We describe a statistical model for natural images that is built upon a multi-scale wavelet decomposition. The model consists of first- and higher-order statistics that capture certain statistical regularities of natural images. We show how this model can be useful in several digital forensic applications, specifically in detecting various types of digital tampering.

This article discusses existing steganographic systems and presents recent research in detecting them via statistical steganalysis. Other surveys focus on the general usage of information hiding and watermarking or else provide an overview of detection algorithms

Steganographic messages can be embedded into digital images in ways that are imperceptible to the human eye. These messages, however, alter the underlying statistics of an image. We previously built statistical models using first-and higher-order wavelet statistics, and employed a non-linear support vector machines (SVM) to detect steganographic messages. In this paper we extend these results to exploit color statistics, and show how a one-class SVM greatly simplifies the training stage of the classifier.

Computer graphics rendering software is capable of generating highly photorealistic images that can be impossible to differentiate from photographic images. As a result, the unique stature of photographs as a definitive recording of events is being diminished (the ease with which digital images can be manipulated is, of course, also contributing to this demise). To this end, we describe a method for differentiating between photorealistic and photographic images. Specifically, we show that a statistical model based on first- and higher-order wavelet statistics reveals subtle but significant differences between photorealistic and photographic images.

Techniques for information hiding (steganography) are becoming increasingly more sophisticated and widespread. With high-resolution digital images as carriers, detecting hidden messages is also becoming considerably more difficult. We describe a universal approach to steganalysis for detecting the presence of hidden messages embedded within digital images. We show that, within multi-scale, multi-orientation image decompositions (e.g., wavelets), first- and higher-order magnitude and phase statistics are relatively consistent across a broad range of images, but are disturbed by the presence of embedded hidden messages. We show the efficacy of our approach on a large collection of images, and on eight different steganographic embedding algorithms.

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

A mathematical framework for steganalysis is presented in this paper with linear steganography being the main focus. A mathematically formal definition of steganalysis is given. Then, active steganalysis defined as the extraction of a hidden message with little or no a priori information is formulated as blind system identification problem within this framework. Conditions for identifiability (i.e., successful steganalysis) are derived. A procedure to systematically exploit any available spatial and temporal diversity information for e#cient steganalysis is also discussed.

This paper proposes multiple regression models as a method for quantitative evaluation of the accuracy in steganalysis with respect to various moderating factors, such as parameter choice of the detector and properties of the carrier object. The case for multivariate statistical inference in steganalysis is particularly relevant: recent findings suggest that type and characteristics of carrier do matter, but the precise relations remain still opaque. In this paper we provide an exemplary comparison between two length-estimating attacks against LSB steganography. Extensions and applications for improved steganalysis are addressed.

The recent approach of a model-based framework for steganography fruitfully contributes to the discussion on the security of steganography. In addition, the first proposal for an embedding algorithm constructed under the model-based paradigm reached remarkable performance in terms of capacity and security. In this paper, we review the emerging of model-based steganography in the context of decent steganalysis as well as from theoretical considerations, before we present a method to attack the above-mentioned scheme on the basis of first order statistics. Experimental results show a good detection ratio for a large test set of typical JPEG images. The attack is successful because of weaknesses in the model and does not put into question the generalised theoretical framework of model-based steganography. So we discuss possible implications for improved embedding functions.

Abstract. In this paper 1, a steganalysis scheme based on multiple features formed by statistical moments of wavelet characteristic functions is proposed. Our theoretical analysis has pointed out that the defined n-th statistical moment of a wavelet characteristic function is related to the n-th derivative of the corresponding wavelet histogram, and hence is sensitive to data embedding. The selection of the first three moments of the characteristic functions of wavelet subbands of the three-level Haar wavelet decomposition as well as the test image has resulted in total 39 features for steganalysis. The effectiveness of the proposed system has been demonstrated by extensive experimental investigation. The detection rate for Cox et al.’s non-blind spread spectrum (SS) data hiding method, Piva et al.’s blind SS method, Huang and Shi’s 8 × 8 block SS method, a generic LSB method (as embedding capacity being 0.3 bpp), and a generic QIM method (as embedding capacity being 0.1 bpp) are all above 90 % over all of the 1096 images in the CorelDraw image database using the Bayes classifier. Furthermore, when these five typical data hiding methods are jointly considered for steganalysis, i.e., when the proposed steganalysis scheme is first trained sequentially for each of these five methods, and is then tested blindly for stegoimages generated by all of these methods, the success classification rate is 86%, thus pointing out a new promising approach to general blind steganalysis. The detection results of steganalysis on Jsteg, Outguess and F5 have further demonstrated the effectiveness of the proposed steganalysis scheme. 1