Non-intrusive Multimedia Authentication and Integrity Verification
Funding Body: EPSRC & Forensic Pathways Ltd (EPSRC CASE Award)
Project Duration: 01 February 2012 to 31 July2015
Imaging devices leave invisible and unique fingerprints in the images they produce. Digital fingerprints can be sensor pattern noise caused by imperfections during the sensor manufacturing stage, lens aberration, camera response function and algorithms or parameters of camera components such as colour filter array, colour interpolation matrices and quantisation tables. Researchers have been exploiting these types of device fingerprints for source device identification and content-integrity verification. However, most approaches use only one type in their systems. As a result, the systems' applicability is limited. This project is about using a vector of digital fingerprint left by imaging devices in images to authenticate image content and detect forgery. To devise a combined vector of various types of device fingerprints, we need to characterise hardware components, fuse fingerprints of different modalities and attenuate the influence of details from the scene. The detection of tampering can be mapped to a two-class pattern recognition problem and be approached by starting with a feature extraction followed by a statistical pattern classification process, which can employ machine learning techniques.