Anti-Forensic Technique for Median Filtering using L1-L2 TV Model

Median filtering is one of the most popular image anti-forensic technique. Due to its nonlinearity, median filtering effectively hides traces of different types of image manipulation and therefore is integrated into several Anti-Forensic algorithms. As a result, the detection of median filtering has become of great significance for forensic analysis. To this end, several median filtering detectors have been proposed. In order to counter this, various anti-forensic techniques have also been suggested. In this paper we propose an effective anti-forensic technique that performs the median filtering operation on the image while ensuring minimal change in its spatial characteristics. The anti-forensic attack is formulated as an optimization problem and is constrained by a difference of Anisotropic and Isotropic Total Variation regularization. We demonstrate the effectiveness of the proposed technique by performing various experiments on popular median filtering detectors and observing the degradation of their performance on the anti-forensically generated images.