Matthias Prandtstetter

Title:
Reconstructing sheets of manually torn paper

Abstract:
In this work, we try to reconstruct sheets of paper manually torn for
being able to extract the information contained in the original
document. One application - among others - is of political interest,
since many secret documents were manually destructed in the former
German Democratic Republic (GDR) during 1989. A previous approach for
reconstructing such paper sheets relies on classical image recognition
techniques considering the content of pieces, their exact shape of
borders, and associated matching algorithms. Practice has shown that
such an approach is often misled due to various effects.
We present two alternative approaches in which we focus our attention on
more reliable structural properties of pieces.
The first one formulates the problem as a variant of a knapsack problem,
whereas the second model tries to exploit the underlying tearing process
for reconstructing the original document.