Automatic Reassembly of Fragments for Restoration of Heritage Site Structures

Abstract

A major bottleneck activity in the process of restoration of Heritage Structures is the reassembly of its fragments. Computer-aided reassembly could assist in finding the relation between them thereby reducing time, manpower and potential degradation to fragile fragments. Using geometric compatibility between the adjacent fragments as the central idea, a reassembly framework for a three-dimensional shell is proposed as a logical extension of the twodimensional framework. Edges are extracted as polygons and relevant features are computed at each of its vertices. Sequences of the match for two fragments in the feature space are found using a modified version of Smith-Waterman Algorithm. Each match is assessed using a connectivity score. The final choice of best match is left to the user by displaying the resultant assembled fragments of prospective candidates along with the score. After pairwise matching, the global reassembly is done through a clustering-based method. This framework can handle fragments even with curved edges which can be reasonably approximated by a set of edges. We verify the methodology using a simulated dataset for both 2D pieces and a shattered 3D surface object.

Deepzzle: Solving Visual Jigsaw Puzzles With Deep Learning and Shortest Path Optimization

“…Finding the correct reassemblies is a crucial step to understand our past better. Usually, this task relies on computer vision algorithms, such as contours or features detection [23]. The recent upsurge of deep-learning opens bright perspectives for finding better reassemblies more efficiently.…”