Digital forensics : blind image tampering detection

Abstract

Advances in both hardware and commercial software have made the effortless editing of digital media possible. Unfortunately, this has led to a significant drop in the trust and reliability people place in photography. Research on countermeasure schemes capable of detecting such fraud is still in its infancy and the need for new methods and computational algorithms to help detect forged or tampered-with digital images, has arisen. This is especially true since photos are used as evidence in court, in the news, in insurance claims and also in national intelligence analysis scenarios amongst others. To prove that an image's authentic, current methods rely on fragile or semi-fragile watermarks or signatures. However, this requires dedicated cameras or recording devices. Moreover, a lot of forgeries are produced with pictures downloaded from the internet which may have easily been taken by different people using different equipment and can also have originally been encoded with different image compression algorithms. To be able to be applied within a forensic setting, techniques that detect and identify forged regions must perform the respective analysis on a single file. In this work we study the state-of-the-art blind methods that authenticate photos by searching for irregularities within the image itself. Such techniques analyse the images on different levels. For instance, some try to detect manipulations by looking at low level properties such as noise patterns or the quality of edges. Other methods consider the optical properties of the image using checks for light consistencies or by determining that all objects are within the expected perspective. At times, although an experienced forgerer may produce a very authentic looking photo, humans can immediately realize that the image is forged just by looking at the content.