledgments

This work was funded in part by the European Union’s Horizon 2020 research and innovation programme under the grant agreement No. 964854 (the RePAIR project)..