A computational approach for printed document forensics using SURF and ORB features

Document forgery is quite common nowadays due to the availability of cost-effective scanners and printers. Important documents like certificates, passport, identification cards, etc., are protected using watermarks or signatures. These are made secured with a protective printing mechanism with extrinsic fingerprints. Therefore, it is easy to authenticate such documents. Other documents required a passive approach for their authentication. These approaches look for document inconsistencies for chances of modification. Some of these attempt to detect and fix the source of the printed document. This paper proposes a classifier-based model to identify the source printer and classify the questioned document in one of the printer classes. A novel approach of utilizing Speeded Up Robust Features and Oriented Fast Rotated and BRIEF feature descriptors is proposed for printer attribution. Naive Bayes, k-NN, random forest and different combinations of these classifiers have been experimented for classification. The proposed model can efficiently classify the questioned documents to their respective printer class. An accuracy of 86.5% has been achieved using a combination of Naive Bayes, k-NN, random forest classifiers with a simple majority voting scheme and adaptive boosting methodology.