Forensic document examination system using boosting and bagging methodologies

Document forgery has increased enormously due to the progression of information technology and image processing software. Critical documents are protected using watermarks or signatures, i.e., active approach. Other documents need passive approach for document forensics. Most of the passive techniques aim to detect and fix the source of the printed document. Other techniques look for the irregularities present in the document. This paper aims to fix the document source printer using passive approach. Hand-crafted features based on key printer noise features (KPNF), speeded up robust features (SURF) and oriented FAST rotated and BRIEF (ORB) are used. Then, feature-based classifiers are implemented using K-NN, decision tree, random forest and majority voting. The document classifier proposed model can efficiently classify the questioned documents to their respective printer class. Further, adaptive boosting and bootstrap aggregating methodologies are used for the improvement in classification accuracy. The proposed model has achieved the best accuracy of 95.1% using a combination of KPNF + ORB + SURF with random forest classifier and adaptive boosting methodology.