Image Forensics in the Wild

In the last years, visual digital data gained a key role in sharing information. Millions of images are shared everyday in the wild web among worldwide users. On the other hand, visual data can be easily tampered by means of available retouching tools to create realistic forgeries. This fact poses the problem of relying on digital images as source of information, especially when these data are used for intelligence purposes or are exhibited as a source of potential evidence in legal acts. In recent years Image Forensics has been proposed as a solution to determine the authenticity of digital images. This technology concerns the analysis of the traces left by processing applied to the digital image to determine its lifecycle (e.g., the acquisition device, the processing it has been subjected to). To date, several tools have been provided by the research community to look into an image at different levels of depth. Anyway, most of them are not ready to work in real case scenarios. Specifically, when a query is demanded by a legal part, problems may arise, mainly related to the methodology to be applied, the reliability of the algorithms and the interpretation of the achieved results to produce a final record on the investigated image. This work focuses on improving the applicability of some forensic technologies in real case scenarios, where images belong to less controlled environments. This thesis addresses the following issues: i) Provide a methodology to investigate a digital image with several tools and provide results to be presented in court; ii) Improve the available of some forensic tools to solve current limitations; iii) Assess the accuracy variability of some available technologies under specific conditions; iv) Develop new applications for the available technologies that take advantage of side information.