Multimedia Forensics (MF) is a relatively recent discipline whose goal is to analyse multimedia signals (images, audio tracks, video sequences) in order to recover potential evidences from them. Multimedia Forensics includes a set of scientific methodologies and technologies allowing to recover some information about digital content:
- Identification of the source device, i.e. the device that captured the content;
- Assessment of the integrity and the authenticity of content;
- Extraction of information from the multimedia signal.
The term “forensic” directly comes from the legal environment: nowadays the traditional interpretation of evidences and witnesses is changing, and a large part of lawyers, prosecutors and judges have to deal with digital evidences. By “digital evidences” we refer to all the evidences that are created, manipulated, stored or transmitted using digital devices (computer, wireless communication systems, internet, mobile phones, smart cards, sat-navs, and many other). While Computer Forensics mainly deals with scientific methods for the extraction of data from digital devices, Multimedia Forensics comes at a later stage and applies scientific methods to analyse such digital objects.
Multimedia Forensics is built on the basic idea that when a digital content is processed (starting from the acquisition phase, through the following possible compression, resampling and other operations) some kind of intrinsic traces are left in it, as a sort of digital fingerprints. Hence, by extracting such fingerprints and analysing their properties, it is possible to recover some information about the life-cycle of the digital data.
Regarding the source identification, MF assumes that the capturing device leaves some traces related to its intrinsic properties (e.g., sensor noise, lens distortion, and so on). Basing on such features, proper statistical tests can be devised to discriminate between computer-generated image/video from scanner- or camera- captured image/video; it is also possible to identify image/video coming from devices of different brands and models, or even to associate an image/video to the exact device that captured it.
Similarly, forgery detection algorithms try to assess the authenticity of the content, assuming that different processing operations leave peculiar statistical or visual features (e.g., artefacts due to JPEG compression); or that traces that are regularly introduced by the capturing device have been altered during tampering operations, thus leading to inconsistencies; it may also happen that some physical or geometric characteristic of the scene is not consistent due to manipulations (e.g., inconsistent lighting or perspective).
Finally, Multimedia Forensics also applies methods for signal enhancement (like noise suppression or distortion compensation) so to improve content understanding and interpretation. Moreover, MF applies some algorithms for the signal analysis in order to infer information from visual and audio data, e.g., colour analysis, pattern recognition, photogrammetric and biometric measures.