Security of Forensic Techniques for Digital Images

Digital images are used everywhere in modern life and mostly replace traditional photographs. At the same time, due to the popularity of image editing tools, digital images can be altered, often leaving no obvious evidence. Thus, evaluating image authenticity is indispensable. Image forensic techniques are used to detect forgeries in digital images in the absence of embedded watermarks or signatures. Nevertheless, some legitimate or illegitimate image post-processing operations can affect the quality of the forensic results. Therefore, the reliability of forensic techniques needs to be investigated. The reliability is understood in this case as the robustness against image post-processing operations or the security against deliberated attacks. In this work, we first develop a general test framework, which is used to assess the effectiveness and security of image forensic techniques under common conditions. We design different evaluation metrics, image datasets, and several different image post-processing operations as a part of the framework. Secondly, we build several image forensic tools based on selected algorithms for detecting copy-move forgeries, re-sampling artifacts, and manipulations in JPEG images. The effectiveness and robustness of the tools are evaluated by using the developed test framework. Thirdly, for each selected technique, we develop several targeted attacks. The aim of targeted attacks against a forensic technique is to remove forensic evidence present in forged images. Subsequently, by using the test framework and the targeted attacks, we can thoroughly evaluate the security of the forensic technique. We show that image forensic techniques are often sensitive and can be defeated when their algorithms are publicly known. Finally, we develop new forensic techniques which achieve higher security in comparison with state-of-the-art forensic techniques