Video camera identification using audio-visual features

One of the major issues in multimedia forensics is the identification of video acquisition devices. Most of the relevant state-of-the-art solutions rely on either visual or audio analysis, using feature arrays that are highly correlated with the characteristics of the respective camera or microphone. In this work, we present a multi-modal approach that uses both video and audio information to improve the detection accuracy. For this purpose, microphone detection based on the blind estimation of the frequency response is complemented with a video camera detection based on a set of video features related to the Color Filter Array interpolation. Experimental results show that the combined approach results in an improved overall classification accuracy over the mono-modal cases.