Introduction
Reassembling shattered 3D objects is a significant challenge in archaeology and forensics, particularly when fragments are small and thin. This paper introduces a hybrid algorithm that combines template guidance—matching pieces to a reference model—with fracture-region matching—fitting pieces together based on edge geometry. This dual approach achieves highly accurate reconstructions of archaeological artifacts and forensic skeletal remains where traditional methods often fail.
3D fragment reassembly is critical for "thin-shell" objects like pottery or skulls. Existing methods typically rely on either fracture-region matching, which fails when fragments are too small to provide a reliable match, or template guidance, which fails when the specific object differs significantly from the template. The proposed system addresses these limitations by integrating both global and local constraints.
Methodology
The authors propose a three-step pipeline that integrates global and local information:
- Initial Template Matching: Using Intrinsic Shape Signatures (ISS) and SHOT descriptors, the system performs a partial match between each fragment and a global template to provide a rough spatial "neighborhood" for each piece.
- Pairwise Fracture Matching: The algorithm extracts "feature curves" from broken edges and uses a modified 4-Point Congruent Sets (4PCS) algorithm to find potential alignments between adjacent fragments.
- Multi-piece Matching Optimization: Reassembly is represented as a graph where nodes are fragments and edges are potential alignments. Beam Search explores multiple assembly paths simultaneously to avoid local minima. Finally, a global refinement step minimizes gaps and overlaps across all pieces.
A key innovation is the weighting factor ($\alpha$), which allows the user to tune the system's reliance on the template versus the inter-fragment matches based on the quality of the reference model.
Key Insights and Results
A major breakthrough of this method is its ability to handle tiny shards without requiring explicit manual segmentation of "intact" versus "fracture" regions. By incorporating both geometries, the algorithm reassembles pieces that previous state-of-the-art methods could not handle. Even a generic template provides enough global constraint to prevent small pieces from drifting during the assembly of the whole.
Experimental results on ceramic models showed matching errors consistently under 1mm. The system successfully reassembled a 19-piece skull using a generic template and proved highly effective in a real forensic case involving a gunshot victim. In that instance, the algorithm aligned fragile fragments that forensic anthropologists found difficult to handle manually. Computation time ranged from 4 to 14 minutes, making it a viable tool for practical laboratory applications.
Conclusion
While the current system relies on a single template, it represents a significant advancement in the automated reassembly of thin-shell fragments. Future work will explore multi-template guidance to better accommodate anatomical or stylistic variety by leveraging databases of multiple reference models.
Summary updated