Camera identification based on sensor noise pattern: a practical procedure for open scenarios

In this paper, the problem of device identification based on sensor noise pattern in open scenarios is addressed. This context is common in practice and quite challenging because of the lack of reference to evaluate statistical similarity measures between image and suspect camera noise patterns. A device identification procedure based on an artificial neural network classifier is proposed, whose parameters are optimised by extreme learning machine algorithm and repeated double cross validation techniques. In addition, a strategy to select training patterns, aiming at open scenario situations, is presented. Experimental results are shown for the sake of performance assessment.