<div>Steganalysis is the study of detecting secret information hidden in objects such as images, videos, texts, time series and games via steganography. Among those objects, the image is the most widely used object to hide secret messages. Detection of possible secret information hidden in images has attracted a lot of attention over the past ten years. People may conduct covert communications by exchanging images in which secret messages may be embedded in bits. One of main advantages of steganography over cryptography is that the former makes this communication insensible for human beings. So statistical methods or tools are needed to help distinguish cover images from stego images. <br></div><div><br></div><div>In this thesis, we start with a discussion of image steganography. Different kinds of embedding schemes for hiding secret information in images are investigated. We also propose a hiding scheme using a reference matrix to lower the distortion caused by embedding. As a result, we obtain Peak Signal-to-Noise Ratios (PSNRs) of stego images that are higher than those given by a Sudoku-based embedding scheme. Next, we consider statistical steganalysis of images in two different frameworks. We first study staganalysis in the framework of statistical hypothesis testing. That is, we cast a cover/stego image detection problem as a hypothesis testing problem. For this purpose, we employ different statistical models for cover images and simulate the effects caused by secret information embedding operations on cover images. Then the staganalysis can be characterized by a hypothesis testing problem in terms of the embedding rate. Rao’s score statistic is used to help make a decision. The main advantage of using Rao’s score test for this problem is that it eliminates an assumption used in the previous work where approximated log likelihood ratio (LR) statistics were commonly employed for the hypothesis testing problems.<br></div><div><br></div><div>We also investigate steganalysis using the deep learning framework. Motivated by neural network architectures applied in computer vision and other tasks, we propose a carefully designed a deep convolutional neural network architecture to classify the cover and stego images. We empirically show the proposed neural network outperforms the state-of-the-art ensemble classifier using a rich model, and is also comparable to other convolutional neural network architectures used for steganalysis.<br></div><div><br></div>The image databases used in the thesis are available on websites cited in the thesis. The stego images are generated from the image databases using source code from the website. <a href="http://dde.binghamton.edu/download/">http://dde.binghamton.edu/download/</a>

This thesis is divided into two parts. In the first part, we show how problems of statistical inference and combinatorial optimization may be approached within a unified framework that employs tools from fields as diverse as machine learning, statistical physics and information theory, allowing us to i) design algorithms to solve the problems, ii) analyze the performance of these algorithms both empirically and analytically, and iii) to compare the results obtained with the optimal achievable ones. In the second part, we use this framework to study two specific problems, one of inference (compressed sensing) and the other of optimization (information hiding). In both cases, we review current approaches, identify their flaws, and propose new schemes to address these flaws, building on the use of message-passing algorithms, variational inference techniques, and spin glass models from statistical physics. / Esta tese está dividida em duas partes. Na primeira delas, mostramos como problemas de inferência estatística e de otimização combinatória podem ser abordados sob um framework unificado que usa ferramentas de áreas tão diversas quanto o aprendizado de máquina, a física estatística e a teoria de informação, permitindo que i) projetemos algoritmos para resolver os problemas, ii) analisemos a performance destes algoritmos tanto empiricamente como analiticamente, e iii) comparemos os resultados obtidos com os limites teóricos. Na segunda parte, este framework é usado no estudo de dois problemas específicos, um de inferência (compressed sensing) e outro de otimização (ocultação de dados). Em ambos os casos, revisamos abordagens recentes, identificamos suas falhas, e propomos novos esquemas que visam corrigir estas falhas, baseando-nos sobretudo em algoritmos de troca de mensagens, técnicas de inferência variacional, e modelos de vidro de spin da física estatística.

Covert channels are becoming more popular as security risks grow in networks. One area that is promising for covert channels is wireless networks, since many use a collision avoidance scheme such as carrier sense multiple access with collision avoidance (CSMA/CA). These schemes often introduce randomness in the network, which provides good cover for a covert timing channel. In this thesis, we use the 802.11 standard as an example to demonstrate a wireless covert channel. In particular, most 802.11 configurations use a distributed coordinated function (DCF) to assist in communications. This DCF uses a random backoff to avoid collisions, which provides the cover for our covert channel. Our timing channel provides great improvements on other recent covert channels in the field of throughput, while maintaining high accuracy. We are able to achieve throughput over 8000 bps using Covert DCF, or by accepting a throughput of 1800 bps we can achieve higher covertness and 99% accuracy as well.

Steganographie ist die Kunst der vertraulichen Kommunikation. Anders als in der Kryptographie, wo der Austausch vertraulicher Daten für Dritte offensichtlich ist, werden die vertraulichen Daten in einem steganographischen System in andere, unauffällige Coverdaten (z.B. Bilder) eingebettet und so an den Empfänger übertragen. Ziel eines steganographischen Algorithmus ist es, die Coverdaten nur geringfügig zu ändern, um deren statistische Merkmale zu erhalten, und möglichst in unauffälligen Teilen des Covers einzubetten. Um dieses Ziel zu erreichen, werden verschiedene Ansätze der so genannten minimum-embedding-impact Steganographie basierend auf Syndromkodierung vorgestellt. Es wird dabei zwischen Ansätzen basierend auf stochastischen und auf deterministischen Matrizen unterschieden. Anschließend werden die Algorithmen bewertet, um Vorteile der Anwendung von Syndromkodierung herauszustellen.

This thesis is divided into two parts. In the first part, we show how problems of statistical inference and combinatorial optimization may be approached within a unified framework that employs tools from fields as diverse as machine learning, statistical physics and information theory, allowing us to i) design algorithms to solve the problems, ii) analyze the performance of these algorithms both empirically and analytically, and iii) to compare the results obtained with the optimal achievable ones. In the second part, we use this framework to study two specific problems, one of inference (compressed sensing) and the other of optimization (information hiding). In both cases, we review current approaches, identify their flaws, and propose new schemes to address these flaws, building on the use of message-passing algorithms, variational inference techniques, and spin glass models from statistical physics. / Esta tese está dividida em duas partes. Na primeira delas, mostramos como problemas de inferência estatística e de otimização combinatória podem ser abordados sob um framework unificado que usa ferramentas de áreas tão diversas quanto o aprendizado de máquina, a física estatística e a teoria de informação, permitindo que i) projetemos algoritmos para resolver os problemas, ii) analisemos a performance destes algoritmos tanto empiricamente como analiticamente, e iii) comparemos os resultados obtidos com os limites teóricos. Na segunda parte, este framework é usado no estudo de dois problemas específicos, um de inferência (compressed sensing) e outro de otimização (ocultação de dados). Em ambos os casos, revisamos abordagens recentes, identificamos suas falhas, e propomos novos esquemas que visam corrigir estas falhas, baseando-nos sobretudo em algoritmos de troca de mensagens, técnicas de inferência variacional, e modelos de vidro de spin da física estatística.

Unlike encryption, steganography hides the very existence of secret information rather than hiding its meaning only. Image based steganography is the most common system used since digital images are widely used over the Internet and Web. However, the capacity is mostly limited and restricted by the size of cover images. In addition, there is a tradeoff between both steganographic capacity and stego image quality. Therefore, increasing steganographic capacity and enhancing stego image quality are still challenges, and this is exactly our research main aim. Related to this, we also investigate hiding secret information in communication protocols, namely Simple Object Access Protocol (SOAP) message, rather than in conventional digital files. To get a high steganographic capacity, two novel steganography methods were proposed. The first method was based on using 16x16 non-overlapping blocks and quantisation table for Joint Photographic Experts Group (JPEG) compression instead of 8x8. Then, the quality of JPEG stego images was enhanced by using optimised quantisation tables instead of the default tables. The second method, the hybrid method, was based on using optimised quantisation tables and two hiding techniques: JSteg along with our first proposed method. To increase the steganographic capacity, the impact of hiding data within image chrominance was investigated and explained. Since peak signal-to-noise ratio (PSNR) is extensively used as a quality measure of stego images, the reliability of PSNR for stego images was also evaluated in the work described in this thesis. Finally, to eliminate any detectable traces that traditional steganography may leave in stego files, a novel and undetectable steganography method based on SOAP messages was proposed. All methods proposed have been empirically validated as to indicate their utility and value. The results revealed that our methods and suggestions improved the main aspects of image steganography. Nevertheless, PSNR was found not to be a reliable quality evaluation measure to be used with stego image. On the other hand, information hiding in SOAP messages represented a distinctive way for undetectable and secret communication.

Web services provide a platform neutral and programming language independent technology that supports interoperable machine-to-machine interaction over a network. Clients and other systems interact with Web services using a standardised XML messaging system, such as the Simple Object Access Protocol (SOAP), typically conveyed using HTTP with an XML serialisation in conjunction with other related Web standards. Nevertheless, the idea of applications from different parties communicating together raises a security threat. The challenge of Web services security is to understand and consider the risks of securing a Web-based service depending on the existing security techniques and simultaneously follow evolving standards in order to fill the gap in Web services security. However, the performance of the security mechanisms is fraught with concerns due to additional security contents in SOAP messages, the higher number of message exchanges to establish trust, as well as the extra CPU time to process these additions. As the interaction between service providers and requesters occurs via XML-based SOAP messages, securing Web services tends to make these messages longer than they would be otherwise and consequently requires interpretation by XML parsers on both sides, which reduces the performance of Web services. The work described in this thesis can be broadly divided into three parts, the first of which is studying and comparing the performance of various security profiles applied on a Web service tested with different initial message sizes. The second part proposes a multi-criteria decision making framework to aid Web services developers and architects in selecting the best suited security profile that satisfies the different requirements of a given application during the development process in a systematic, manageable, and effective way. The proposed framework, based on the Analytical Hierarchy Process (AHP) approach, incorporates not only the security requirements, but also the performance considerations as well as the configuration constraints of these security profiles. The framework is then validated and evaluated using a scenario-driven approach to demonstrate situations where the decision making framework is used to make informed decisions to rank various security profiles in order to select the most suitable one for each scenario. Finally, the last part of this thesis develops a novel steganography method to be used for SOAP messages within Web services environments. This method is based on changing the order of XML elements according to a secret message. This method has a high imperceptibility; it leaves almost no trail because it uses the communication protocol as a cover medium, and keeps the structure and size of the SOAP message intact. The method is empirically validated using a feasible scenario so as to indicate its utility and value.

Sammanhang. Genom att använda steganografi i bilder är det möjligt att lagra en stor mängd data utan att påverka bilden som information lagras i. En vanlig metod som används inom steganografi är LSB (Least Significant Bit), som ofta anses vara en av de första metoderna som användes inom steganografi. Förutom valet av metod har personen ett val att göra när hen ska välja en bild som ska representera bärare av information. Vad man ofta försöker åstadkomma är att skapa en komplex metod men glömmer bort bilden som ska användas som bärare. I slutändan är det bilden som kommer att användas och testas vid olika mätningar. Mål. Den här studien kommer att undersöka om olika typer av bildformat, BMP, PNG, JPEG och TIFF, har någon påverkan när jämförelse görs av originalbilden och den modifierade, givet att en steganografisk metod används för att lagra informationen. Studien kommer även att undersöka om bildrepresentationen har någon påverkan på mätningarna. Metod. En utbyggd metod av Least Significant Bit metoden kommer att implementeras och användas för att lagra information i olika typer av bilder med olika bildformat. Ett experiment sätts upp för att undersöka formaten med hjälp av mätningsverktygen MSE (Mean Squared Error), PSNR (Peek Signal-to-Noise Ratio) och SSIM (Structural Similarity). Resultat. Vid jämförelse av de grafer och tabeller som togs fram, sågs JPEG ha ett bättre resultat genom att ha ett lägre differentiellt värde mellan varje test. BMP, PNG och TIFF hade minimala skillnader mellan varandra för varje test. För bildrepresentationen visade två bilder högre differentiellt värde än de resterande tre. Sammanfattning. Resultaten från experimentet visade att vilken komprimeringsmetod som ett bildformat använder kommer ha påverkan på mätningsvärdena. Resultaten visade också att bildrepresentation kan ha en påverkan på mätningarna av en bild men mer data behövs för att dra en slutsats. / Context. By using image steganography it is possible to hide a large amount of data without making big differences to the initial picture. One commonly used method is Least Significant Bit (LSB), which often is considered one of the first method implemented and used in Image Steganography. Apart from the method, the user also have a choice when deciding what picture he or she should use as the carrier of information. What people often try to accomplish is to have a very complex method that hides the data in an efficient way, but forgets about the picture used as a carrier. In the end, all measurements will be done on the picture. Objectives. This study will investigate if different image formats, BMP, PNG, JPEG and TIFF, have an impact on the differences when comparing the original picture with the modified, given that data is stored with a steganographic method and is gradually increased. The study will also investigate if what the picture visually represent will have an effect on the measurements. Methods. An extended method of the Least Significant Bit method will be implemented and used to create different pictures with different kinds of image formats. An experiment will investigate these formats by taking measurements with MSE (Mean Squared Error), PSNR (Peek Signal-to-Noise Ratio) and SSIM (Structural Similarity). Results. When comparing different formats one could say that JPEG showed better performance by having a lower differential value between each test, by looking at the graphs and tables. BMP, PNG and TIFF had minimal changes between each other for each test. As for the visual representation of the pictures, two pictures showed a higher differential value after each test than the remaining three. Conclusions. The results from the experiment showed that which compression method a format uses will have an impact on the measurement. The results also showed that the pictures’ visual representation could have some impact on the measurement of a picture but more data is needed to conclude this theory.

De nos jours, le développement de la stéganalyse et de la stéganographie est incontournable, et peut être utilisé à des fins légales comme illégales, comme dans toute autre application. Le travail présenté dans cette thèse, se concentrant sur ces questions, est divisée en trois parties. La première partie concerne les paramètres permettant d’accroître le niveau de sécurité de la stéganographie afin de faire face aux techniques de stéganalyse. La contribution apportée dans cette première partie concerne l’étude de l’effet de la charge utile, l’extraction des caractéristiques, ainsi que le groupe d’images utilisées dans la phase d’apprentissage et la phase de test. Les résultats des simulations montrent que les techniques de stéganalyse de l’ état de l’art échouent dans la détection des messages secrets intégrés dans les images quand les paramètres changent entre l’apprentissage et le test. Dans la deuxième partie, nous étudions l’impact de la combinaison de plusieurs méthodes stéganographiques sur la détection des messages secrets. Ce travail prend en considération qu’il n’existe pas une procédure idéale, mais que le stéganographieur pourra utiliser n’importe quel schéma ainsi que n’importe quel taux d’embarquement. Dans la troisième et dernière partie, on propose une méthode qui calcule une carte de distorsion précise, en fonction de la dérivée seconde de l’image. La dérivée seconde est utilisée afin de calculer les courbes de niveau, ensuite le message va être caché dans l’image en écartant les courbes de niveaux inférieurs à un certain seuil. Les résultats expérimentaux démontrent que le niveau de sécurité est acceptable comparé aux méthodes stéganographiques de l’état de l’art. / In the recent time, the field of image steganalysis and steganography became more important due to the development in the Internet domain. It is important to keep in mind that the whole process of steganography and steganalysis can be used for legal or illegal operations like any other applications. The work in this thesis can be divided inthree parts. The first one concentrates on parameters that increase the security of steganography methods against steganalysis techniques. In this contribution the effect of the payload, feature extractions, and group of images that are used in the learning stage and testing stage for the steganalysis system are studied. From simulation, we note that the state of the art steganalyzer fails to detect the presence of a secret message when some parameters are changed. In the second part, we study how the presence of many steganography methods may influence the detection of a secret message. The work takes into consideration that there is no ideal situation to embed a secret message when the steganographier can use any scheme with any payloads. In the third part, we propose a method to compute an accurate distortion map depending on a second order derivative of the image. The second order derivative is used to compute the level curve and to embed the message on pixels outside clean level curves. The results of embedding a secret message with our method demonstrate that the result is acceptable according to state of the art steganography.

Abstract Broadband communication networks and multimedia data available in a digital format opened many challenges and opportunities for innovation. Versatile and simple-to-use software and decreasing prices of digital devices have made it possible for consumers from all around the world to create and exchange multimedia data. Broadband Internet connections and near error-free transmission of data facilitate people to distribute large multimedia files and make identical digital copies of them. A perfect reproduction in digital domain have promoted the protection of intellectual ownership and the prevention of unauthorized tampering of multimedia data to become an important technological and research issue. Digital watermarking has been proposed as a new, alternative method to enforce intellectual property rights and protect digital media from tampering. Digital watermarking is defined as imperceptible, robust and secure communication of data related to the host signal, which includes embedding into and extraction from the host signal. The main challenge in digital audio watermarking and steganography is that if the perceptual transparency parameter is fixed, the design of a watermark system cannot obtain high robustness and a high watermark data rate at the same time. In this thesis, we address three research problems on audio watermarking: First, what is the highest watermark bit rate obtainable, under the perceptual transparency constraint, and how to approach the limit? Second, how can the detection performance of a watermarking system be improved using algorithms based on communications models for that system? Third, how can overall robustness to attacks to a watermark system be increased using attack characterization at the embedding side? An approach that combined theoretical consideration and experimental validation, including digital signal processing, psychoacoustic modeling and communications theory, is used in developing algorithms for audio watermarking and steganography. The main results of this study are the development of novel audio watermarking algorithms, with the state-of-the-art performance and an acceptable increase in computational complexity. The algorithms' performance is validated in the presence of the standard watermarking attacks. The main technical solutions include algorithms for embedding high data rate watermarks into the host audio signal, using channel models derived from communications theory for watermark transmission and the detection and modeling of attacks using attack characterization procedure. The thesis also includes a thorough review of the state-of-the-art literature in the digital audio watermarking.