Catálogo de publicaciones - libros

This paper presents a spread spectrum (SS) watermarking technique that is secure against carriers estimation in a Watermark Only Attack framework. After reviewing the sufficient conditions to design secure algorithms for watermarking and steganography, we present a setup based on Blind Source Separation (BSS) theory to assess the lack of security of classical SS techniques such as classical SS or ISS. We motivate a new SS watermarking algorithm called Natural Watermarking (NW) where the estimation of the secret carriers is impossible and which achieves perfect secrecy thanks to unchanged Gaussian distributions of the secret carriers. The theoretical evaluation of the NW security is carried out and the case of multi-bit embedding is addressed. Finally, a robust extension of NW is presented and the properties of NW and Robust-NW are both practically verified.

In the asymmetric watermarking problem, we wish to embed a signal in a piece of multimedia and later prove that we have done so, but without revealing information that can be used by an adversary to remove the signal later.

There have been several published solutions to this problem, which suffer from the twin problems of size and time complexity. We provide a protocol similar to the one described in [5], but with substantial improvements in time and space. Our protocol is non-interactive, and each exchange of information between the prover and verifier involves a smaller payload of data. The algorithm uses a form of matrix embedding with pseudo-random Gaussian matrices. Aside from an improvement in efficiency, it possesses other practical advantages over previous protocols.

We present a semi-public key implementation of quantization index modulation (QIM) watermarking called Secure QIM (SQIM). Given a signal, a watermark detector can learn the presence of an SQIM watermark without learning anything anything else from the detection process. The watermark detector first transforms the signal with a secret transform, unknown to the detector, and then quantizes the transform coefficients with secret quantizers, also unknown to the detector. This is done with the use of homomorphic cryptosystems, where calculations are performed in an encrypted domain. A low-power, trusted, secure module is used at the end of the process and reveals only if the signal was watermarked or not. Even after repeated watermark detections, no more information is revealed than the watermarked status of the signals. The methods we present are for watermark systems with quantizers of stepsize 2.

In this paper we focus on estimating the amount of information that can be embedded in the sequencing of packets in ordered channels. Ordered channels, e.g. TCP, rely on sequence numbers to recover from packet loss and packet reordering. We propose a formal model for transmitting information by packet-reordering. We present natural and well-motivated channel models and jamming models including the -distance permuter, the -buffer permuter and the -stack permuter. We define the natural information-theoretic (continuous) game between the channel processes (max-min) and the jamming process (min-max) and prove the existence of a Nash equilibrium for the mutual information rate. We study the zero-error (discrete) equivalent and provide error-correcting codes with optimal performance for the distance-bounded model, along with efficient encoding and decoding algorithms. One outcome of our work is that we extend and complete D. H. Lehmer’s attempt to characterize the number of distance bounded permutations by providing the asymptotically optimal bound - this also tightly bounds the first eigen-value of a related state transition matrix [1].

A crucial security practice is the elimination of network covert channels. Recent research in IPv6 discovered that there exist, at least, 22 different covert channels, suggesting the use of advanced active wardens as an appropriate countermeasure. The described covert channels are particularly harmful not only because of their potential to facilitate deployment of other attacks but also because of the increasing adoption of the protocol without a parallel deployment of corrective technology. We present a pioneer implementation of active wardens that eliminates the covert channels exploiting the and the field as well as the well-known Attack. Network-aware active wardens take advantage of network-topology information to detect and defeat covert protocol behavior. We show, by analyzing their performance over a controlled network environment, that the wardens eliminate a significant percentage of the covert channels and exploits with minimal impact over the end-to-end communications (approximately 3% increase in the packet roundtrip time).

Our paper proposes an enhanced video watermarking approach. The fundamental idea is to use geometric warping for watermarks with high predictable robustness to lossy compression. We explain the basic watermarking approach which uses a block based statistic (Normed Centre of Gravity - NCG) to describe the geometric structure of blocks. The NCG also is used to choose robust blocks. To embed the watermark information the chosen blocks are changed by geometric warping. To extract the watermark, the original video is not necessary. The NCG is used to detect the watermarked blocks and compute the embedded watermark bit. In some cases, the independent geometric warping of blocks which contain the same object results in visible artifacts. We propose to link blocks in space and time to block groups. In contrast to the basic approach, the blocks of one block group can be warped in dependence on each other. Thus, the visible artifacts are prevented.

In audio watermarking, the robustness to desynchronization attacks such as TSM (Time-Scale Modification) operations, is still an open issue. In this paper, both mathematical proof and experimental testing show that the histogram shape (represented as the relative relation in the number of samples among three different histogram bins) and the audio mean are two robust features to the TSM attacks. Accordingly, a multi-bit robust audio watermarking algorithm based on the two statistical features is proposed by modifying the histogram. The audio histogram with equal-sized bins is extracted from a selected amplitude range referred to the audio mean, and then the relative relations in the number of samples among groups of three neighboring bins are designed to carry the watermark by reassigning the number of samples in the bins. The watermarked audio signal is perceptibly similar to the original one. Simulation results demonstrated that the hidden message is very robust to the TSM, cropping, and a variety of other distortions in Stirmark Benchmark for Audio.

We introduce content-aware steganography as a new paradigm. As opposed to classic steganographic algorithms that only embed information in the syntactic representation of a datagram, content-aware steganography embeds secrets in the semantic interpretation which a human assigns to a datagram. In this paper, we outline two constructions for content-aware stegosystems, which employ, as a new kind of security primitive, problems that are easy for humans to solve, but difficult to automate. Such problems have been successfully used in the past to construct Human Interactive Proofs (HIPs), protocols capable of automatically distinguishing whether a communication partner is a human or a machine.

We develop the algebraic theory of timed capacity for channels with binary inputs and outputs in the presence of noise, by obtaining a formula for capacity in terms of the unique solution of a nonlinear algebraic equation. We give provably correct numerical algorithms for solving this equation, specifically tailored toward calculating capacity. We use our results to establish that information theory has an inherent discontinuity in it: the function which assigns the unique capacity achieving distribution to the noise matrix of a binary channel has no continuous extension to the set of all noise matrices. Our results provide new formulae in the case of untimed binary channels as well. Our results are important in the study of real-world systems, such as the NRL Network Pump® system and traffic analysis in anonymity systems.

Multimedia security schemes often combine cryptographic schemes with information hiding techniques such as steganography or watermarking. Example applications are dispute resolving, proof of ownership, (asymmetric/anonymous) fingerprinting and zero-knowledge watermark detection. The need for formal security definitions of watermarking schemes is manifold, whereby the core need is to provide suitable abstractions to construct, analyse and prove the security of applications on top of watermarking schemes. Although there exist formal models and definitions for information-theoretic and computational security of cryptographic and steganographic schemes, they cannot simply be adapted to watermarking schemes due to the fundamental differences among these approaches. Moreover, the existing formal definitions for watermark security still suffer from conceptual deficiencies.

In this paper we make the first essential steps towards an appropriate formal definition of watermark robustness, the core security property of watermarking schemes: We point out and discuss the shortcomings of the existing proposals and present a formal framework and corresponding definitions that cover those subtle aspects not considered in the existing literature. Our definitions provide suitable abstractions that are compatible to cryptographic definitions allowing security proofs of composed schemes.