Ilan SHIMSHONI | Niv DERECH | Magali SEGAL-STOLARSKY | Ayellet TAL
(University of Haifa & Technion, Haifa, Israel)

Keywords: GRAVITATE, Reassembly, Complimentary matching, 2D, 3D

Abstract:
One of the three main capabilities of the GRAVITATE project is reassembly. A reassembly algorithm is given as input a set of fragments of archaeological artefacts and its goal is to reassemble the object from these fragments. The task is quite challenging since the fragments are abraded, their colours have faded and some of the fragments might be missing or not be related to the artefact to be restored.
In this project we worked on two problems. The first one reconstructs a planar object such as a fresco from images of its parts taking into account the colours of the fragments and their shape. The algorithm first extrapolates the images and matches them to other images. The matches are ranked by the quality of their matches and their uniqueness. The best matching pair of parts is chosen for reassembly. Then more parts are added improving their local transformation. The final output is the complete object.
The second algorithm deals with scans of 3D fragments. Colour information is not assumed to exist. The first step of the algorithm is applying faceting to the meshes, yielding the facets of the fragment and their boundaries. Here instead of extrapolating the fragments we analyse the differential properties of the points on the facets trying to eliminate possible matches between pairs of fragments. On the remaining pairs an approximate relative transformation is computed and only on pairs of boundary curves which satisfy this condition the curves are matched. The matches are then ranked by measuring the distance between closest points on the two fragments. The best ranked matches are returned by the algorithm.
In future we plan to complete the total reassembly of 3D fragments and combine the ideas of both algorithms for dealing with coloured 3D objects.

Relevance for the conference: The algorithms reassemble archeological artefacts
Relevance for the session: This work is part of the GRAVITATE EU2020 project
Innovation: Dealing with genrally shaped, abraded and faded fragments is difficult compared to regular rectangular pieces
References:
• Paikin, Genady, and Ayellet Tal. “Solving multiple square jigsaw puzzles with missing pieces.” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015.
• Sizikova, Elena, and Thomas Funkhouser. “Wall painting reconstruction using a genetic algorithm.” Journal on Computing and Cultural Heritage (JOCCH) 11.1 (2017): 3.