Automated Reconstruction of Fragmented Objects - Application to the reassembly of celebrated wall-paintings


Many thousands of important archaeological objects are unearthed fragmented throughout the world in various archaeological sites. Hence, the development of an integrated information system that may assist the reassembly of fragmented archaeological finds is of essential importance for the preservation of Cultural Heritage.

The research group “Applications of Computer Engineering in Pattern Recognition, Arcaeometry, Bioengineering, Music and Arts” (Prof. C. Papaodysseus, Ass.Prof. M. Panagopoulos, Dr. D. Arabadjis, Dr. M.Exarhos, Dr. P. Roussopoulos, Dr. S. Zannos, Dr. F. Giannopoulos, Dr. Th. Panagopoulos, Mr. Th. Mamatsis) has developed a novel information system that gives a practically unique solution to the problem of fragmented wall-paintings reconstruction. The developed system takes into account various aspects regarding the unearthed fragments, such as:

  • Matching of the contours of the painted/plain sides of adjacent fragments
  • Three dimensional (3D) matching of the lateral surfaces of adjacent fragments
  • Continuation of the thematic content between the painted sides of matching fragments
  • Crack or texture continuation between adjacent fragments

We will give a brief description of the main aspects of the developed system.

1. Two-Dimensional Automated Reconstruction

The first subsystem takes into account only the two-dimensional geometric matching of the plain sides contours. The approach consists of a number of criteria, which exploit contour similarities between 2D objects. It is shown below an application of the system to Akrotiri, Thera wall-paintings.

The method has been applied to the prehistoric (c. 1650 B.C.) wall-paintings excavated in thousands of fragments at Akrotiri, Thera, Greece. The authors have photographed the painted side of the wall-painting fragments and the developed system offered numerous matches between them for the first time. The matches have been verified by the dedicated scholars. Three reconstructed “islands” of fragments are shown below:

Image 1: Example of wall-painting parts automatically reconstructed by the 2D system for the first time.

Image 2: Another example of wall-painting parts automatically reconstructed by the 2D system for the first time

Image 3: A third example of numerous wall painting fragments, automatically reconstructed by the 2D system for the first time.

2. Three Dimensional Automated Reconstruction

The necessity of exploiting the 3D representation of the fragments lies on the fact that the 2D subsystem spots the existing matches between fragments, but often gives erroneous matches. We have attributed this to the limited information obtained in 2D, in connection with the large number of fragments and the possible small matching length. Consequently, we have developed a subsystem that finds the actual matches between Fragments in three dimensions, using principles of 3D-Geometry, Calculus of Variations and 3D pattern recognition. We have applied this subsystem to the virtual reassembly of islands of fragments excavated at Tyrins, in cooperation with Ms. Lena Papazoglou, ex director of the Prehistoric Department of the National Archaeological Museum of Greece.

To archieve this, we have obtained three-dimensional representations of the unearthed fragments via a 3D-scanner. A number of such 3D representations are depicted below.

Image 4: Example of 3D representation of fragments obtained by the authors via a 3D-scanner.

Image 5: Example of 3D representation of fragments obtained by the authors via a 3D-scanner.

The developed Mathematic algorithms offered optimal matching between initially adjacent fragments. We have applied the aforementioned methodology to 41 fragments of prehistoric wall-paintings (14th and 13th century B.C.) from Tyrins and Mycenae kept in National Archaeological Museum of Greece. The developed system offered 9 islands of matching fragments consisting of 2-5 pieces. These results have been verified by Museum’s scholars, who didn’t find any additional matches.

A number of associated results are shown below:

Image 6: Two fragments optimally matching in 3 dimensions; the matching position has been automatically suggested by the system.

Image 7: Two fragments optimally matching in 3 dimensions; the matching position has been automatically suggested by the system.

Image 8: The image of four reassembled fragments as suggested by the system.

Image 9: Verification of the validity of the matching offered by the system.

Image 10: The image of three reassembled fragments as suggested by the system.

Image 11: Verification of the validity of the matching offered by the system.

Exploitation of the thematic continuation

We have developed another subsystem that achieves optimal matching between fragments that are not necessarily in contact, by exploitation of the thematic content on the fragments’ painted sides. We exploit the fact that the same geometric scheme may appear on different fragments belonging to the same theme. A number of associated results are shown below.

Image 12: A first characteristic example. We note that it is intrinsically very difficult for the dedicated personnel to spot this matching.

Image 13: Matching of fragments in loose contact following an Archimedes spiral pattern.

Related Publications here.