Treffer: Universal JPEG steganalysis in the compressed frequency domain

We present in this paper a new approach for universal JPEG steganalysis and propose studying statistics of the compressed DCT coefficients. This approach is motivated by the Avalanche Criterion of the JPEG lossless compression step. This criterion makes possible the design of detectors whose detection rates are independent of the payload. We design an universal steganalytic scheme using blocks of the JPEG file binary output stream. We compute higher order statistics over their Hamming weights and combined them with a Kullbak-Leibler distance between the probability density function of these weights and a benchmark one. We evaluate the universality of our detector through its capacity to efficiently detect the use of a new algorithm not used during the training step. To that goal, we examinate training sets produced by Outguess, F5 and JPhide-and-Seek. The experimental results we obtained show that our scheme is able to detect the use of new algorithms with high detection rate (~ 90%) even with very low embedding rates (< 10-5).