A digital watermark can be seen as an information channel, which is hidden in a cover signal. It is usually designed to be imperceptible to human observers. Although imperceptibility is often achieved, the inherent modification of the cover signal may be viewed as a potential disadvantage. In this paper, we present a reversible watermarking technique for digital audio signals. In our context reversibility refers to the ability to restore the original input signal in the watermark detector. In summary, the approach works as follows. In the encoder, the dynamic range of the input signal is limited (i.e. it is compressed), and part of the unused bits is deployed for encoding the watermark bits. Another part of these bits is used to convey information for the bit-exact reconstruction of the cover signal. It is the purpose of the watermark detector to extract the watermark and reconstruct the input signal by restoring the original dynamic range. In this study we extensively tested this new algorithm with a variety of settings using audio items with different characteristics. These experiments showed that for 16bit PCM audio, capacities close to 1-bit per sample can be achieved, while perceptual degradation of the watermarked signal remained acceptable.

Distribution via Internet is of increasing importance. Easy access, transmission and consumption of digitally represented music is very attractive to the consumer but led also directly to an increasing problem of illegal copying. To cope with this problem watermarking is a promising concept since it provides a useful mechanism to track illicit copies by persistently attaching property rights information to the material. Especially for online music distribution the use of so-called transaction watermarking, also denoted with the term bitstream watermarking, is beneficial since it offers the opportunity to embed watermarks directly into perceptually encoded material without the need of full decompression/compression. Besides the concept of bitstream watermarking, former publications presented the complexity, the audio quality and the detection performance. These results are now extended by an assessment of the robustness of such schemes. The detection performance before and after applying selected attacks is presented for MPEG-1/2 Layer 3 (MP3) and MPEG-2/4 AAC bitstream watermarking, contrasted to the performance of PCM spread spectrum watermarking.

In recent years, the spread spectrum watermarking technology has become the most promising technique that not only widely used for still image and video watermarking, but also used for audio watermarking. However, some technique problems such as requiring psycho-acoustic shaping used for reducing audible noise have greatly limited the utility of spread spectrum watermarking technology in audio watermarking. In this paper, we propose a novel audio watermarking method using spread spectrum watermarking technology by which we can embed watermark audio signals inaudibly with a robust to a wide range of unintended and intended attacks. In proposed method, the watermark is represented by sinusoidal patterns consisting of sinusoids with the phase-modulated by the elements of pseudo-random sequence. We theoretically and experimentally confirmed that the sinusoidal patterns based on pseudo-random sequences keep the same correlation property of pseudo-random sequences and have the characteristics of high robustness with less noise, being easy to manipulate, and without requirement of psycho-acoustic shaping. The watermark detection is done by blind detection and the effectiveness of proposed method have been certificated by the test of STEP2001.

Audio watermarking has been used mainly for digitally stored content. Using real-time watermark embedding, its coverage can be extended to live broadcasts and live performances. In general, a conventional embedding algorithm receives a host signal (HS) and outputs the summation of the HS and a watermark signal (WS). However, when applied to real-time embedding, there are two problems: (1) delay of the HS, and (2) possible interruption of the broadcast. To solve these problems, we propose a watermark generation algorithm that outputs only a WS, and a system composition method in which a mixer outside the computer mixes the WS generated by the algorithm and the HS. In addition, we propose a new composition method "sonic watermarking." In this composition method, the sound of the HS and the sound of the WS are played separately by two speakers, and the sounds are mixed in the air. Using this composition method, it would be possible to generate a watermarking sound in a concerto hall so that the watermark could be detected from content recorded by audience members who have recording devices at their seats. We report on the results of experiments and discuss the merits and flaws of various real-time watermarking composition methods.

Today a number of audio watermarking algorithms have been proposed, some of them at a quality making them suitable for commercial applications. The focus of most of these algorithms is copyright protection. Therefore, transparency and robustness are the most discussed and optimised parameters. But other applications for audio watermarking can also be identified stressing other parameters like complexity or payload. In our paper, we introduce a new mp2 audio watermarking algorithm optimised for high payload. Our algorithm uses the scale factors of an mp2 file for watermark embedding. They are grouped and masked based on a pseudo-random pattern generated from a secret key. In each group, we embed one bit. Depending on the bit to embed, we change the scale factors by adding 1 where necessary until it includes either more even or uneven scale factors. An uneven group has a 1 embedded, an even group a 0. The same rule is later applied to detect the watermark. The group size can be increased or decreased for transparency/payload trade-off. We embed 160 bits or more in an mp2 file per second without reducing perceived quality. As an application example, we introduce a prototypic Karaoke system displaying song lyrics embedded as a watermark.

Classification of audio documents as bearing hidden information or not is a security issue addressed in the context of steganalysis. A cover audio object can be converted into a stego-audio object via steganographic methods. In this study we present a statistical method to detect the presence of hidden messages in audio signals. The basic idea is that, the distribution of various statistical distance measures, calculated on cover audio signals and on stego-audio signals vis-à-vis their denoised versions, are statistically different. The design of audio steganalyzer relies on the choice of these audio quality measures and the construction of a two-class classifier. Experimental results show that the proposed technique can be used to detect the presence of hidden messages in digital audio data.

This paper presents the results of embedding short covert message utterances on a host, or cover, utterance by modifying the phase or amplitude of perceptually masked or significant regions of the host. In the first method, the absolute phase at selected, perceptually masked frequency indices was changed to fixed, covert data-dependent values. Embedded bits were retrieved at the receiver from the phase at the selected frequency indices. Tests on embedding a GSM-coded covert utterance on clean and noisy host utterances showed no noticeable difference in the stego compared to the hosts in speech quality or spectrogram. A bit error rate of 2 out of 2800 was observed for a clean host utterance while no error occurred for a noisy host. In the second method, the absolute phase of 10 or fewer perceptually significant points in the host was set in accordance with covert data. This resulted in a stego with successful data retrieval and a slightly noticeable degradation in speech quality. Modifying the amplitude of perceptually significant points caused perceptible differences in the stego even with small changes of amplitude made at five points per frame. Finally, the stego obtained by altering the amplitude at perceptually masked points showed barely noticeable differences and excellent data recovery.

In this paper, the influence of the estimated perceptual mask from the watermarked image is analyzed with the purpose of decreasing the probability of error in detection. We concentrate on a well known perceptual model that has been applied to 8x8 block-wise DCT coefficients. A new procedure based on extracting the mask of the original image from the watermarked image is developed. A closed form solution is derived to estimate the coefficients of the original image. The mask in then computed from these coefficients. The model can also be recursively estimated. Results show that the proposed solution outperforms those solutions based on not considering at all the perceptual mask in detection or on applying the same procedure that was used when embedding the watermark but using the watermarked image instead of the original image.

One of the most active research area in the watermarking community is the research in dealing with geometric distortion. The geometric distortion problem has two aspects, namely its effect on watermark detectability and its effect on the perceptual quality of the watermarked data. Most research in this area has been concentrated on addressing the first aspect of the problem, and research on objective visual quality assessment of geometrically distorted images is not widely discussed in the literature. As a consequence, there is a lack of objective visual quality measurement for this class of distortion. In this paper we propose a method of objectively assessing the perceptual quality of geometrically distorted images. Our approach is based on the modeling of a complex, global geometric distortion using local, simpler geometric transformation models. The locality of this simpler geometric transformation determines the visual quality of the distorted images.

An image watermark parameter optimization procedure is proposed for selecting the most effective DCT coefficients for watermark embedding. Using this set of coefficients improves the watermark robustness and reliability against attack while it maintains the transparency of the embedded watermark. With the aid of prior knowledge of attacks, the visual masking effect and the attack distortion on each (DCT) transform coefficient are pre-calculated so that a maximum strength watermark within visual threshold can be inserted. There are two stages in the design phase. First, taking into account the combined effect of watermark embedding and attack, we pick up the robust coefficients that resist a specific type of attacks and in the meanwhile we keep the distortion lower than the visual threshold. Although typically the watermark detection reliability increases with the increasing number of embedded coefficients, the less effective coefficients may degrade the overall detection performance. Thus, in the second stage, some initially selected coefficients are discarded by an iterative process to reduce the overall error detection probability. Since digital images are often compressed for efficient storage and transmission, we adopt JPEG compression as the attacking source. The simulation results show that the detection error probability is significantly reduced when the selected robust coefficients are in use. These coefficients with watermark embedded on them can also survive color reduction, Gaussian filtering, and frequency mode Laplacian removal (FMLR) attacks.

This paper presents a new approach in the selection of auspicious sites to be watermarked. The selection takes into account human visual system properties including luminance adaptation, contrast sensitivity and spatio-frequential selectivity. The paper exploits also the local band limited contrast to determine the maximum watermark strength to be applied without inducing visible degradations. Compared to the well known approaches, a contrast masking model is used here to adjust, site by site, the watermark strength. To test the approach efficiency, obtained results are considered in the context of an adaptive watermarking algorithm. The performance of this latter is evaluated in terms of watermark invisibility. Robustness to most common attacks such as JPEG compression, cropping (with zero padding) and low pass filtering is also considered.

Most watermarking applications require that the embedded watermark be imperceptible. Accordingly, perceptual masking models that identify unperceived regions of the signal were adapted in a straightforward manner to watermarking. The derived mask -- or slack -- is often interpreted as the maximal allowed distortion within a given region of the signal; it is used in many watermarking embedding methods to shape a white spectrum message, in the relevant transform domain (space, frequency). Such a usage of the mask is intuitively satisfying since imperceptibility is indeed guaranteed; yet, it discards any guarantee of robustness to attacks -- another fundamental, necessary property of watermarks. The trade-off between fidelity and robustness has been little addressed so far due in great part to the absence of an accurate measure of perceptual distortion. In this paper we study this trade-off using Watson's measure of perceptual distance between two images as the measure of fidelity. Based on a constrained perceptual distance, the embedder must maximize the watermark's robustness while assuming a knowledgeable attacker will attempt to remove the watermark. Solving this problem leads to an optimized watermark strength for each location of the content.

Steganalysis of sequential steganography is presented in this paper. Abrupt change in statistics due to sequential embedding is exploited by the proposed technique. Analytical derivations are presented for several cases along with experimental results. Experiments show that the proposed method can be used quite effectively to detect locations and length of messages embedded using spread spectrum steganography.

The process of information hiding is modeled in the context of additive noise. Under an independence assumption, the histogram of the stegomessage is a convolution of the noise probability mass function (PMF) and the original histogram. In the frequency domain this convolution is viewed as a multiplication of the histogram characteristic function (HCF) and the noise characteristic function. Least significant bit, spread spectrum, and DCT hiding methods for images are analyzed in this framework. It is shown that these embedding methods are equivalent to a lowpass filtering of histograms that is quantified by a decrease in the HCF center of mass (COM). These decreases are exploited in a known scheme detection to classify unaltered and spread spectrum images using a bivariate classifier. Finally, a blind detection scheme is built that uses only statistics from unaltered images. By calculating the Mahalanobis distance from a test COM to the training distribution, a threshold is used to identify steganographic images. At an embedding rate of 1 b.p.p. greater than 95% of the stegoimages are detected with false alarm rate of 5%.

In this paper, we present general methodology for developing attacks on steganographic systems for the JPEG image format. The detection first starts by decompressing the JPEG stego image, geometrically distorting it (e.g., by cropping), and recompressing. Because the geometrical distortion breaks the quantized structure of DCT coefficients during recompression, the distorted/recompressed image will have many macroscopic statistics approximately equal to those of the cover image. We choose such macroscopic statistic S that also predictably changes with the embedded message length. By doing so, we estimate the unknown message length by comparing the values of S for the stego image and the cropped/recompressed stego image. The details of this detection methodology are explained on the F5 algorithm and OutGuess. The accuracy of the message length estimate is demonstrated on test images for both algorithms. Finally, we identify two limitations of the proposed approach and show how they can be overcome to obtain accurate detection in every case. The paper is closed with outlining a condition that must be satisfied by all secure high-capacity steganographic algorithms for JPEGs.

Universal blind steganalysis attempts to detect steganographic data without knowledge about the applied steganographic system. Farid proposed such a detection algorithm based on higher-order statistics for separating original images from stego images. His method shows an astonishing performance on current steganographic schemes. Starting from the statistical approach in Farid's algorithm, we investigate the well known steganographic tool Jsteg as well as a newer approach proposed by Eggers et al., which relies on histogram-preserving data mapping. Both schemes show weaknesses leading to a certain detectability. Further analysis shows which statistic characteristics make both schemes vulnerable. Based on these results, the histogram preserving approach is enhanced such that it achieves perfect security with respect to Farid's algorithm.

There are two primary challenges to monitoring the Web for steganographic media: finding suspect media and examining those found. The challenge that has received a great deal of attention is the second of these, the steganalysis problem. The other challenge, and one that has received much less attention, is the search problem. How does the steganalyzer get the suspect media in the first place? This paper describes an innovative method and architecture to address this search problem. The typical approaches to searching the web for covert communications are often based on the concept of “crawling” the Web via a smart “spider.” Such spiders find new pages by following ever-expanding chains of links from one page to many next pages. Rather than seek pages by chasing links from other pages, we find candidate pages by identifying requests to access pages. To do this we monitor traffic on Internet backbones, identify and log HTTP requests, and use this information to guide our process. Our approach has the advantages that we examine pages to which no links exist, we examine pages as soon as they are requested, and we concentrate resources only on active pages, rather than examining pages that are never viewed.

Two fundamental questions in steganography are addressed in this paper, namely, (a) definition of steganography security and (b) definition of steganographic capacity. Since the main goal of steganography is covert communications, we argue that these definitions must be dependent on the type of steganalysis detector employed to break the embedding algorithm. We propose new definitions for security and capacity in the presence of a steganalyst. The intuition and mathematical notions supporting these definitions are described. Some numerical examples are also presented to illustrate the need for this investigation.

In this paper, we describe a new higher-order steganalytic method called Pairs Analysis for detection of secret messages embedded in digital images. Although the approach is in principle applicable to many different steganographic methods as well as image formats, it is ideally suited to 8-bit images, such as GIF images, where message bits are embedded in LSBs of indices to an ordered palette. The EzStego algorithm with random message spread and optimized palette order is used as an embedding archetype on which we demonstrate Pairs Analysis and compare its performance with the chi-square attacks and our previously proposed RS steganalysis. Pairs Analysis enables more reliable and accurate message detection than previous methods. The method was tested on databases of GIF images of natural scenes, cartoons, and computer-generated images. The experiments indicate that the relative steganographic capacity of the EzStego algorithm with random message spread is less than 10% of the total image capacity (0.1 bits per pixel).

In this paper, we present a new steganographic paradigm for digital images in raster formats. Message bits are embedded in the cover image by adding a weak noise signal with a specified but arbitrary probabilistic distribution. This embedding mechanism provides the user with the flexibility to mask the embedding distortion as noise generated by a particular image acquisition device. This type of embedding will lead to more secure schemes because now the attacker must distinguish statistical anomalies that might be created by the embedding process from those introduced during the image acquisition itself. Unlike previously proposed schemes, this new approach, that we call stochastic modulation, achieves oblivious data transfer without using noise extraction algorithms or error correction. This leads to higher capacity (up to 0.8 bits per pixel) and a convenient and simple implementation with low embedding and extraction complexity. But most importantly, because the embedding noise can have arbitrary properties that approximate a given device noise, the new method offers better security than existing methods. At the end of this paper, we extend stochastic modulation to a content-dependent device noise and we also discuss possible attacks on this scheme based on the most recent advances in steganalysis.

With the growing commercialization of watermarking techniques in various application scenarios it has become increasingly important to quantify the performance of watermarking products. The quantification of relative merits of various products is not only essential in enabling further adoption of the technology by society as a whole, but will also drive the industry to develop testing plans/methodologies to ensure quality and minimize cost (to both vendors & customers.) While the research community understands the theoretical need for a publicly available benchmarking system to quantify performance, there has been less discussion on the practical application of these systems. By providing a standard set of acceptance criteria, benchmarking systems can dramatically increase the quality of a particular watermarking solution, validating the product performances if they are used efficiently and frequently during the design process. In this paper we describe how to leverage specific design of experiments techniques to increase the quality of a watermarking scheme, to be used with the benchmark tools being developed by the Ad-Hoc Watermark Verification Group. A Taguchi Loss Function is proposed for an application and orthogonal arrays used to isolate optimal levels for a multi-factor experimental situation. Finally, the results are generalized to a population of cover works and validated through an exhaustive test.

In 1998, Paul Kocher showed that when a smart card computes cryptographic algorithms, for signatures or encryption, its consumption or its radiations leak information. The keys or the secrets hidden in the card can then be recovered using a differential measurement based on the intercorrelation function. A lot of silicon manufacturers use desynchronization countermeasures to defeat power analysis. In this article we detail a new resynchronization technic. This method can be used to facilitate the use of a neural network to do the code recognition. It becomes possible to reverse engineer a software code automatically. Using data and clock separation methods, we show how to optimize the synchronization using signal processing. Then we compare these methods with watermarking methods for 1D and 2D signal. The very last watermarking detection improvements can be applied to signal processing for smart cards with very few modifications. Bayesian processing is one of the best ways to do Differential Power Analysis, and it is possible to extract a PIN code from a smart card in very few samples. So this article shows the need to continue to set up effective countermeasures for cryptographic processors. Although the idea to use advanced signal processing operators has been commonly known for a long time, no publication explains that results can be obtained. The main idea of differential measurement is to use the cross-correlation of two random variables and to repeat consumption measurements on the processor to be analyzed. We use two processors clocked at the same external frequency and computing the same data. The applications of our design are numerous. Two measurements provide the inputs of a central operator. With the most accurate operator we can improve the signal noise ratio, re-synchronize the acquisition clock with the internal one, or remove jitter. The analysis based on consumption or electromagnetic measurements can be improved using our structure. At first sight the same results can be obtained with only one smart card, but this idea is not completely true because the statistical properties of the signal are not the same. As the two smart cards are submitted to the same external noise during the measurement, it is more easy to reduce the influence of perturbations. This paper shows the importance of accurate countermeasures against differential analysis.

Our paper addresses two issues of a biometric authentication algorithm for ID cardholders previously presented namely the security of the embedded reference data and the aging process of the biometric data. We describe a protocol that allows two levels of verification, combining a biometric hash technique based on handwritten signature and hologram watermarks with cryptographic signatures in a verification infrastructure. This infrastructure consists of a Trusted Central Public Authority (TCPA), which serves numerous Enrollment Stations (ES) in a secure environment. Each individual performs an enrollment at an ES, which provides the TCPA with the full biometric reference data and a document hash. The TCPA then calculates the authentication record (AR) with the biometric hash, a validity timestamp, and a document hash provided by the ES. The AR is then signed with a cryptographic signature function, initialized with the TCPA's private key and embedded in the ID card as a watermark. Authentication is performed at Verification Stations (VS), where the ID card will be scanned and the signed AR is retrieved from the watermark. Due to the timestamp mechanism and a two level biometric verification technique based on offline and online features, the AR can deal with the aging process of the biometric feature by forcing a re-enrollment of the user after expiry, making use of the ES infrastructure. We describe some attack scenarios and we illustrate the watermarking embedding, retrieval and dispute protocols, analyzing their requisites, advantages and disadvantages in relation to security requirements.

In this paper, a new ID card system based-on the digital information hiding and optical extraction technique is suggested and implemented. In the information-hiding process, stego keys are digitally generated by using 2-D PRS and used to hide some specific data related to the card holders into their digital photo-image (cover image) in order to verify the ID cards. To extract the hidden data from the photo-image (stego image) in real-time, a matched filter-based optical correlator system is introduced. Optical correlation between the input stego image and the stego keys in the matched filter bank makes the sharp correlation peaks in the correlation plane only if the stego key hidden in the stego image is identically same with the stego key stored in the matched filter bank. This correlation peak signals and its position data can be used for verification of the ID card. In the experiment, a cover image of “Lena” is divided into 10 x 7 blocks and then, the stego key consisting of 64 x 64 pixels selected from 50 valid stego keys is embedded into the 3 different blocks of the cover image. Through the optical correlation between the “Lena” stego image and its valid stego key, resultant three correlation outputs and their position data are finally used for authentication. From these experimental results, a possibility of implementing a new ID card verification system using opto-digital information hiding and extraction scheme is suggested.

Biometric authentication systems require a fast and accurate method of matching biometric data for identification purposes. This paper introduces a data reduction technique based on image processing to better embed biometric data in small images. For the most part, biometric data cannot be directly embedded in small images, because of limited embedding capacities and a large amount of data in biometric images. An image processing technique to extract features from biometric data, like fingerprints and retinal scans, has been developed and tested. This new technique developed to extract features is based on the Hough transform and has been tested on a large volume of real image data. The data reduction technique was applied to these images and the data reduced to size, which could be easily embedded in small pictures, like those on identity cards. Existing embedding algorithms were utilized.

Earlier, we presented a highly robust invisible watermarking method having a payload of one bit -- indicating the presence or absence of the watermark. Other invisible watermarking methods also possess this property. Our method, by design, places a major portion of the scant energy of the watermark into the low spatial frequencies of the watermarked image while leaving the higher spatial frequencies, contributed by the original image, largely unaltered. In this paper we will show that application of blurring filters to a presumed watermarked image before attempting watermark detection can improve the probability of detection. Blurring filters, because they suppress high spatial frequencies, are generally quite destructive of image quality. However, for a watermark which has dominant low frequency content, the application of a blurring filter can serve to improve the statistical environment for watermark detection and thereby improves detection probability. This is especially true for detections attempted from scans of printed images. The halftone screening process used in preparing images for printing adds significant high frequency noise to the printed image, and that noise is detrimental to watermark detection. Application of a rudimental blurring filter to a watermarked image before detection will be demonstrated showing that increased amounts of blurring improve the probability of watermark detection. However, the amount of blurring can not be increased without limit, and the amount of beneficial blurring will be shown to be related to the amount of low-pass filtering used in constructing the watermark.

Nowadays, a popular method used for additive watermarking is wide spread spectrum. It consists in adding a spread signal into the host document. This signal is obtained by the sum of a set of carrier vectors, which are modulated by the bits to be embedded. To extract these embedded bits, weighted correlations between the watermarked document and the carriers are computed. Unfortunately, even without any attack, the obtained set of bits can be corrupted due to the interference with the host signal (host interference) and also due to the interference with the others carriers (inter-symbols interference (ISI) due to the non-orthogonality of the carriers). Some recent watermarking algorithms deal with host interference using side informed methods, but inter-symbols interference problem is still open. In this paper, we deal with interference cancellation methods, and we propose to consider ISI as side information and to integrate it into the host signal. This leads to a great improvement of extraction performance in term of signal-to-noise ratio and/or watermark robustness.

In this work we investigate the performance of multiplicative spread spectrum (SS) watermarking. In order to make the analysis mathematically tractable, we make some simplifying assumptions, in that we assume that the host features are normally distributed and that attacks take the form of additive white Gaussian noise. After deriving the optimum decoder for the considered scenario, the performance of the system are derived in terms of achievable bit error rate. The results are compared to those achievable through additive spread spectrum watermarking and a classical quantization based scheme. The overall conclusion of the paper is that while keeping the general behavior of SS systems, the multiplicative SS scheme outperforms its additive counterpart for low to moderate attacks.

In this work we propose a new domain for digital watermarking. The proposed domain is the joint time-frequency cells of the Wigner distribution of the image. This domain provides an expanded, richer environment for watermark embedding. Watermark is placed in selected time-frequency cells of Wigner distribution. An inverse mapping then generates the watermarked signal. It is recognized that this approach is similar to time-frequency filtering problem. Since the watermarked Wigner may no longer be an admissible distribution any longer, the algorithm estimates a signal whose Wigner distribution is closest to the desired watermarked distribution. Selecting watermarkable time-frequency cells is similar, but not the same, problem encountered in DCT watermarking. For watermarking that is robust to JPEG, only those cells whose time-frequency signatures are relatively untouched by compression are selected. As a result, watermark remains detectable across all JPEG quality factors.

In spread-spectrum watermarking, the watermarked document is obtained from the addition of an attenuated watermark signal to a cover multimedia document. A traditional strategy consists of optimizing the detector for a given embedding function. In general, this leads to sub-optimal detection and much improvement can be obtained by exploiting side-information available at the embedder. In some prior art, the authors showed that for blind detection of small signals, maximum detection power is obtained to first order by setting the watermark signal to the gradient of the detector. Recently, Malvar et al. improved the performance of direct-sequence spread-spectrum watermarking by using a signal dependent modulation. In the first part of the paper, we develop this idea further and extend Costa's decoding theory to the problem of watermarking detection. In the second part, we propose a practical implementation of this work using non-linear detectors based on our family of polynomial functions. We show some improved performance of the technique.

Watermark detection plays a crucial role in multimedia copyright protection and has traditionally been tackled using correlation-based algorithms. However, correlation-based detection is not actually the best choice, as it does not utilize the distributional characteristics of the image being marked. Accordingly, an efficient watermark detection scheme for DWT coefficients is proposed as optimal for non-additive schemes. Based on the statistical decision theory, the proposed method is derived according to Bayes’ decision theory, the Neyman-Pearson criterion, and the distribution of the DWT coefficients, thereby minimizing the missed detection probability subject to a given false alarm probability. The proposed method was tested in the context of robustness, and the results confirmed the superiority of the proposed technique over conventional correlation-based detection method.

In this paper, we investigate the improvement achieved by turbo coding for robustness in still image watermarking. We use an error correcting scheme based on the concatenation of a BCH product code and a repetition code. The product code is iteratively decoded using the Chase-Pyndiah algorithm (turbo decoded). For this study, we set the watermarking distortion to around 38 dB and we consider different payloads that can correspond to different applications and services. We compare different coding strategies (i.e. repetition code only and a concatenation of product and repetition codes) in terms of robustness to different photometric attacks, in particular additive noise or lossy compression. Typically, for a payload of 121 bits, the robustness gain for a given message error probability is significant: first error appears at a JPEG quality factor of about 25% with the new coding scheme instead of about 50% when using only repetition codes.

Due to the inherent combination of image, video, audio and 3D-models in new MPEG standards like MPEG-4, robust 3D-watermarking is becoming more and more popular to ensure data authenticity and integrity. Beside the design of 3D-watermarking schemes, the evaluation is necessary. In the paper we analyze currently proposed 3D watermarking algorithms for weaknesses that may possibly be exploited in watermarking attacks and derive according design principles for improving algorithms. The paper makes contributions in three respects: First we analyze properties with respect to preservation of surface continuity and symmetries. Second, we analyze how algorithms proposed by Ohbuchi et al. and Praun et al. compensate for errors introduced through imperfect mesh resampling. For Ohbuchi et al. algorithm, we explain how spectral decomposition can be applied in directions other than canonical coordinate system axes in order to minimize errors introduced to resampling. With an experimental prototyped algorithm based on spectral decomposition, we demonstrate a significant increase of robustness of features with respect to errors introduced in resampling. Third, we analyzed the general effects caused by polygon simplification on watermarking algorithms, which, from the results from Praun et al. can be considered as a “critical” operation. As a first result of our analysis, we present a modification of Praun et al. watermarking scheme detector trying to improve compensation for these effects.

In this paper, we propose an improved autocorrelation function (ACF) based watermarking. The autocorrelation function based watermarking has been known to be one of the effective watermarking schemes that are resilient to geometric transform attacks. In this watermarking scheme, both the peaks in the autocorrelation function of a marked image and the embedded watermark should survive after various attacks. Generally, the autocorrelation peaks are less robust than the actual watermark signal. The main focus of this paper is the improvement of the robustness of the peaks in an autocorrelation function. Since an original image and the detector structure are available at the embedding time, instead of simple addition of a watermark to an image, the watermark embedder analyzes the original image and uses this information actively during the embedding process so that the marked image has higher periodic autocorrelation peaks. The proposed watermarking scheme has higher robustness than a conventional ACF based watermarking against geometric attacks that are combined with removal attacks.

This paper deals with the issue of watermarking 2D-vector data (point coordinates) which are used in Geographical Information Systems (GIS). The tolerance range of the coordinates is used to embed the information. As an embedding constraint no violation of the tolerance range at each point may occur. In the first step of our approach we lay a rectangular grid over the map, where the origin, the extension, and the subdivision of this grid is only known by the embedder. Based on a secret key we choose two disjunctive sets A and B of grid elements namely “patches.” Now these selected elements are further divided into smaller subareas or “subpatches.” In subpatches belonging to set A the distances of points to a reference line within each subpatch are decreased while the distances of subpatches of set B are unchanged. In the retrieving procedure we consider all subpatches belonging to the sets A and B and calculate for each of the two sets their sample variance. The decision whether the map is watermarked with a certain key or not is based on this feature and is made with the help of a hypothesis test. The proposed method is robust against attacks changing the coordinates within the tolerance range and against polygon simplification procedures.

Digital watermarking systems are required to embed as much information as possible in a digital media without the perceptual distortion as well as to extract it correctly with high probabilities even though the media is subjected to many kinds of operations. To this end, guided scrambling (GS) techniques, usually used for a recording channel, are applied to digital watermarking systems. A simple GS scheme can make the power of a watermark signal larger against the power of media noise under the condition of preserving the perceptual fidelity, resulting in smaller error probabilities of the retrieved watermark bits. In addition, watermarking systems based on the GS can have more robustness to some specified operations if the prior information on the operations is given to the embedder. JPEG compression is a good example of such an operation when still images are transmitted over the Internet. In order for watermark signals to be more tolerable to the known JPEG attack, the GS-based watermark embedder having advance knowledge of the JPEG compression is realized. Further, the GS concatenated with turbo coding is introduced to lower the bit error rate more.

Current research on data hiding is more and more demonstrating that many applications can benefit from these technologies: among these, medical data management. Current medical record formats store in separated fields image data and the textual information, so that the link between image and patient occasionally could get mangled by protocol converters or tampering attacks. Moreover, if an intruder can access to the database, he is able to modify the attached text. Embedding patient's information directly into the image through data hiding technology can represent an useful safety measure. Data hiding technologies suitable for such an application must satisfy specific requirements, the most important are: a high payload reliably identifying a patient; the preservation of the quality of the host medical image, the robustness to content modification. According to this analysis, a comparison between different data hiding approaches will be presented, to evaluate the most suitable algorithms for medical applications. In particular two different kind of algorithms will be taken into account: one algorithm based on Bayes theory will be compared with algorithms following the new approach of modelling data hiding as communication with side-information at the transmitter. These methods will be tested and compared in the framework of medical data management in order to identify benefits and drawbacks of both the different approaches for such an application.

Wireless sensor networks have emerged as the major criteria that enable the next scientific, technological, engineering, and economic revolution. Since digital rights management is of the crucial importance for sensor networks, there is an urgent need for development of intellectual property protection (IPP) techniques. We have developed the first system of watermarking techniques to embed cryptologically encoded authorship signatures into data and information acquired by wireless embedded sensor networks. The key idea is to impose additional constraints during the data acquisition or sensor data processing. Constraints correspond to the encrypted signature and are selected in such a way that they provide favorable tradeoffs between the accuracy and the strength of proof of the authorship. The techniques for watermarking raw sensor data include one that modifies the location and orientation of a sensor, time management discipline (e.g. frequency and phase of intervals between consecutive data capturing), and its resolution. The second set of techniques embeds signature during data processing. There are at least three degrees of freedom that can be exploited: error minimization procedures, physical world model building, and solving of computationally intractable problems. We have developed several watermarking techniques that leverage on the error minimization degree of freedom and have demonstrated their effectiveness for watermarking location discovery information.

In this paper we present a new method for hiding data in music scores. In contrast to previous published algorithms we investigate the possibilities of embedding information in clefs. Using the clef as information carrier has two advantages: First, a clef is present in each staff line which guarantees a fixed capacity. Second, the clef defines the reference system for musical symbols and music containing symbols, e.g. the notes and the rests, are not degraded by manipulations. Music scores must be robust against greyscale to binary conversion. As a consequence, the information is embedded by modifying the black and white distribution of pixels in certain areas. We evaluate simple image processing mechanisms based on erosion and dilation for embedding the information. For retrieving the watermark the b/w-distribution is extracted from the given clef. To solve the synchronization problem the watermarked clef is normalized in a pre-processing step. The normalization is based on moments. The areas used for watermarking are calculated by image segmentation techniques which consider the features of a clef. We analyze capacity and robustness of the proposed method using different parameters for our proposed method. This proposed method can be combined with other music score watermarking methods to increase the capacity of existing watermarking techniques.

A novel technique allowing secure transmission/storage of electronic documents in printed form is described. First, given a document to protect, an error resilient "visibly encrypted" version is printed. Later, when the original document is to be recovered, the system scans the "visibly encrypted" document and decrypts it after asking for a secret key. Unfortunately, one faces the problem that when a document is printed and scanned, the rescanned document may look similar to the original, but will be distorted during the process. Therefore, to ensure reliable and high rate transmission over the print-and-scan channel it is essential a judicious theoretical model for characterizing the problem and providing reliable communications schemes. The proposed method is based on Pulse Amplitude Modulation (PAM), using small square-shaped pulses and a Maximum Likelihood (ML) detector that is derived after estimating the distortions introduced by the print-and-scan channel. Furthermore, it is essential to employ synchronization techniques to correctly demodulate the printed pulses. In our case, we use an adaptive scheme that resembles the well-known phase locked loops (PLL's). Finally, we will discuss schemes that can make the bit stream resilient to transmissions errors and how to combine them with cryptographic algorithms in order to produce a secure system.

The problem of embedding a binary image into another binary image is considered. We propose a method where the embedding is performed spatially, i.e., a smaller sized binary image (Image A) is embedded in a larger sized binary image (Image B). Image B is the only visible image when the composite is viewed directly. To view Image A, we can perform a suitable extraction operation on Image B. The embedding algorithm can be applied recursively so that the final binary image can contain a number of progressively smaller sized binary images. Examples are used to show the quality of the embedding. Admissible conditions and security issues of the embedding algorithm are considered.

Watermarking of printed materials has usually focused on process inks of cyan, magenta, yellow and black (CMYK). In packaging, almost three out of four printed materials include spot colors. Spot colors are special premixed inks, which can be produced in a vibrant range of colors, often outside the CMYK color gamut. In embedding a watermark into printed material, a common approach is to modify the luminance value of each pixel in the image. In the case of process color work pieces, the luminance change can be scaled to the C, M, Y and K channels using a weighting function, to produce the desired change in luminance. In the case of spot color art designs, there is only one channel available and the luminance change is applied to this channel. In this paper we develop a weighting function to embed the watermark signal across the range of different spot colors. This weighting function normalizes visibility effect and signal robustness across a wide range of different spot colors. It normalizes the signal robustness level over the range of an individual spot color’s intensity levels. Further, it takes into account the sensitivity of the capturing device to the different spot colors.

A novel watermarking algorithm for watermarking low bit-rate MPEG-4 compressed video is developed and evaluated in this paper. Spatial spread spectrum is used to invisibly embed the watermark into the host video. A master synchronization template is also used to combat geometrical distortion such as cropping, scaling, and rotation. The same master synchronization template is used for watermarking all video objects (VOP) in the bit-stream, but each object can be watermarked with a unique payload. A gain control algorithm is used to adjust the local gain of the watermark, in order to maximize watermark robustness and minimize the impact on the quality of the video. A spatial and temporal drift compensator is used to eliminate watermark self-interference and the drift in quality due to AC/DC prediction in I-VOPs and motion compensation in P- and B-VOPs, respectively. Finally, a bit-rate controller is used to maintain the data-rate at an acceptable level after embedding the watermark. The developed watermarking algorithm is tested using several bit-streams at bit-rates ranging from 128-750 Kbit/s. The visibility and the robustness of the watermark after decompression, rotation, scaling, sharpening, noise reduction, and trans-coding are evaluated.

MPEG-4 is an international object-based standard that provides technological basis for digital television, interactive graphics and multimedia applications. These objects can be natural or synthetic e.g. textures, 3D objects, videos or sounds. In this paper we suggest an integrity approach to protect the content of MPEG-4 data. The essential part of this approach is to embed a robust watermark into each visual, audio and 3D object. The content fragile watermark verifying the integrity of a scene is the sum of all information retrieved from the robust watermarks extracted from the objects of the scene. The information of the fragile watermark will be distributed redundantly to all robust watermarks of the scene. Another essential part of our approach is to embed a part of the scene description or object descriptors as a watermark message into the video or audio streams. The amount of embedded information depends on the payload of the watermarking algorithms. We also analyze the possibility of embedding equivalent information into 3D models, depending on the application.

Imperceptibility and robustness including complexity are important issues in watermarking system, so it is desirable to consider these issues on design of watermarking system. In this paper, we will propose one of video watermarking in which motion between three continuous frames is utilized in order to provide the imperceptibility, robustness and acceptable complexity. Region R for watermarking is retrieved from three frames f(t), f(t+1) and f(t+2), and the watermarks which conveys a payload of L bits are embedded into DCT coefficients of Region R. Experimental results show that the proposed method could provide good fidelity and the robustness against MPEG-2 compression as well as some frame attacks.

In this paper, we describe a novel framework for watermarking 3D video objects via their texture information. Instead of classical existing algorithms dealing with 3D objects and that operate on meshes in order to protect the object itself, the main goal of our work is to retrieve information originally hidden in the texture image of the object, from resulting images or videos having used the 3D synthetic object. After developing the theory and practical details of our 3D object watermarking scheme, we present preliminary results of several experiments carried out in various conditions, ranging from ideal conditions (e.g. a priori known rendering parameters) to more realistic conditions (e.g. rendering projection estimated from 2D view) or within the context of possible attacks (e.g. mesh reduction).

Many watermarking applications may benefit from the availability of both the original and the watermarked media content to perform detection. However, one must take into account that the watermarked media might have undergone deformations such that direct correspondence between original and watermarked signal is not possible anymore. Extra processing is therefore needed to benefit from the availability of both signals. In video applications, besides the spatial deformations, one must also consider a possible temporal structure modification of the watermarked content. Such distortions include frame rate modifications, scene removal and temporal cropping. In this paper, we present how one can perform an automatic frames alignment of spatially and temporally deformed video sequences. Our approach consist in establishing a correspondence between automatically detected key-frames in the two sequences. The key-frame detection is inspired from existing methods dealing with scene cuts localization and semantics (MPEG7-like) scene classification. Different simulations show that the method can cope with common temporal deformations.

In refernce one, we proposed a technique or protocol for efficient temporal synchronization of video watermarks. Our technique is based on constructing a watermark with temporal redundancy, which allows the detector to efficiently establish and maintain synchronization without performing extensive search or explicit template signal embedding. In this paper, we describe several enhancements to our technique. A new class of key generators is presented which uses a cryptographic hash function to define the set of states and the state transition function of the finite state machine (FSM). In addition to the very large number of states and key-dependent state transition functions, which enhance security, the new class of key generators allows a limited degree of randomization. This randomization can take the form of multiple start states in the FSM, or by randomized state transitions. The non-deterministic behavior of a randomized FSM requires the detector to perform more search, but also makes the key sequence less predictable and improves security. We also describe a new method for temporal redundancy control which adaptively changes the watermark key based on the characteristics of the video. This new strategy prevents a loss of temporal redundancy (which leads to a loss of robustness) which can occur using the earlier naive method. The enhancements were implemented and the detection performance after frame dropping, transposition, and averaging attack are compared.

Security is an increasingly important attribute for multimedia applications that require prevention of unauthorized access to copyrighted data. Two approaches have been used to protect scalable video content in distribution: Partial encryption and progressive encryption. Partial encryption provides protection for only selected portions of the video. Progressive encryption allows transcoding with simple packet truncation, and eliminates the need to decrypt the video packets at intermediate network nodes with low complexity. Centralized Key Management with Secret Sharing (CKMSS) is a recent approach in which the group manager assigns unique secret shares to the nodes in the hierarchical key distribution tree. It allows the reconstruction of different keys by communicating different activating shares for the same prepositioned information. Once the group key is established, it is used until a member joins/leaves the multicast group or periodic rekeying occurs. In this paper, we will present simulation results regarding the communication and processing requirements of the CKMSS scheme applied to scalable video. In particular, we have measured the rekey message size and the processing time needed by the server for each join/leave request and periodic rekey event.

In this paper, we propose a key-based video watermarking system in which the watermark embedding and the video encoding are processed at the same time on MPEG-2. Since the watermark information would propagate to inter-frames through the motion compensated coding, the watermark is embedded in a single intra-frame but can be extracted from all frames in the same group of pictures (GOP). The watermark is embedded in the low frequency DCT coefficients of the intra-frames based on the block polarity. The block polarity is Tri-state Exclusive-Or (TXOR) with the watermark to generate the secret key, which labels the block locations of the embedded watermark. In the decoding end, the block polarity over a GOP is calculated by a weighted voting procedure according to the frame weighting. Finally, the watermark over a GOP can be obtained by TXOR operation of the key and the block polarity. The simulation results show that the system has great imperceptibility that the PSNRs of the watermarked frames are almost the same as the un-watermarked ones and more accurate normalized correlation (NC) can be obtained as well.

The illegal copying of movies in the cinema is now common practice. Although the quality is fairly low, the economic impact of these illegal copies can be enormous. Philips' digital cinema watermarking scheme is designed for the upcoming digital cinema format and will assist content owners and distributors with tracing the origin of illegal copies. In this paper we consider this watermarking scheme in more detail. A characteristic of this watermarking scheme is that it only exploits the temporal axis to insert a watermark. It is therefore inherently robust to geometrical distortions, a necessity for surviving illegal copying by camcorder recording. Moreover, the scheme resists frame rate conversions resulting from a frame rate mismatch between the camcorder and the projector. The watermarking scheme has been tested in a 'real' digital cinema environment with good results.

Forensic digital watermarking is a promising tool in the fight against piracy of copyrighted motion imagery content, but to be effective it must be (1) imperceptibly embedded in high-definition motion picture source, (2) reliably retrieved, even from degraded copies as might result from camcorder capture and subsequent very-low-bitrate compression and distribution on the Internet, and (3) secure against unauthorized removal. No existing watermarking technology has yet to meet these three simultaneous requirements of fidelity, robustness, and security. We describe here a forensic watermarking approach that meets all three requirements. It is based on the inherent robustness and imperceptibility of very low spatiotemporal frequency watermark carriers, and on a watermark placement technique that renders jamming attacks too costly in picture quality, even if the attacker has complete knowledge of the embedding algorithm. The algorithm has been tested on HD Cinemascope source material exhibited in a digital cinema viewing room. The watermark is imperceptible, yet recoverable after exhibition capture with camcorders, and after the introduction of other distortions such as low-pass filtering, noise addition, geometric shifts, and the manipulation of brightness and contrast.

The enormous economic incentives for safeguarding intellectual property in the digital domain have made forensic watermarking a research topic of considerable interest. However, a recent examination of some of the leading product development efforts reveals that at present there is no effective watermarking implementation that addresses both the fidelity and security requirements of high definition digital cinema. If Boeing Digital Cinema (BDC, a business unit of Boeing Integrated Defense Systems) is to succeed in using watermarking as a deterrent to the unauthorized capture and distribution of high value cinematic material, the technology must be robust, transparent, asymmetric in its insertion/detection costs, and compatible with all the other elements of Boeing's multi-layered security system, including its compression, encryption, and key management services.

Digital watermarking may be used to enforce copyright protection of digital cinema, by embedding in each projected movie an unique identifier (fingerprint). By identifying the source of illegal copies, watermarking will thus incite movie theatre managers to enforce copyright protection, in particular by preventing people from coming in with a handy cam. We propose here a non-blind watermark method to improve the watermark detection on very impaired sequences. We first present a study on the picture impairments caused by the projection on a screen, then acquisition with a handy cam. We show that images undergo geometric deformations, which are fully described by a projective geometry model. The sequence also undergoes spatial and temporal luminance variation. Based on this study and on the impairments models which follow, we propose a method to match the retrieved sequence to the original one. First, temporal registration is performed by comparing the average luminance variation on both sequences. To compensate for geometric transformations, we used paired points from both sequences, obtained by applying a feature points detector. The matching of the feature points then enables to retrieve the geometric transform parameters. Tests show that the watermark retrieval on rectified sequences is greatly improved.

Digital cinema can be defined as the digital electronic distribution and display of theatrical film content or live material to the theatre. However, this simple definition does not reflect the complex mesh of balanced business relationships between the different parties involved in the worldwide cinematic distribution. Its transition from an analogue to a digital business will happen progressively, starting with only content being digitally distributed and going until an all-digital business. From these facts derive strong requirements on digital rights management (DRM) systems for digital cinema and therefore on the digital rights language supporting it. This paper explores the requirements imposed by the cinematic distribution model and by its progressive transition to digital, and the impact these have on digital rights languages. We analyze the support provided by different digital rights languages, identifying weaknesses and exploring solutions in fulfilling the requirements of the digital cinema distribution model.

Digital cinema, a new frontier and crown jewel of digital multimedia, has the potential of revolutionizing the science, engineering and business of movie production and distribution. The advantages of digital cinema technology over traditional analog technology are numerous and profound. But without effective and enforceable copyright protection measures, digital cinema can be more susceptible to widespread piracy, which can dampen or even prevent the commercial deployment of digital cinema. In this paper we propose a novel approach of fingerprinting each individual distribution copy of a digital movie for the purpose of tracing pirated copies back to their source. The proposed fingerprinting technique presents a fundamental departure from the traditional digital watermarking/fingerprinting techniques. Its novelty and uniqueness lie in a so-called semantic or subjective transparency property. The fingerprints are created by editing those visual and audio attributes that can be modified with semantic and subjective transparency to the audience. Semantically-transparent fingerprinting or watermarking is the most robust kind among all existing watermarking techniques, because it is content-based not sample-based, and semantically-recoverable not statistically-recoverable.

In this paper we consider the problem of capacity analysis in the framework of information-theoretic model of data hiding. Capacity is determined by the stochastic model of the host image, by the distortion constraints and by the side information about watermarking channel state available at the encoder and at the decoder. We emphasize the importance of proper modeling of image statistics and outline the possible decrease in the expected fundamental capacity limits, if there is a mismatch between the stochastic image model used in the hider/attacker optimization game and the actual model used by the attacker. To obtain a realistic estimation of pssible embedding rates we propose a novel stochastic non-stationary image model that is based on geometrical priors. This model outperforms the previously analyzed EQ and spike models in reference application such as denoising. Finally, we demonstrate how the proposed model influences the estimation of capacity for real images. We extend our model to different transform domains that include orthogonal, biorthogonal and overcomplete data representations.

The paper discusses the limits of error correction coding for spread spectrum-based video watermarking. The error correction code has as input the watermark data bits and as output the values which will be scaled and used to modify the video pixels (transform coefficients). The data rate of the watermark can increase only at the expense of increasing code rate. Theoretically, the scheme is seen as a communication channel with Gaussian additive noise interference. Shannon's (ideal) spherical codes are used as the error correcting code to calculate the minimum signal to noise ratio (SNR) necessary for a coding scheme with a given block length to achieve a given error probability. This limit is different from Shannon's asymptotic limit, which is valid for infinite block lengths and zero error probability. In practice, in order to verify the Gaussian channel assumption, the error correction code is a concatenation of codes, of which the innermost is a repetition code. Several practical codes of different length and rates, such as turbo codes and BCH codes are investigated and their performance compared to that of the ideal code of the same size. The compromise block length/code rate is investigated for several marking schemes and attacks.

An undesirable side effect of many watermarking and data-hiding schemes is that the host signal into which auxiliary data is embedded is distorted. Finding an optimal balance between the amount of information embedded and the induced distortion is therefore an active field of research. In recent years, with the rediscovery of Costa's seminal paper Writing on Dirty Paper, there has been considerable progress in understanding the fundamental limits of the capacity versus distortion of watermarking and data-hiding schemes. For some applications, however, no distortion resulting from auxiliary data, however small, is allowed. In these cases the use of reversible data-hiding methods provide a way out. A reversible data-hiding scheme is defined as a scheme that allows complete and blind restoration (i.e. without additional signaling) of the original host data. Practical reversible data-hiding schemes have been proposed by Fridrich et al., but little attention has been paid to the theoretical limits. Some first results on the capacity of reversible watermarking schemes have been derived. The reversible schemes considered in most previous papers have a highly fragile nature: in those schemes, changing a single bit in the watermarked data would prohibit recovery of both the original host signal as well as the embedded auxiliary data. It is the purpose of this paper to repair this situation and to provide some first results on the limits of robust reversible data-hiding. Admittedly, the examples provided in this paper are toy examples, but they are indicative of more practical schemes that will be presented in subsequent papers.

Blind spread-spectrum watermarking schemes with correlation based detection algorithms suffer significantly from host signal interference. Improvements are possible when considering the host signal as side-information to the watermark encoder. Costa showed that, in a specific case, host signal interference can be avoided completely. The scalar Costa scheme, which operates on structured codebooks based on scalar quantizers, performs relatively close to the ideal Costa scheme in terms of maximum rate of reliable communication (capacity) over a channel with AWGN attack. The concept of amplitude limited channels in conjunction with repetition of the signal constellation is also known in the communications community by the name modulo channel, which has been studied into great detail. Starting from the modulo channel, which implies a constraint on the amplitude in the receiver, we model the watermarking scenario based on scalar quantizers. The capacity of this model is presented and compared to the theoretical results for SCS watermarking in the case of AWGN attacks. A final analysis of SCS in terms of optimality, based on the results with the modulo channel, shows that even binary SCS is near optimum for one-dimensional (symbol-by-symbol) embedding in terms of capacity for an i.i.d. Gaussian host signal and an appropriate AWGN attack.

Data hiding using quantization has revealed as an effective way of taking into account side information at the encoder. When quantizing more than one host signal samples there are two choices: (1) using the Cartesian product of several one-dimensional quantizers, as made in Scalar Costa Scheme (SCS); or (2) performing vectorial quantization. The second option seems better, as rate-distortion theory affirms that higher dimensional quantizers yield improved performance due to better sphere-packing properties. Although the embedding problem does resemble that of rate-distortion, no attacks or host signal characteristics are usually considered when designing the quantizer in this way. We show that attacks worsen the performance of the a priori optimal lattice quantizer through a counterexample: the comparison under Gaussian distortion of hexagonal lattice quantization against bidimensional Distortion-Compensated Quantized Projection (DC-QP), a data hiding alternative based in quantizing a linear projection of the host signal. Apart from empirical comparisons, theoretical lower bounds on the probability of decoding error of hexagonal lattices under Gaussian host signal and attack are provided and compared to the already analyzed DC-QP method.

The many recent publications that focus upon watermarking with side information at the embedder emphasize the fact that this side information can be used to improve practical capacity. Many of the proposed algorithms use quantization to carry out the embedding process. Although both powerful and simple, recovering the original quantization levels, and hence the embedded data, can be difficult if the image amplitude is modified. In our paper, we present a method that is similar to the existing class of quantization-based techniques, but is different in the sense that we first apply a projection to the image data that is invariant to a class of amplitude modifications that can be described as order preserving. Watermark reading and embedding is done with respect to the projected data rather than the original. Not surprisingly, by requiring invariance to amplitude modifications we increase our vulnerability to other types of distortions. Uniform quantization of the projected data generally leads to non-uniform quantization of the original data, which in turn can cause greater susceptibility to additive noise. Later in the paper we describe a strategy that results in an effective compromise between invariance to amplitude modification and noise susceptibility.

Robust hash functions are central to the security of multimedia content authentication systems. Such functions are sensitive to a key but robust to many allowed signal processing operations on the underlying content. Robustness of the hash function to changes in the original content implies the existence of a cluster in the feature space around the original contents feature vector, any point within which getting hashed to the same output. The shape and size of the cluster determines the trade-off between the robustness offered and the security of the authentication system based on the robust hash function. The clustering itself is based on a secret key and hence unknown to the attacker. However, we show in this paper that the specific clustering arrived at by a robust hash function may be possible to learn. Specifically, we look at a well known robust hash function for image data called the Visual Hash Function (VHF). Given just an input and its hash value, we show how to construct a statistical model of the hash function, without any knowledge of the secret key used to compute the hash. We also show how to use this model to engineer arbitrary and malicious collisions. Finally, we propose one possible modification to VHF so that constructing a model that mimics its behavior becomes difficult.

Digital watermarking has become an accepted technology for enabling multimedia protection schemes. Based on the introduced media independent protocol schemes for invertible data authentication in references 2, 4 and 5 we discuss the design of a new 3D invertible labeling technique to ensure and require high data integrity. We combine digital signature schemes and digital watermarking to provide a public verifiable integrity. Furthermore the protocol steps in the other papers to ensure that the original data can only be reproduced with a secret key is adopted for 3D meshes. The goal is to show how the existing protocol can be used for 3D meshes to provide solutions for authentication watermarking. In our design concept and evaluation we see that due to the nature of 3D meshes the invertible function are different from the image and audio concepts to achieve invertibility to guaranty reversibility of the original. Therefore we introduce a concept for distortion free invertibility and a concept for adjustable minimum distortion invertibility.

This paper proposes a unique watermarking technique for printed documents by superposing dot pattern blocks on backgrounds of the document images. The dot pattern block contains information, but it is not visually distracting as the printed documents look as if uniform dot patterns are superposed on the background. The dots in the dot pattern block are lined up to a certain direction so that these dots generate 2-D wavelet in a scanned image of the printed documents. We have described the difference of symbols "0" and "1" as the difference of wavelet direction, and each wavelet is detected by 2-D Gabor filter, which makes the symbol detection error rate very low. In addition, accuracy of detection is not affected by format or layout of documents. Alteration detection can be achieved by embedding original document image feature into document image itself as alteration detection data. When watermark is extracted, the alteration detection data and the image feature of scanned image, are compared to determine if alterations are made or not.

Digital watermarking is a promising technology for protecting intellectual property rights on digital content. Resolving authorship-disputes was one of the first and most propelling applications of robust digital watermarks, and much research effort has gone into protocols for resolving authorship-disputes by means of digital watermarks. Unfortunately, previous proposals lack formal definitions of their trust model, their assumptions, and requirements they should fulfill. This lack of formal definitions makes security proofs for such protocols impossible and many dispute resolving protocols, claimed to be secure, can be shown to be insecure. In this paper we set off to rigorously defining dispute resolving schemes based on a reasonable formal definition of "authorship." Building on this formal fundament, we analyze the most important proposals for dispute resolving, and discuss their connection to our authorship model. We show that existing proposals suffer from two major problems: First, they require an unnecessary high level of trust in the dispute resolving party. The second and even more serious is that the winner of the dispute is not guaranteed to be the rightful author of the disputed work (conclusiveness problem). As solutions, we propose dispute resolving schemes based on zero-knowledge watermark detection and asymmetric watermarking schemes.

A novel framework is proposed for lossless authentication watermarking of images which allows authentication and recovery of original images without any distortions. This overcomes a significant limitation of traditional authentication watermarks that irreversibly alter image data in the process of watermarking and authenticate the watermarked image rather than the original. In particular, authenticity is verified before full reconstruction of the original image, whose integrity is inferred from the reversibility of the watermarking procedure. This reduces computational requirements in situations when either the verification step fails or the zero-distortion reconstruction is not required. A particular instantiation of the framework is implemented using a hierarchical authentication scheme and the lossless generalized-LSB data embedding mechanism. The resulting algorithm, called localized lossless authentication watermark (LAW), can localize tampered regions of the image; has a low embedding distortion, which can be removed entirely if necessary; and supports public/private key authentication and recovery options. The effectiveness of the framework and the instantiation is demonstrated through examples.

This paper explores techniques that involve the use of the embedder's knowledge of the cover work to help determine the watermark signal to be added to it. While the receiver always seeks to maximize a detection statistic which is a function of an apriori known pseudorandom sequence, the signal added to the cover work by the embedder is allowed to vary, on a per-chip basis, based upon the characteristics of the cover work. Although adaptation of an added watermark signal can be aimed at minimization of visual artifacts, this paper focuses on adaptation of the watermark signal to improve the readability of the signal outside of any human visual system constraints. This idea can be applied in various scenarios. Two specific examples are discussed. When source models are available and maximum-likelihood detection is used, the added watermark signal can be allowed to adapt to host signal variations in order to maximize the likelihood ratio detection statistic used at the receiver. Another instance where per-chip variation can be put to use is when a pre-filter is used to suppress the cover work prior to reading the watermark signal. In this case, the watermark signal is varied in such a way as to maximize the signal at the output of the pre-filter.

For most data hiding applications, the main source of concern is the effect of lossy compression on hidden information. The objective of watermarking is fundamentally in conflict with lossy compression. The latter attempts to remove all irrelevant and redundant information from a signal, while the former uses the irrelevant information to mask the presence of hidden data. Compression on a watermarked image can significantly affect the retrieval of the watermark. Past investigations of this problem have heavily relied on simulation. It is desirable not only to measure the effect of compression on embedded watermark, but also to control the embedding process to survive lossy compression. In this paper, we focus on oblivious watermarking by assuming that the watermarked image inevitably undergoes JPEG compression prior to watermark extraction. We propose an image-adaptive watermarking scheme where the watermarking algorithm and the JPEG compression standard are jointly considered. Watermark embedding takes into consideration the JPEG compression quality factor and exploits an HVS model to adaptively attain a proper trade-off among transparency, hiding data rate, and robustness to JPEG compression. The scheme estimates the image-dependent payload under JPEG compression to achieve the watermarking bit allocation in a determinate way, while maintaining consistent watermark retrieval performance.

In this paper an empirical evaluation of the lossy compression properties of the JPEG2000 standard for watermarking purposes is performed. The JPEG2000 standard is used as a basic tool for determining both how and where the embedded watermark should be placed in the image. The original image is slightly modified in order to generate a similar image (indistinguishable by the observer), and the mark is embedded in the pixels presenting differences between both images. Previous experiments show that the properties of the resulting watermarking scheme depend strongly on the transformation stage characteristics of the lossy image compression system. Several parameters related to the JPEG2000 standard are tested, in addition to the compression ratio: the wavelet transform, block size, and the number of levels of decomposition for each block. These parameters affect not only the capacity but also the robustness of the watermarking scheme: the number of different pixels between the original image and the slightly modified one, and the distribution of such differences. For evaluation purposes we use the Stirmark benchmark. We compare the impact on system performance of each one of the JPEG2000 standard parameters, for several kinds of attacks, namely filtering (including sharpening), JPEG lossy compression, cropping, row and column removal, and a combination of rotation, cropping and scaling.

Wavelet filters can be parametrized to create an entire family of different wavelet filters. We discuss wavelet filter parametrization as a means to add security to wavelet based watermarking schemes. We analyze the influence of the number of filter parameters and use non-stationary multi-resolution decomposition where different wavelet filters are used at different levels of the decomposition. Using JPEG and JPEG2000 compression we assess the normalized correlation and Peak Signal to Noise Ratio (PSNR) behavior of the watermarks. The security against unauthorized detection is also investigated. We conclude that the proposed systems show good robustness against compression and depending on the resolution we choose for the parameters we get between 2⁹⁹ and 2¹⁸⁵ possible keys.

One way of recovering watermarks in geometrically distorted images is by performing a geometrical search. In addition to the computational cost required for this method, this paper considers the more important problem of false positives. The maximal number of detections that can be performed in a geometrical search is bounded by the maximum false positive detection probability required by the watermark application. We show that image and key dependency in the watermark detector leads to different false positive detection probabilities for geometrical searches for different images and keys. Furthermore, the image and key dependency of the tested watermark detector increases the random-image-random-key false positive detection probability, compared to the Bernoulli experiment that was used as a model.