Article Preview
TopIntroduction
The advent of image processing technologies easily enables modification and manipulation of digital visual data, so that we are no longer confident that what we are seeing in a photo is a true representation of what really happened: the value of photography as a record of events must be carefully evaluated. Such a need comes from different fields of application: one of the most important is the forensic scenario, in which the trustworthiness of images must be assured before using them as potential evidences. Image Forensics (IF) (under the umbrella of the more general Digital Forensics) is the science addressing the validation, identification, analysis, interpretation of digital images as potential evidences. One of the most interesting tasks in IF is splicing detection, that aims at understanding if a given photo is a composition of different shots. Several approaches for splicing detection have been proposed recently (Piva, 2013), sharing the same basic idea: creating a forgery usually requires some processing steps, and these leave some statistical footprints into the signal.
In this context, the Image Forensic Analyst (IFA from now on) is the professional that applies technological means for extracting information on image history and for assuring its credibility, after the chain of custody (COC) procedures have been applied for acquiring, transferring and storing the visual data (see Figure 1). Usually, the IFA has not any previous knowledge about the history of the images that he is considering (i.e., what device acquired them, whether a processing software has been used to edit them or not, and so on), and must produce a report about the credibility of the analysed contents. To reach this goal, the IFA today could use algorithms developed in the IF literature, but in practice several problems rise: algorithms are stand-alone, in that they focus on a specific footprint and ignore the others; they assume some prior knowledge about the kind of processing that could have been carried on the media; and, finally, they do not expose a user interface helping the IFA in setting up the analysis and interpreting the results.
Figure 1. A simplified version of the chain of custody, where the positioning of the proposed tool is highlighted
As a result, several issues are still open when we consider to apply the results coming from academic research to practical cases, where the IFA needs technological instruments that facilitate him in reaching a conclusion:
- •
There are no tools that help the IFA to exploit the different capabilities of existing algorithms. We should consider that, in the end, the IFA is mainly concerned about image integrity (i.e. Algorithm output), and only indirectly concerned about footprint detection (i.e. Algorithm functioning);
- •
Usually the presence/absence of forensic fingerprints can be verified on the image as a whole, or on a selected suspected region; only few examples of tools that provide a fine-grained localization of forgery within a digital image have been proposed;
- •
Each tool usually considers to reveal one specific trace of tampering, but the IFA cannot know in advance which traces should be searched for. Therefore, there is need for a tool that helps in interpreting and putting together the outputs from different algorithms.
For these reasons, we believe that providing a comprehensive system for image splicing detection is an important contribution for the diffusion of image forensic technologies.
In this work we present a tool for evaluating the integrity of a digital image, by revealing whether the image is a malicious composition of different contents or an original shot of an event. This tool combines different image forensic algorithms for splicing detection through a specifically tailored decision fusion framework, improving the detection performance with respect to single tools, and provides an intuitive and functional interface that allows the IFA to easily access this multi-clue analysis. The system we propose contributes to solve each of the previously listed problems by: