Micenková, Barbora3; van Beusekom, Joost2; Shafait, Faisal2
1 Department of Computer Science, Science and Technology, Aarhus University2 DFKI Kaiserslautern3 Department of Computer Science, Science and Technology, Aarhus University
Stamps, along with signatures, can be considered as the most widely used extrinsic security feature in paper documents. In contrast to signatures, however, for stamps little work has been done to automatically verify their authenticity. In this paper, an approach for verification of color stamps is presented. We focus on photocopied stamps as non-genuine stamps. Our previously presented stamp detection method is improved and extended to verify that the stamp is genuine and not a copy. Using a variety of features, a classifier is trained that allows successful separation between genuine stamps and copied stamps. Sensitivity and specificity of up to 95% could be obtained on a data set that is publicly available.