A Markov Process Based Approach to Effective Attacking JPEG Steganography

Abstract

Abstract. In this paper, a new steganalysis scheme is presented to effectively detect the advanced JPEG steganography. For this purpose, we first choose to work on JPEG 2-D arrays formed from the magnitudes of JPEG quantized block DCT coefficients. Difference JPEG 2-D arrays along horizontal, vertical and diagonal directions are then used to enhance changes caused by JPEG steganography. Markov process is applied to modeling these difference JPEG 2-D arrays so as to utilize the second order statistics for steganalysis. In addition to the utilization of difference JPEG 2-D arrays, a thresholding technique is developed to greatly reduce the dimensionality of transition probability matrices, i.e., the dimensionality of feature vectors, thus making the computational complexity of the proposed scheme manageable. The experimental works are presented to demonstrate that the proposed scheme has outperformed the existing steganalyzers in attacking OutGuess, F5, and MB1.

Steganalysis of Content-Adaptive Steganography in Spatial Domain

“…What is really needed for steganalysis is an unbiased estimate of the central pixel obtained from the neighboring pixels, excluding the pixel being estimated. The recently proposed SPAM feature set [13], as well as the earlier work [2,15], use the value of the neighboring pixel as the prediction:…”

“…To reduce its dimensionality, features are usually constructed as some integral quantities. Considering the noise residual as a Markov chain, one can take its sample transition probability matrix [2,13,15] or the sample joint probability matrix (the co-occurrence matrix) as a feature. To capture higher-order dependencies among pixels, higher-order co-occurrence matrices are usually formed.…”

YASS: Yet Another Steganographic Scheme That Resists Blind Steganalysis

“…Cover memory has been shown to be very important to steganalysis [22], and is incorporated into the feature vector in several ways, e.g. [23,15]. 3.…”

“…It is known that, the higher order statistics, in general, are difficult to match, model, or restore. Recently, blind steganalysis algorithms [9][10][11][12][13][14][15][16] have been proposed that employ supervised learning to distinguish between the plain cover and stego images, and also identify the particular hiding algorithm used for steganography. These techniques bank on the fact that there are some image features that are modified during the embedding process which can be used as an input to the learning machine.…”

Breaking HUGO – The Process Discovery

“…This is where the embedding changes are localized and thus most pronounced. This strategy, originally coined in 2004 [6], was later confirmed in [6,10,17,15,18,4]. Because HUGO's embedding domain is known, after the early failures described in the previous two sections, we revisited the pixel domain and achieved a major breakthrough on September 23, 2010.…”