Spring 2007   InfAs 535

Steganography, Watermarking & Steganalysis

Summaries of Past Projects

2006

Steganography in the TCP/IP Data Stream, Andy Hoernecke.  Various forms of steganography have been used for many years. With the proliferation of information technologies, it is important that society be aware of the power of this useful technology. The Internet provides an excellent medium for the transmission of this hidden data. Although some steganographic techniques have been invented or theorized using an Ethernet data stream, many are easily detected. This paper will explore the possibility of hiding data not in the data stream itself, but instead in part of a protocol used to send data—specifically TCP’s acknowledgement error control mechanism.

SVD Embedding In Wavelet Domain, Andrew Regenscheid. This paper explores a steganographic algorithm that
embeds a covert message in an image file. The algorithm presented in this paper is a modification of an algorithm first proposed by Bergman and Davidson in [1]. Their algorithm uses the singular value decomposition to transform each n x n block from a spatial image into three matrices U, S and V. I review their method for embedding and extracting bits into the matrix U. My modification to the algorithm involves transforming the image data into the wavelet domain, where we then apply Bergman and Davidson’s algorithm. Experimental results indicate that this method has the potential to reduce the perceptibility of embedding, but there is a significant increase in the error rate. Further improvements are needed to make this a practical embedding scheme. (Implemented.)

The Viability of Digital Watermarking in Data
Management at a Public Health Laboratory, Brian Wels.  Analytical laboratories receive samples from customers and produce reports that contain data specific to the samples received. Data produced by the laboratory is considered confidential and belongs to the customer that is paying for the analytical services. While the customers are free to distribute this data to anyone they choose the laboratory is obligated to restrict access to the data. If an unauthorized party obtained a copy of the laboratory report, was it obtained covertly from the customer or from the laboratory? The consequences for the laboratory could negatively impact their reputation and future earnings. An invisible watermark applied to the report that contains the date and time of transmission could help track and identify the source of the leak.

Audio Watermarking to Improve Caller-ID Security in VoIP Communication, Bryan Ellingson.  The focus of this paper is to resolve a problem in current VoIP deployments. No extensive knowledge of VoIP or watermarking is required for the intended audience of watermarking and communication enthusiasts. A resolution is needed to protect users from being victims of fraud or other unwanted harm commonly perpetrated by ‘spoofing'. This paper presents an overview of VoIP deployments, with detail on the SIP protocol. Flaws in these deployments relating to VoIP caller-id services are also exposed. As a result of these flaws a new approach to VoIP security is examined. Two solutions are proposed which utilize audio watermarking to improve caller-id security and can be deployed independent of the watermarking algorithm used. To that end several watermarking concepts are evaluated for suitability in the proposed solutions. The final goal is to utilize the knowledge gained from these evaluations in order to perform future tests on the solutions. The solutions and test results will be used to further ongoing research by Dr. Guan and myself in the area of VoIP security. (Implemented.)

Internet Surveillance and Filtering: Steganography-based Circumvention Techniques, Christophe Veltsos.  This paper provides an overview of the means and techniques used by governments to monitor, filter, and censor the Internet for their citizens; an overview of the means currently used to circumvent those policies; and a proposed steganography-based technique to protect those who seek content that has been censored or filtered.

Chi-Square Attack Applied To Image-Set Embedding Using LSB Encoding, David Miller.   There is an increasing interest in techniques for the detection of steganographic images on the internet.  Images posted on the internet can be used as a transfer media to move large amounts of hidden data.  In some cases, the movement of hidden data in the internet may be related to activities that are illegal or are considered a threat to national security.  One such technique for detecting steganographic images is the chi-square test based on the LSB message embedding algorithm.  The chi-square test looks for properties of an image commonly occurring when LSB embedding is applied.  This paper considers an embedding algorithm call ‘image-set’ embedding, where a single message is embedded in a set of images.  Four variations of this image-set embedding approach are considered.  For each variation of the image-set embedding technique the effectiveness of the application of the chi-square attack is considered, both on individual images and on the entire image-sets.  Additionally, an attack based on the mean and variance of the weighted color-scaled histograms is considered.  This mean-variance attack is applied to all of the images in all images sets. (Implemented.)

Steganography and Malicious Code:  Past, Present and Future Integrations, H. Duane Long.  This project discusses the history, current uses, and potential threats of various malware and steganographic techniques. A review of recent professional literature regarding these topics is presented. Included are the differences between and similarities of spam, phishing, pharming, worms, viruses, Trojans, zombies, backdoors and keyloggers. Also discussed is the growing concern that steganography may soon be merged with these malicious code attacks creating possibly catastrophic results to current business practices. Internet use for exchange of information may be seriously compromised due to increasing expansion of stego-malware. Little progress is currently reported in the attempts to
prevent, curtail, or produce recovery processes to be applied to these merged attacks. The findings of this paper suggest that future electronic communication may be highly compromised by steganographic intrusions, and research into preventing such a scenario should be made a priority.

DWT (COVER vs. STEGO) Energy Comparison for Common Embedding Software, Enanga Fale.  Researchers are constantly looking in new methods for steganography and steganlaysis for JEPEG images. Throughout various studies and technical reports, the wavelet transform has been presented as one of the many ideal embedding and detection scheme for modified images.  (Implemented.)

Analyzing the Chi-Square Attack on an Singular Value Decomposition Embedding Scheme, Eric Hansen.  The singular value decomposition embedding algorithm by Bergman and Davidson[1] is a unique approach to steganography, but one whose robustness has not been thoroughly tested. Implementing the linear chi-square attack by Westfeld and Pfitzmann[2] on images containing information embedded with this singular value decomposition embedding algorithm yielded two main results. First, chi-square attacking the spatial domain of images with embedded messages did not prove to be informative. Second, chi-square attacking the singular value decomposition matrices of an image with data embedded did correctly identify embedded images and rejected clean images. This would imply that the SVD embedding technique is resistant to a surface chi-square attack but vulnerable to an attack of an adversary who knows what they’re looking for.  (Implementation)

Digimarc Corporation, Gargi Bhattacharyya.   In my project I have tried to give a detailed article on Digimarc Corporation’s Drivers License issuance solutions, discuss issues relating to fraud detection using digimarc’s drivers license solution. I have also discussed the role of the drivers license in homeland security isuues.

Textbased Steganography in Internet Chat and Messaging Services, Gregory P. Probst.  In this paper, we introduce two new methods for performing textbased steganography by combining the mediums of online chat and instant messaging as well as exploiting the grammatical nuances within those mediums for providing text based cover. Steganography has enjoyed a long history and today there are several solutions available for textbased steganography including mimicry, contextbased and contextless steganography. However, while some of these methods apply the use of generally accepted grammatical rule sets, their focus and need to ensure “proper” use of grammar within the text based steganography causes limitations to the extent textbased steganography can be used. A casual observation of communications within the cyber world will illustrate to many that grammatical rules applied to typical daytoday conversation and written correspondence are largely broken and and these violations are accepted violations within many of the online mediums. Because of this greater acceptance of grammar rule violations, the grammar base is largely expanded with the use of lexicon symbol sets and their textual representations, word substitution schemes that results in more freedom within both the contextfree and contextsensitive methods of steganography. This paper will demonstrate how these grammatical evolutions combined with the very transport mediums that exist within the cyber world will yield a very flexible encoding base one can exploit to hide message information.

Term Paper Final Report:
In-band Watermarking of Digital Audio, Karl F. Hoech.  This paper discusses the purpose and challenge associated with watermarking of digital multimedia, including discussion of two principal characteristics of watermark signals. Further, the paper presents significant developments in the field related to development of effective techniques for watermarking of digital audio. The paper then proposes a novel approach using frequency-hopped spread spectrum and adaptive gain control to in-band watermarking of audio for transmission over analog media, including development of requirements for performance of the watermark, the steganographic signal design, modeling and simulation, and discussion of testing and results.

New Techniques in Watermarking Binary Images of Music Scores, Matthew K. Wilden.  This paper contains information on two new watermarking techniques for musical scores. It details the various other methods that are not capable of correctly watermarking these musical scores and goes into detail on why those methods fail and why these new techniques will succeed.
 

Digital Watermarking Technologies – A Practical View, Ryan Smith.  Digital watermarking faces significant challenges to be used effectively and with wide-spread acceptance. To meet these challenges, technical, legal, and infrastructure concerns and issues need to be addressed. Along the quest for general use and acceptance of digital watermarking, there have been small successes and many failures. However, due to the problem of rampant digital content piracy and the monetary values at stake, a viable solution will be reached, sooner of later.

Genetic Programming Approach to Blind Steganalysis, Thad M. Gillispie.   Successful identification of messages hidden in images with steganography typically requires some knowledge of the original image data. Blind steganalysis attempts to accomplish this identification without knowledge of the original image data by using complex pattern recognition and statisical analysis. And in most cases, knowledge of the embedding technique is required for many of the approches to work. In a novel attempt to find a way of doing blind steganalysis without knowledge of the embedding method, genetic programming techniques were applied to a known set of training data using the Simple Computer Language Algorithm (SCLA) evolvable data structure. Use of genetic programming enables one to use the same program for various embedding techniques since the data structures have the ability to evolve toward an optimum solution each time they are implemented. In this investigation, the implementation of the evolvable data structure is handled by an extensive Class in C++ named SCLA_World, which enables the program to easily be ported to a variety of problems. A set of 2,388 jpeg images were analyzed using a features set containing 82 features, all with an associated quantitative value. A series of simulations was run, each simulation consisting of two phases: a Successful identification of messages hidden in images with steganography typically requires some knowledge of the original image data. Blind steganalysis attempts to accomplish this identification without knowledge of the original image data by using complex pattern recognition and statisical analysis. And in most cases, knowledge of the embedding technique is required for many of the approches to work. In a novel attempt to find a way of doing blind steganalysis without knowledge of the embedding method, genetic programming techniques were applied to a known set of training data using the Simple Computer Language Algorithm (SCLA) evolvable data structure. Use of genetic programming enables one to use the same program for various embedding techniques since the data structures have the ability to evolve toward an optimum solution each time they are implemented. In this investigation, the implementation of the evolvable data structure is handled by an extensive Class in C++ named SCLA_World, which enables the program to easily be ported to a variety of problems. A set of 2,388 jpeg images were analyzed using a features set containing 82 features, all with an associated quantitative value. A series of simulations was run, each simulation consisting of two phases: a training period to determine the optimal SCLA structure to solve the problem and the analysis of all 2,388 images by the SCLA with the highest fitness after training. A majority of the simulations were conducted on images embedded with F5. The maximum fitness obtained during the F5 simulations was roughly 51, which correlates to about 80% of embedded images being correctly identified and 30% of clean messages being falsely identified. Although initial results are unfavorable, the genetic programming technique shows enough promise to warrant further pursuit.training period to determine the optimal SCLA structure to solve the problem and the analysis of all 2,388 images by the SCLA with the highest fitness after training. A majority of the simulations were conducted on images embedded with F5. The maximum fitness obtained during the F5 simulations was roughly 51, which correlates to about 80% of embedded images being correctly identified and 30% of clean messages being falsely identified. Although initial results are unfavorable, the genetic programming technique shows enough promise to warrant further pursuit.  

Digital Fingerprinting An Introduction, Thieu Pham.  Abstract—As similar to human fingerprints, digital watermarks can be added to digital objects and serve as fingerprints for these objects. These digital fingerprints then can be used to trace back to those particular objects. With an increasing use of multimedia and the rising of illegal copying and piracy, fingerprinting can be used to trace back to the individuals who are responsible and hold those accountable. Catching the perpetrators help deter other people from committing the same crime. Fingerprinting is a new and effective approach to the prevention of illegal copying and the protection of intellectual properties. This paper introduces and defines digital fingerprinting. This paper focuses on what fingerprinting is and its objectives, what the current solutions and schemes are to overcome the collusion attack, and finally, a successful real-life example of using fingerprinting in catching a traitor.

Investigation of Current Watermarking Technology in Practical Applications.

Secure Watermarking in Very Low Bitrate Speech Codecs, Linfeng Zhang.  In this paper, we discuss the current techniques used in speech watermarking. We focus on how to embed secure watermarks into very low bitrate speech codecs. We propose several candidate watermarking algorithms, and our experimental results show that the quality of speech does not drop very much. We analyze the security level of our algorithms, and some of them can achieve very high confidentiality as a 124-bit key encryption system.

Last Updated February 01, 2007