## Data-Hiding Codes (2005)

Venue: | Proc. IEEE |

Citations: | 37 - 3 self |

### BibTeX

@INPROCEEDINGS{Moulin05data-hidingcodes,

author = {Pierre Moulin and Ralf Koetter},

title = {Data-Hiding Codes},

booktitle = {Proc. IEEE},

year = {2005},

pages = {2083--2127}

}

### OpenURL

### Abstract

This tutorial paper reviews the theory and design of codes for hiding or embedding information in signals such as images, video, audio, graphics, and text. Such codes have also been called watermarking codes; they can be used in a variety of applications, including copyright protection for digital media, content authentication, media forensics, data binding, and covert communications. Some of these applications imply the presence of an adversary attempting to disrupt the transmission of information to the receiver; other applications involve a noisy, generally unknown, communication channel. Our focus is on the mathematical models, fundamental principles, and code design techniques that are applicable to data hiding. The approach draws from basic concepts in information theory, coding theory, game theory, and signal processing, and is illustrated with applications to the problem of hiding data in images. Keywords—Coding theory, data hiding, game theory, image processing, information theory, security, signal processing, watermarking. I.

### Citations

9316 |
Elements of information theory
- Cover, Thomas
- 1991
(Show Context)
Citation Context ... adapt to this known interference. Indeed, there are clear connections between data hiding and information-theoretic problems of communication with side information at the encoder and/or decoder [28]–=-=[30]-=-. These connections have been identified independently by several researchers in 1999 [31]–[34] and subsequently developed in great detail. Other researchers, perhaps most notably Cox and Miller [4], ... |

2994 | New directions in cryptography
- Diffie, Hellman
- 1976
(Show Context)
Citation Context ...lowing cryptographic terminology, two basic types of systems can be used: private-key systems, based on Shannon’s theory of security [118]; and public-key systems as introduced by Diffie and Hellmann =-=[119]-=-.22 The study of practical, secure QIM codes is still in its infancy. Ideas include randomized sparse QIM [33], randomized dithering [65] and look-up tables [121], and randomized lattice rotations [70... |

1783 |
Vector Quantization and Signal Compression
- Gersho, Gray
- 1992
(Show Context)
Citation Context ...des to images is illustrated in Section XII. The paper concludes with a discussion in Section XIII. Three short appendixes summarizing relevant notions of coding theory [36], vector quantization (VQ) =-=[37]-=-, and detection theory [38] have been included. H. Notation We use uppercase letters to denote random variables, lowercase letters for their individual values, calligraphic fonts for sets, and boldfac... |

932 |
Noiseless coding of correlated information sources
- Slepian, Wolf
- 1973
(Show Context)
Citation Context ... performance index of interest. IV. BINNING SCHEMES: GENERAL PRINCIPLES Binning is an important information-theoretic technique used in many different scenarios ranging from distributed source coding =-=[55]-=- and the problem of encoding data with side information at the transmitter only [28], to the classical problem of decoding data with side information that is available at the receiver only [30], [56].... |

899 | Secure spread spectrum watermarking for multimedia
- Cox, Kilian, et al.
- 1997
(Show Context)
Citation Context ...s, which are elementary and nonrobust against noise. They are however closely related to more advanced binning techniques. The period 1995–1998 saw the development of SSM codes, which are more robust =-=[50]-=- and have been used in several commercial products; see [1] and references therein. Both SSM and LSB methods are reviewed next. We also comment briefly on system performance methods that were often us... |

807 |
Statistical Methods for Speech Recognition
- Jelinek
- 1999
(Show Context)
Citation Context ...e averaging measure on , e.g., a probability distribution on . There exist fairly good statistical models for host signals such as images, speech, etc. that can be used to select an appropriate [47], =-=[48]-=-. MOULIN AND KOETTER: DATA-HIDING CODES 2089sThe distortion introduced by the attacker can be measured in terms of or in terms of . A natural requirement for the attacker is that this distortion, meas... |

764 | The rate-distortion function for source coding with side information at the decoder
- Wyner, Ziv
- 1976
(Show Context)
Citation Context ... [55] and the problem of encoding data with side information at the transmitter only [28], to the classical problem of decoding data with side information that is available at the receiver only [30], =-=[56]-=-. Since blind data hiding is especially related to the problem of transmission with side information at the transmitter, we provide an overview of binning in this section. We start with a simple illus... |

744 |
Writing on dirty paper
- Costa
- 1983
(Show Context)
Citation Context ...2.8), we obtain [88] if , if (7.9) where . When , the optimal distributions turn out to be Gaussian test channels [30], [46], [88]; see Fig. 20. Closely related to this result is one derived by Costa =-=[29]-=- in 1983 for communications on an additive white Gaussian noise channel (with power ) in the presence of an i.i.d. Gaussian interference (with power ) that is known at the encoder but not at the decod... |

303 | Collusion-Secure Fingerprinting for Digital Data
- Bonen, Shaw
- 1998
(Show Context)
Citation Context ...y damaging) the document itself. Copyright protection led to the emergence of digital watermarking1 at the beginning of the Internet revolution, in the early 1990s. Fingerprinting and Traitor Tracing =-=[11]-=-. This is analogous to the copyright protection problem, with a twist: the distribution list for the digital document is limited, and a distinct digital signature is embedded in each document, making ... |

256 |
Information hiding techniques for steganography and digital watermarking, Artech house publishers
- Katzenbeisser, Petitcolas
- 2000
(Show Context)
Citation Context ...nson et al. [1] and by Petitcolas [2]. The recent books by Barni and Bartolini [3], Cox, Miller, and Bloom [4], Eggers and Girod [5], Johnson, Duric, and Jajodia [6], and Katzenbeisser and Petitcolas =-=[7]-=- are also valuable resources. The goal of this paper is to provide an overview of this field, focusing on the core principles and the mathematical methods that can be used for data hiding. We do not a... |

242 |
Coding for a Channel with Random Parameters
- Gelfand, Pinsker
- 1980
(Show Context)
Citation Context ...mally adapt to this known interference. Indeed, there are clear connections between data hiding and information-theoretic problems of communication with side information at the encoder and/or decoder =-=[28]-=-–[30]. These connections have been identified independently by several researchers in 1999 [31]–[34] and subsequently developed in great detail. Other researchers, perhaps most notably Cox and Miller ... |

240 | Nested linear/lattice codes for structured multiterminal binning
- Zamir, Shamai, et al.
- 2002
(Show Context)
Citation Context ...en the sets and is now The decoder’s output is 1 2 3 dist (5.14) defining dist . The quantity dist is a coordinatewise sum of squared quantization errors. 1) General Construction: The papers [64] and =-=[65]-=- presented a general approach for constructing structured binning schemes to approach capacity. The approach is based on nested lattices. A lattice in -dimensional Euclidean space is defined as a set ... |

239 | Information-theoretic Analysis of Information Hiding
- Moulin, O’Sullivan
- 1999
(Show Context)
Citation Context ...lves vulnerable to more advanced steganalysis methods. Where does the cat-and-mouse game stop? Again, statistical detection theory provides a natural and fundamental framework to answer this question =-=[34]-=-, [136]–[141]. The steganalyzer is essentially faced with a binary choice: decide whether data are hidden or not. Assume a statistical model (pdf ) for images or image features is available. If the st... |

210 | An information-theoretic model for steganography
- Cachin
- 2004
(Show Context)
Citation Context ...ation bits, � a ��� �w�. In this experiment, h ah a 4dB and x a 256. The figure shows representative points for the best known codes [256,128,38], [256,64,62], [256,32,96], [256,16,113], [256,8,128], =-=[256,4,136]-=-, [256,2,170], [256,1,256]. Case . Denoting by the -th block of The receiver first implements the modulo lattice operation received data, the receiver first implements the modulo lattice operation Ass... |

208 |
The prisoners’ problem and subliminal channel
- Simmons
- 1984
(Show Context)
Citation Context ...s to search through an expected number 2 of images to find a match. The image Alice selects is a perfectly natural one, and the steganalyzer (also called warden Willie by analogy to a prisoner’s game =-=[133]-=-) is fooled. The above method is computationally infeasible if the length of the message sequence is large. For such applications, other steganographic methods must be devised. The LSB embedding metho... |

204 |
A Wavelet Tour
- Mallat
- 2001
(Show Context)
Citation Context ...h that started in 1999 and includes [35], [63], [67], [68], [155]–[157]. A. Capacity Estimates Several transforms, including the two-dimensional (2-D) block DCT and the 2-D discrete wavelet transform =-=[158]-=- decompose images into approximately independent components that describe the local spatial-frequency contents of the image. To simplify the presentation, we focus on the 2-D block DCT using 8 8 block... |

200 |
Signal compression based on models of human perception
- Jayant, Johnston, et al.
- 1993
(Show Context)
Citation Context ... files, and other binary data files. While tractable, such distortion measures fail to capture the complexities of human perception, including masking and threshold effects. The reader is referred to =-=[41]-=- for an excellent overview of this subject. Detailed perceptual models for images and speech have been constructed and refined over time. A popular example in image processing is Watson’s metric [42],... |

194 | Multimedia dataembedding and watermarking technologies
- Swanson, Kobayashi, et al.
- 1998
(Show Context)
Citation Context ...nications Magazine (August 2001), Signal Processing (June 2001), and the IEEE Signal Processing Magazine (September 2000). The state of the art before 2000 is surveyed in the papers by Swanson et al. =-=[1]-=- and by Petitcolas [2]. The recent books by Barni and Bartolini [3], Cox, Miller, and Bloom [4], Eggers and Girod [5], Johnson, Duric, and Jajodia [6], and Katzenbeisser and Petitcolas [7] are also va... |

162 | Perceptual watermarks for digital images and video
- Wolfgang, Podilchuk, et al.
- 1999
(Show Context)
Citation Context ...an be locally adapted to host signal characteristics, e.g., (3.1) can be replaced with (3.6) where depends on the local characteristics of the host (e.g., frequency and temporal characteristics) [4], =-=[51]-=-. 2) To reduce decoder’s noise in (3.5), one can preprocess the host prior to embedding the watermark [4, Sec. 5.1]. This can be done using a causal preprocessor, leveraging information-theoretic resu... |

156 | Image coding based on mixture modeling of wavelet coefcients and a fast estimationquantization framework - LoPresto, Ramchandran, et al. - 1997 |

144 | Rotation, scale and translation invariant spread spectrum digital image watermarking”, signal process - Ruanaidh, Pun - 1998 |

142 | Watermarking as communications with side information - Cox, Miller, et al. |

142 |
Visual pattern discrimination
- Julesz
- 1962
(Show Context)
Citation Context ...ric [42], which is based on the concept of just noticeable differences and captures both threshold effects and spatial-frequency sensitivity of the human visual system. Psychovisual studies by Julesz =-=[43]-=- suggest that image textures with the same second-order statistics are perceived as identical by the human visual system. More accurate models have been developed later; given a natural texture one ca... |

139 |
Resolving Rightful Ownerships with Invisible Watermarking Techniques
- Craver, Memon, et al.
- 1998
(Show Context)
Citation Context ...arder to develop an effective copy attack against robust watermarking methods, in which watermark and content cannot be easily separated. Ambiguity Attacks. The main application is proof of ownership =-=[129]-=-. The attacker creates a forgery: a fake original host , together with a fake watermark (indexed by a fake message and a forged key ). He claims to have produced the disputed marked signal using . Thi... |

125 | Reliable communication under channel uncertainty
- Lapidoth, Narayan
- 1998
(Show Context)
Citation Context ...allowed to be unbounded. Capacity can again be achieved using a random binning scheme closely related to the one described above, and a particular universal decoder based on the method of types [30], =-=[91]-=-. This decoder evaluates the empirical first-order statistics of the pairs , for all possible codewords . The binning scheme is such that the probability distribution for (averaged over ) is memoryles... |

117 |
An Invisible Watermarking Technique for Image Verification
- Yeung, Mintzer
- 1997
(Show Context)
Citation Context ... theory, Willems in 1988 had already formulated a setup for quantization-based codes [58], but his ideas remained undeveloped for about ten years. Meanwhile, Swanson et al. [59] and Yeung and Mintzer =-=[60]-=- invented quantization codes that are based on sound ideas but introduce excessive distortion relative to QIM. To introduce QIM, we start out the simplest case of embedding one information bit in a si... |

113 | A Lapidoth, The Gaussian watermarking game
- Cohen
- 2002
(Show Context)
Citation Context ...his case, we have WNR 1 WNR (7.10) which admits an elegant MMSE (minimum mean squared error) interpretation [85]; also see (6.17). Additional extensions of Costa’s result have recently appeared [65], =-=[86]-=-, [87]. In particular, the capacity formula 1/2 1 is still valid if the interference is any finite-power sequence, for any values of and . Also, the capacity for the following two data-hiding games ar... |

113 | Watermarking Resisting to Translation, Rotation and Scaling
- Kutter
- 1998
(Show Context)
Citation Context ...chronization problems. MOULIN AND KOETTER: DATA-HIDING CODES 2113sinformation about the actual message, so less power is available for the information-bearing signals. • Embedding in invariant domain =-=[103]-=-–[105]. The difficulty with these methods is to construct suitable invariants. This has been done for operations such as scaling, translation, and rotation, but it is difficult to extend this approach... |

109 | R.Du, “Reliable detection of LSB steganography in color and grayscale images
- Fridrich, Goljan
(Show Context)
Citation Context ...ell. These ideas are described in papers by Fridrich et al., who developed a simple but surprisingly powerful algorithm (RS steganalysis) to detect the presence of hidden information in the LSB plane =-=[134]-=-, [135]. Recently improved LSB steganographic methods have been developed that can resist RS steganalysis, but these new methods are themselves vulnerable to more advanced steganalysis methods. Where ... |

108 | Digital watermarking for telltale tamper proofing and authentication
- Kundur, Hatzinakos
- 1999
(Show Context)
Citation Context ...de automatic video surveillance [12] and authentication of drivers’ licenses [13]. Media Forensics. The goal here is to extract information about any processing that may have been applied to a signal =-=[14]-=-. For instance, authentication methods would reveal that an image has been tampered with, but not how. Forensic methods would take the analysis one step further, e.g., by indicating which part(s) of t... |

106 | Information Hiding-a Survey
- Petitcolas, Anderson, et al.
- 1999
(Show Context)
Citation Context ...gust 2001), Signal Processing (June 2001), and the IEEE Signal Processing Magazine (September 2000). The state of the art before 2000 is surveyed in the papers by Swanson et al. [1] and by Petitcolas =-=[2]-=-. The recent books by Barni and Bartolini [3], Cox, Miller, and Bloom [4], Eggers and Girod [5], Johnson, Duric, and Jajodia [6], and Katzenbeisser and Petitcolas [7] are also valuable resources. The ... |

101 |
Robust Template Matching for Affine Resistant Image Watermarks
- Pereira, Pun
(Show Context)
Citation Context ...f error, ��� € � versus number of information bits, � a ��� �w�. In this experiment, h ah a 4dB and x a 256. The figure shows representative points for the best known codes [256,128,38], [256,64,62], =-=[256,32,96]-=-, [256,16,113], [256,8,128], [256,4,136], [256,2,170], [256,1,256]. Case . Denoting by the -th block of The receiver first implements the modulo lattice operation received data, the receiver first imp... |

100 |
Communication theory of secrecy systems. Bell system technical journal
- Shannon
- 1949
(Show Context)
Citation Context ..., we restrict our attention to conventional cryptographic keys. Following cryptographic terminology, two basic types of systems can be used: private-key systems, based on Shannon’s theory of security =-=[118]-=-; and public-key systems as introduced by Diffie and Hellmann [119].22 The study of practical, secure QIM codes is still in its infancy. Ideas include randomized sparse QIM [33], randomized dithering ... |

100 | Detecting hidden messages using higherorder statistical models - Farid |

95 | Scalar Costa Scheme for Information Embedding
- Eggers, Bauml, et al.
- 2003
(Show Context)
Citation Context ...he real line in Fig. 4. They form two lattices13 4 4 (5.3) 1) Original QIM: In [57], the marked signal is defined as . (5.4) 11Later termed “scalar Costa scheme” when scalar quantizers are used [61], =-=[62]-=-. We retain the original QIM terminology in this paper. 12Dithering is classical technique used in signal compression for improving the perceptual aspect of quantized signals. 13Strictly speaking, two... |

84 | Anti-Collusion Fingerprinting For Multimedia - Trappe, Wu, et al. - 2003 |

79 |
Duality between source coding and channel coding and its extension to the side information case
- Pradhan, Chou, et al.
- 2003
(Show Context)
Citation Context ... function is achieved by time-sharing. See Fig. 18. In both cases, the worst attack is a binary symmetric channel (BSC) with crossover probability . The capacity formula (7.8) was derived in [80] and =-=[81]-=- under the assumption of a fixed attack channel. B. Random Binning In principle, the capacity bound can be approached using a random binning coding technique [28], [30], which exemplifies the role of ... |

79 |
Memon, ”Counterfeiting attacks on oblivious block-wise independent invisible watermarking schemes
- Holliman, N
- 2000
(Show Context)
Citation Context ... bound on log probability of error, ��� € � versus number of information bits, � a ��� �w�. In this experiment, h ah a 4dB and x a 256. The figure shows representative points for the best known codes =-=[256,128,38]-=-, [256,64,62], [256,32,96], [256,16,113], [256,8,128], [256,4,136], [256,2,170], [256,1,256]. Case . Denoting by the -th block of The receiver first implements the modulo lattice operation received da... |

78 |
Improved spread spectrum: a new modulation technique for robust watermarking
- Malvar, Florencio
- 2003
(Show Context)
Citation Context ... information available causally to the encoder [31]. In data hiding, however, the encoder need not be restricted to causal strategies. Good results have been obtained using linear preprocessing [52], =-=[53]-=-. The embedding rule in [52] is of the form (3.7) where is a matrix that depends on the second-order statistics of and can be optimized against worst case filtering and colored noise attacks. The embe... |

77 | On lattice quantization noise
- Zamir, Feder
- 1996
(Show Context)
Citation Context ...ent dither vector is potentially selected for each length- host-data block. For simplicity we first consider the case (one single data block). Case . Write Under high-rate lattice quantization theory =-=[75]-=-, the quantization noise may be modeled as random, independent of , and uniformly distributed over ; moreover, that model is exact if a dithered lattice quantizer is used [76], [77]. The embedding dis... |

77 | Geometrically invariant watermarking using feature points - Bas, Chassey, et al. |

73 | A blind watermarking scheme based on structured codebooks
- Eggers, Su, et al.
- 2000
(Show Context)
Citation Context ...s on the real line in Fig. 4. They form two lattices13 4 4 (5.3) 1) Original QIM: In [57], the marked signal is defined as . (5.4) 11Later termed “scalar Costa scheme” when scalar quantizers are used =-=[61]-=-, [62]. We retain the original QIM terminology in this paper. 12Dithering is classical technique used in signal compression for improving the perceptual aspect of quantized signals. 13Strictly speakin... |

73 | The duality between information embedding and source coding with side information and some applications
- Chen, Wornell
- 2003
(Show Context)
Citation Context ... capacity function is achieved by time-sharing. See Fig. 18. In both cases, the worst attack is a binary symmetric channel (BSC) with crossover probability . The capacity formula (7.8) was derived in =-=[80]-=- and [81] under the assumption of a fixed attack channel. B. Random Binning In principle, the capacity bound can be approached using a random binning coding technique [28], [30], which exemplifies the... |

66 |
Embedding secret information into a dithered multilevel image
- Tanaka, Nakamura, et al.
- 1990
(Show Context)
Citation Context ...mark-to-document ratio in the literature). An alternative definition with in place of in the denominator is sometimes used. III. EARLY WORK The first papers on data hiding appeared in the early 1990s =-=[49]-=-. The ideas proposed during that period include least significant bit (LSB) embedding techniques, which are elementary and nonrobust against noise. They are however closely related to more advanced bi... |

59 |
Translation Resilient Watermarking for Images
- Lin, Wu, et al.
(Show Context)
Citation Context ...zation problems. MOULIN AND KOETTER: DATA-HIDING CODES 2113sinformation about the actual message, so less power is available for the information-bearing signals. • Embedding in invariant domain [103]–=-=[105]-=-. The difficulty with these methods is to construct suitable invariants. This has been done for operations such as scaling, translation, and rotation, but it is difficult to extend this approach to mo... |

59 | Watermark copy attack
- Kutter, Voloshynovskiy, et al.
- 2000
(Show Context)
Citation Context ...han using a brute-force approach (which would be infeasible if the key space is large)? The answer is “yes” for a basic spread-spectrum scheme [125] but unknown for more complex schemes. Copy Attacks =-=[127]-=-, [128]. Here the attacker illegally embeds a watermark derived from one document into a new document. For instance, if the auxiliary document is an image marked using an LSB embedding technique, the ... |

57 | How realistic is photorealistic
- Lyu, Farid
- 1999
(Show Context)
Citation Context ...of features and generate synthetic textures that look like the original one [44]. Advances in computer graphics have likewise made it possible to generate synthetic images that look like natural ones =-=[45]-=-. The relevance of this work to image watermarking, for instance, if if is that a sophisticated embedder or attacker could replace a textured portion of an image (say a grass field) with a similar-loo... |

56 |
Information Hiding : Steganography and Watermarking
- JOHNSON, DURIC, et al.
- 2001
(Show Context)
Citation Context ... 2000 is surveyed in the papers by Swanson et al. [1] and by Petitcolas [2]. The recent books by Barni and Bartolini [3], Cox, Miller, and Bloom [4], Eggers and Girod [5], Johnson, Duric, and Jajodia =-=[6]-=-, and Katzenbeisser and Petitcolas [7] are also valuable resources. The goal of this paper is to provide an overview of this field, focusing on the core principles and the mathematical methods that ca... |

55 | Multiresolution scene-based video watermarking using perceptual models - Swanson, Zhu, et al. - 1998 |

51 | On universal quantization by randomized uniform/lattice quantizers
- Zamir, Feder
- 1992
(Show Context)
Citation Context ...t of and uniformly distributed over , it turns out that the quantization 2098 PROCEEDINGS OF THE IEEE, VOL. 93, NO. 12, DECEMBER 2005serror is also independent of and uniformly distributed over [76], =-=[77]-=-. This property considerably simplifies the analysis and understanding of QIM schemes and has therefore been popular in theoretical analyses. Additionally, is shared with the decoder; can thus be used... |