Surface Registration Techniques Applied to Archaeological Fragment Reconstruction

Abstract

Reconstruction of broken archaeological artifacts from fragments is a very time-consuming task that requires a big effort if performed manually. In fact, due to budgetary limitations, this is not even attempted in countless sites around the world, leaving vast quantities of material unstudied and stored indefinitely.This Thesis dissertation faces the application of surface registration techniques to the automatic re-assembly of broken archaeological artifacts from fragments. To efficiently do so, the reconstruction problem has been divided into two groups: 3 degrees of freedom and 6 degrees of freedom problems. This distinction is motivated for two major reasons: archaeological interest of the application and computational complexity of the solution.First kind of problems (3 degrees of freedom) deal with 2D objects or with flat 3D objects, like ripped-up documents or frescoes, respectively. In both cases, the mural paintings and engravings on the fragments' surface are of huge importance in the field of Cultural Heritage Recovery. In this sense, archaeologically speaking, the value of the reconstruction is not the final model itself, but the information stored in the upper surface. In terms of computation complexity, the reduced solution space allows using exhaustive techniques to ensure the quality of the results, while keeping execution times low. A fast hierarchical technique is introduced to face this kind of problems. Starting from an exhaustive search strategy, the technique progressively incorporates new features that lead to a hierarchical search strategy. Convergence and correction of the resulting technique are ensured using an optimistic cost function. Internal search calculations are optimized so the only operations performed are additions, subtractions and comparisons over aligned data. All heavy geometric operations are carried out by the GPU on a pre-processing stage that only happens once per fragment.Second kind of problems (6 degrees of freedom) deal with more general situations, where no special constraints are considered. Typical examples are broken sculptures, friezes, columns... In this case, computational complexity increases considerably with the extra 3 degrees of freedom, making exhaustive approaches prohibitive. To face this problems, an efficient sparse technique is introduced that uses a pre-processing stage to reduce the size of the problem: singular key-points in the original point cloud are selected based on a multi-scale feature extraction process driven by the saliency of each point. By computing a modified version of the PFH descriptor, the local neighborhood of each keypoint is described in a compact histogram. Using exclusively the selected key-points and their associated descriptors, a very fast one-to-one search algorithm is executed for each possible pair of fragments. This process uses a three-level hierarchical search strategy driven by the local similarity between key-points, and applying a set of geometric consistence tests for intermediate results. Finally, a graph-bas...