James Harvey Undergraduate Dissertation 2017/18

Reassembling Fractured 3D Objects

Supervised by S.Maddock

Abstract

Reassembly of archaeological objects consisting of multiple, small and often unidentifiable fragments is a difficult process. Traditional manual approaches therefore had to use creative techniques in order to solve these essentially 3D jigsaw puzzles. But due to advancements in computer-based processes, new and existing techniques are becoming increasingly autonomous, allowing this matching process to become more time efficient and accurate.

In this project, a manual approach using the Leap Motion will be the replacement to a complex extracting algorithm which identifies features on fracture surfaces of archaeological fragments, relying instead on human knowledge to select regions on these fragments which they deem important to be considered in the matching process. This combined with an automatic matching algorithm, which takes the vertex information from the two selected regions to return an error to how well they have been matched, make up the developed system. A particular emphasis is placed on how fragments can be partially matched with one another.

The developed system has shown that it is capable with dealing with pairwise partial matching using controlled data where when such a fragment was presented with multiple alternatives, it was correctly matched with its corresponding partner, indicated by the lowest root mean square error. In the case of the real scanned data, when reduced to a polygon count that is manageable within this project, the system could also reassemble these fragments with their correct corresponding fragment with the lowest error, additionally with a visually correct representation of the original object. Further work should ensure that system can deal with fragments of a higher fidelity, allowing more precise matching and to make use of an alternative error metric to provide better context to the accuracy of this matching.