Torn Document Analysis as a Prerequisite for Reconstruction

An automated assembling of torn documents (2D) will support philologists, archaeologists and forensic experts. Especially if the amount of fragments is large (up to 1000), a human puzzle solver will not be feasible due to cost and time. Ancient manuscripts may be broken due to bad storage conditions, or documents are manually torn to make the information unreadable. In Germany a project to reconstruct the torn "Stasi-files" is running for historical investigations. Also disasters like the collapse of the historical archive of the city of cologne (Germany), where a large part of the archived manuscripts have been destroyed, need algorithms to reconstruct torn manuscripts and books. The automated solving can be divided into shape based matching techniques (apictorial) or techniques that analyze the visual content of the fragments (pictorial) too. Artifacts like broken and lost pieces or overlapping parts of fragments increase the error rate of shape based matching techniques. Therefore a combined approach of document analysis and shape matching is necessary for large instances of this problem. In this paper the preliminary snippet processing is described where the orientation of fragments, as well as the content like paper color and the color of the inks used is analyzed. The methods presented, are evaluated on database consisting of 690 snippets of Stasi files which were manually annotated to provide groundtruth data.