Experiments on Improving Sensor Pattern Noise Extraction for Source Camera Identification

The source camera identification problem is concerned with the identification of the camera that has been used to generate a digital picture. A widely adopted identification technique, proposed by Lukas in [1], relies on the usage of the pattern noise left by the camera sensor as a fingerprint. This technique may perform badly when applied to images that have undergone lossy compression techniques, such as being saved as a low-quality JPEG image. In this paper, we firstly analyze the experimental performance of the identification technique by Lukas, when dealing with JPEG images saved using increasing compression rates. Then, we investigate if and how some of the enhanced sensor pattern noise extraction techniques proposed in literature are able to improve on the original technique in the considered cases. Our results show that, on a side, an increase in the compression rate of a JPEG image deeply affects the effectiveness of the identification process carried out using the Lukas technique. On the other side, we show that at least two of the considered enhanced sensor pattern noise extraction techniques succeed in recovering most part of this degradation.