Research On Automatic Reassembly Of2D Fragments

When confronted with a small amount of fragments, we can reconstruct the originalobjects by simple artificial contrast. However, due to the large number of mixedarchaeological pieces, reconstruction by artificial contrast is complicated andtime-consuming, and may cause damage to the fragments. Similar reconstruction problemsfrom archaeological pieces also exist in the area of judicial identification. Therefore,computer-aided automatic or semi-automatic reassembly becomes necessary. Thinnerfragments of murals, tiles, cloth, and canvas, etc., which are negligible can be called2Dfragments; While fragments of potteries, ceramics and Terra-Cotta Warriors withthree-dimensional features are called3D fragments.Much of work has been done for the reassembly of2D fragments at home and abroad.Image reassembly problems are often achieved by particularities such as smooth puzzleedges, ideal corner points, regular arrangement, etc.Content-based reassembly method are often only used for strip or rectangularfragments produced by shredder through crosscutting, slitting or other means, and theregular hexagon file fragments; Irregular fragment reassembly, according to itscharacteristics, is divided into content-based (color, texture) reassembly and contour-basedreassembly, where in more research are contour-based. The main content of this thesis isbased on contour-based two-dimensional irregular fragment reassembly.Contour-based two-dimensional irregular fragment reassembly comprises thefollowing steps:1) pre-treatment, i.e., digital image fragments acquisition, contourextraction, representation and description;2) local match, to find the space-adjacentrelationship between any two fragments;3) global matching and reassembly, to find theglobal reassembly path and achieve reassembly recovery. Most studies are focused on localmatching algorithm and studies on global matching are relatively less.This thesis proposed a new contour-based two-dimensional fragment reassemblyprocess for two-dimensional irregular fragments. Taking into account the large amount of data when using the uniform sampling points to represent the contour, this thesis uses asimplified representation of the contour curve, i.e., uses the polygonal approximation toapproximate the contour curve to get a number of polygon vertices (feature points), andthen extract the polygon feature (rotation angle sequence and side length sequence).Since the rotation angles at the matching vertex of two matched polygon are equal inmagnitude and opposite in sign, while the lengths of two matched edges are approximatelyequal, this thesis uses an improved local matching method suitable for polygonal featuresto find the space adjacent fragment pairs and measure their degree of matching. Inaccordance with the non-overlapping feature of fragments in space, some pseudo-matchedpairs can be excluded. In global matching stage, this thesis proposed a newmatching-angle-based global path searching method, which can achieve multi-fragmentsreassembly by generating global path and reassembly the fragments thereby.The main innovation of this thesis is as follows:1. Analyzed the characteristics of polygons matching, improved a string-matchingmethod and apply it to the contour approximate polygon matching, and proposed a newmatching metric to achieve local matching and matching metric of multi-fragments.2. In the global matching stage, after analyzing the relationship between matchingangle of fragments, we propose path-generating method based on matching angle. First wedetermine several paths according to the conditions required by matching, and then thosepaths are integrated to finally stitch the fragments along the path. Compared with theexisting best-first method, the method proposed in this thesis can fully utilize the partialmatching result, provide a small amount of computing, and improve the efficiency of pathsearching.