Document Forgery Detection using Printer Source Identification—A Text-Independent Approach

Source printer identification for printed document forgery detection has gained much significance in recent years. Traditional techniques employed in the literature are highly text dependent and therefore may prove insufficient in certain scenarios. In this study, we present a text-independent approach for effective characterization of source printer, using deep visual features. By employing transfer learning on pre-trained Convolutional Neural Networks (CNNs), we achieve significant recognition results on a dataset of 1200 documents from 20 different printers (13 laser and 7 inkjet). A comparison with various conventional features on the same dataset demonstrate that our proposed methodology classifies printed documents more accurately and effectively.