A sequential method for online steganalysis

This paper studies online detection of hidden information in digital images. By online, it is meant that we inspect a flow of images that are transmitted sequentially. This has crucial consequences on the detection of hidden data. By contrast to the usual detection settings, the delay before detection has to be considered in the definition of the correct detection probability, or power function. Similarly, the false alarm probability is considered with respect to a number of inspected cover images. In this paper a new sequential detection method is proposed with the goal to maximize the detection accuracy for a prescribed detection delay. The study of proposed sequential test allows the establishing of detection power as a function of detection delay and also permits us to bound the probability of false alarm for a given number of cover images. Numerical results on real image using both the well-known WS type of detector and the recent ensemble classifier show the relevance of the proposed approach and the accuracy of the theoretical findings.