Abstract. A new, simple, approach for active steganography is proposed in this paper that can successfully resist recent blind steganalysis methods, in addition to surviving distortion constrained attacks. We present Yet Another Steganographic Scheme (YASS), a method based on embedding data in randomized locations so as to disable the selfcalibration process (such as, by cropping a few pixel rows and/or columns to estimate the cover image features) popularly used by blind steganalysis schemes. The errors induced in the embedded data due to the fact that the stego signal must be advertised in a specific format such as JPEG, are dealt with by the use of erasure and error correcting codes. For the presented JPEG steganograhic scheme, it is shown that the detection rates of recent blind steganalysis schemes are close to random guessing, thus confirming the practical applicability of the proposed technique. We also note that the presented steganography framework, of hiding in randomized locations and using a coding framework to deal with errors, is quite simple yet very generalizable. Key words: data hiding, error correcting codes, steganalysis, steganography, supervised learning. 1

Abstract—The problems of batch steganography and pooled steganalysis, proposed in [1], generalize the problems of hiding and detecting hidden data to multiple covers. It was conjectured that, given covers of uniform capacity and a quantitative steganalysis method satisfying certain assumptions, “secure ” steganographic capacity is proportional only to the square root of the number of covers. We now prove that, with respect to a natural definition of secure capacity, and in a suitably asymptotic sense, this conjecture is true. This is in sharp contrast to capacity results for noisy channels. Index Terms—Channel capacity, communication systems, information hiding, steganography.

Abstract—This paper considers the least squares method (LSM) for estimation of the length of payload embedded by least-significant bit replacement in digital images. Errors in this estimate have already been investigated empirically, showing a slight negative bias and substantially heavy tails (extreme outliers). In this paper, (approximations for) the estimator distribution over cover images are derived: this requires analysis of the cover image assumption of the LSM algorithm and a new model for cover images which quantifies deviations from this assumption. The theory explains both the heavy tails and the negative bias in terms of cover-specific observable properties, and suggests improved detectors. It also allows the steganalyst to compute precisely, for the first time, a-value for testing the hypothesis that a hidden payload is present. This is the first derivation of steganalysis estimator performance. Index Terms—Least-significant bit (LSB) embedding, steganography, structural steganalysis. I.

We present a new benchmark for binary steganalysis methods, based on the asymptotic information (in the entropic sense) it gives about the presence of hidden data. The theoretical foundation is quite unlike ad hoc performance measures found in steganalysis literature that are based on false positive and negative rates. It is argued that this new metric is an application-independent long-run measure of true performance. There are some challenges to computing the benchmark empirically, and some suggested methods are presented, but no definitive answer emerges. As a simple case study, some steganalysis methods from the literature are evaluated using these techniques.

This paper draws together two methodologies for the detection of bit replacement steganography: the principle of maximum likelihood, which is statistically well-founded but has lead to weak detectors in practice, and so-called structural detection, which is sensitive but lacks optimality and can suffer from complicated exposition. The key novelty is to extend structural analysis to include a hypothetical “pre-cover”, from which the cover object is imagined to derive. Here, maximum likelihood detection is presented for three structural detectors. Although the algebraic derivation is long, and maximizing the likelihood function difficult in practice, conceptually the new detectors are reasonably simple. Experiments show that the new detectors are the best performers yet, very significantly so in the detection of replacement of multiple bit planes.

This paper proposes a complete practical methodology for minimizing additive distortion in steganography with general (non-binary) embedding operation. Let every possible value of every stego element be assigned a scalar expressing the distortion of an embedding change done by replacing the cover element by this value. The total distortion is assumed to be a sum of per-element distortions. Both the payload-limited sender (minimizing the total distortion while embedding a fixed payload) and the distortion-limited sender (maximizing the payload while introducing a fixed total distortion) are considered. Without any loss of performance, the non-binary case is decomposed into several binary cases by replacing individual bits in cover elements. The binary case is approached using a novel syndromecoding scheme based on dual convolutional codes equipped with the Viterbi algorithm. This fast and very versatile solution achieves state-of-the-art results in steganographic applications while having linear time and space complexity w.r.t. the number of cover elements. We report extensive experimental results for a large set of relative payloads and for different distortion profiles, including the wet paper channel. Practical merit of this approach is validated by constructing and testing adaptive embedding schemes for digital images in raster and transform domains. Most current coding schemes used in steganography (matrix embedding, wet paper codes, etc.) and many new ones can be implemented using this framework.

Recent results on the information theory of steganography suggest, and under some conditions prove, that the detectability of payload is proportional to the square of the number of changes caused by the embedding. Assuming that result in general, this paper examines the implications for an embedder when a payload is to be spread amongst multiple cover objects. A number of variants are considered: embedding with and without adaptive source coding, in uniform and nonuniform covers, and embedding in both a fixed number of covers (so-called batch steganography) as well as establishing a covert channel in an infinite stream (sequential steganography, studied here for the first time). The results show that steganographic capacity is sublinear, and strictly asymptotically greater in the case of a fixed batch than an infinite stream. In the former it is possible to describe optimal embedding strategies; in the latter the situation is much more complex, with a continuum of strategies which approach the unachievable asymptotic optimum.

Abstract. This paper is concerned with the estimation of steganographic capacity in digital images, using information theoretic bounds and very large-scale experiments to approximate the distributions of genuine covers. The complete distribution cannot be estimated, but with carefullychosen algorithms and a large corpus we can make local approximations by considering groups of pixels. A simple estimator for the local quadratic term of Kullback-Leibler divergence (Steganographic Fisher Information) is presented, validated on some synthetic images, and computed for a corpus of covers. The results are interesting not so much for their concrete capacity estimates but for the comparisons they provide between different embedding operations, between the information found in differentlysized and-shaped pixel groups, and the results of DC normalization within pixel groups. This work suggests lessons for the future design of spatial-domain steganalysis, and also the optimization of embedding functions. 1

Abstract. In this paper, we study embedding efficiency, which is an important attribute of steganographic schemes directly influencing their security. It is defined as the expected number of embedded random message bits per one embedding change. Constraining ourselves to embedding realized using linear covering codes (so called matrix embedding), we show that the quantity that determines embedding efficiency is not the covering radius but the average distance to code. We demonstrate that for linear codes of fixed block length and dimension, the highest embedding efficiency (the smallest average distance to code) is not necessarily achieved using codes with the smallest covering radius. Nevertheless, we prove that with increasing code length and fixed rate (i.e., fixed relative message length), the relative average distance to code and the relative covering radius coincide. Finally, we describe several specific examples of q-ary linear codes with q matched to the embedding operation and experimentally demonstrate the improvement in steganographic security when incorporating the coding methods to digital image steganography. 1

We extend the square root law of steganographic capacity, for the simplest case of iid covers, in two ways. First, we show that the law still holds under a more realistic embedding assumption, where the payload is of fixed length (instead of, in the classic result, independent embedding at each location). Second, we consider the case of nonuniform embedding paths, which is forced when the stegosystem’s secret key is of limited size: we show that the secret key must be of length at least linear in the payload size, if a square root law is to hold. The latter is parallel to Shannon’s perfect cryptography bound. Categories and Subject Descriptors D.2.11 [Software Engineering]: Software Architectures— information hiding; H.1.1 [Models and Principles]: Systems