Printed Documents Forensics Based On Intrinsic Degradation

Printed document are widely used more and more, while at the same time the number of cases related to forged documents occur increasingly in recent years, such as bogus contracts, falsified documents evidence, illegal leaflets. In the court trial, it is needed to verify the authenticity of documents as evidence; in the criminal investigation process, it is needed to find the source printer of documents. Thus, judicial and public security departments urgently need the printed documents forensics technology, including testing whether the document have been forged, as well as determining which printer the documents came from.In the printed documents forensics technology, the key lies in how to extract the intrinsic features contained in the printed documents. This paper focuses on the extraction of page's geometric distortion features from the printed documents, which represents the printed contents' overall distortion of the geometry deformation caused by the defects of mechanical devices during the printing process.This paper designs the experiments to find the page's geometric distortion phenomenon, which can be understood as the geometric deformation part of the image degradation. Then projective transformation and elastic transformation models are used respectively to model the page's geometric distortion, from which the page's geometric distortion features are extracted. As the experiment shown, projective transformation model can describe the page's geometric distortion well, and more suitable models sould be experimented in the future research.First we try to use the projective transformation to model the page's geometric distortion. The purpose of pre-processing process was to get the set of feature point pair, least squares method was to estimate the parameters of the model, and the residual error matrix can be obtained through the estimated parameters. Partial parameters of the model and the residual error matrix all contain the intrinsic features of printer. Both the classification of the selected parameters and comparison of the residual error matrix can help us link a document to a printer. The accuracy of the classification was 100% in the experiment containing 10 printers.Then we use the elastic transformation model. The process has been adjusted towards different printed document content, such as the ICP algorithm for matching feature points set is included in the process. We experimented thin plate spline elastic transformation.