#### DMCA

## Attack Modelling: Towards a Second Generation Watermarking Benchmark (2001)

### Cached

### Download Links

- [vision.unige.ch]
- [vision.unige.ch]
- [cvml.unige.ch]
- DBLP

### Other Repositories/Bibliography

Venue: | Signal Processing, Special Issue on Information Theoretic Issues in Digital Watermarking |

Citations: | 47 - 4 self |

### Citations

5126 | Stochastic relaxation, Gibbs distributions, and the Bayesian restoration of images.
- Geman, Geman
- 1984
(Show Context)
Citation Context ...tes the matrix norm, and ae (res) is the energy function for the sGG model. In practice, iterative algorithms are often used to solve the above problem. Examples of such algorithms are the stochastic =-=[23]-=- and deterministic annealing (mean-field annealing) [17], graduated nonconvexity [2], ARTUR algorithm [49] or its generalization [3]. Of course, it is preferable to obtain the closed form solution for... |

1531 |
Embedded Image Coding Using Zerotrees of Wavelet Coefficients,”
- Shapiro
- 1993
(Show Context)
Citation Context ...e redundancy of real images. A well-known example of intrascale model based methods that exploits the zero-correlation across the subbands is the embedded zerotree wavelet (EZW) algorithm proposed in =-=[63]-=- and its extended implementation in SPHIT [61]. The example of interscale coder is EQ-coder proposed by Lopresto et al [41] that utilizes the above stochastic image models for quantization scheme desi... |

1139 | A new, fast and efficient Image codec based on set partitioning in hierarchical trees. IEEE Trans. on circuit and systems for video tech.,
- Said, Pearlman
- 1996
(Show Context)
Citation Context ...ple of intrascale model based methods that exploits the zero-correlation across the subbands is the embedded zerotree wavelet (EZW) algorithm proposed in [63] and its extended implementation in SPHIT =-=[61]-=-. The example of interscale coder is EQ-coder proposed by Lopresto et al [41] that utilizes the above stochastic image models for quantization scheme design. The aim of this section is to show the con... |

1100 | Secure Spread Spectrum Watermarking for Multimedia,”
- Cox
- 1997
(Show Context)
Citation Context ...nd decoder in private watermarking. Class I: SI is not available (switches A and B open). It is a typical case for all earlier watermarking algorithms that were inspired by the original papers of Cox =-=[13]-=- and Tirkel [65]. It assumes that the watermark is embedded in the cover image and is then decoded without reference to information about channel state. The detection of watermarks is mostly based on ... |

1041 |
Writing on dirty paper,”
- Costa
- 1983
(Show Context)
Citation Context ...sists of the fact that the theoretical capacity of oblivious watermarking scheme is equal to that of a decoder with access to the cover data. This conclusion is based on the remarkable paper of Costa =-=[11]-=-. Therefore, there is no more need in the cover data for the decoder, if the cover data is used as SI by the encoder. However, this approach has several drawbacks. First, the complexity of the encoder... |

729 |
Ideal spatial adaptation via wavelet shrinkage
- Donoho, IM
- 1994
(Show Context)
Citation Context ...ers are included in the benchmarking tool Stirmark [56] that can be downloaded from site [55]. The representatives of the MAP-estimates are the adaptive Wiener (Lee) filter, soft- and hard- shrinkage =-=[20]-=- which are considered below in more details with respect to the watermarking applications. A detailed analysis of the denoising methods is given in Section 6.1. Lossy compression has recently been est... |

325 | Attacks on copyright marking systems,"
- Petitcolas, Anderson, et al.
- 1998
(Show Context)
Citation Context ...ian, ffl\Gamma contaminated (mixture model of Gaussian and Laplacian), and Cauchy watermark distributions respectively. The averaging and median filters are included in the benchmarking tool Stirmark =-=[56]-=- that can be downloaded from site [55]. The representatives of the MAP-estimates are the adaptive Wiener (Lee) filter, soft- and hard- shrinkage [20] which are considered below in more details with re... |

311 | Deterministic edge-preserving regularization in computed imaging,”
- Charbonnier, Blanc-Féraud, et al.
- 1997
(Show Context)
Citation Context ...hms are often used to solve the above problem. Examples of such algorithms are the stochastic [23] and deterministic annealing (mean-field annealing) [17], graduated nonconvexity [2], ARTUR algorithm =-=[49]-=- or its generalization [3]. Of course, it is preferable to obtain the closed form solution for the analysis of the obtained estimate. To generalize the iterative approaches to the minimization of the ... |

265 | Information-Theoretic Analysis of Information Hiding,”
- Moulin, J
- 1999
(Show Context)
Citation Context ...e when subtracting the estimated watermark, to take into account the content adaptive nature of the embedding algorithm. Otherwise, distortions will be unnecessarily introduced. Moulin and O'Sullivan =-=[45]-=- consider the influence of the attacks from the information-theoretic point of view and come to the conclusion that the additive white Gaussian noise attack can be asymptotically optimal with respect ... |

224 | Spatially Adaptive Wavelet Thresholding with Context Modeling for Image denoising”,
- Chang, Yu, et al.
- 2000
(Show Context)
Citation Context ...the watermark removal problem. The idea of 28 using lossy compression for denoising has been proposed originally in [62,46]. There are two main recent points of view on this subject based on the work =-=[40,8]-=-. The first approach [40] refers to a theory of complexity regularization and the second one is the generalization of the shrinking principle to the case of quantized data [8]. In our formulation, los... |

217 | Analysis of multiresolution image denoising schemes using generalized Gaussian and complexity priors,”
- Moulin, Liu
- 1999
(Show Context)
Citation Context ...o this problem is soft-shrinkage [68] which is well known in the wavelet domain [20] b x = y + max (0; jy \Gamma yj \Gamma T ) sign (y \Gamma y) (29) where T = p 2 oe 2 w oe x . It was shown recently =-=[44]-=- that the hard-shrinkage denoiser can be determined under the same priors in the limiting case fl ! 0: b x = y + / (jy \Gamma yj ? T ) (y \Gamma y) (30) where / (:) denotes a thresholding function tha... |

189 | Watermarking of uncompressed and compressed video,”
- Hartung, Girod
- 1998
(Show Context)
Citation Context ...performance of these schemes is very poor due to two basic assumptions made: all attacks are modeled as additive stationary Gaussian noise that results in the simple correlation detection 11 receiver =-=[59,24]-=-. Secondly, these schemes assume no geometrical attacks. Class II: SI is available at encoder only (A open, B closed). This scheme has found recently a lot of attention in the watermarking community d... |

183 | DCT quantization matrices visually optimized for individual images”, - Watson - 1993 |

179 |
Image-adaptive Watermarking Using Visual Models. In
- Podilchuk, Zeng
- 1998
(Show Context)
Citation Context ...performance of these schemes is very poor due to two basic assumptions made: all attacks are modeled as additive stationary Gaussian noise that results in the simple correlation detection 11 receiver =-=[59,24]-=-. Secondly, these schemes assume no geometrical attacks. Class II: SI is available at encoder only (A open, B closed). This scheme has found recently a lot of attention in the watermarking community d... |

169 | Watermarking as communications with side information
- Cox, Miller, et al.
- 1999
(Show Context)
Citation Context ...ume no geometrical attacks. Class II: SI is available at encoder only (A open, B closed). This scheme has found recently a lot of attention in the watermarking community due to the publication of Cox =-=[15]-=-. The block diagram of this scheme is shown in Figure 4. Attacks Encoder Image channel state estimation c y y' b x Embedder Extractor Decoder b ^ k Fig. 4. Watermarking as communication with side info... |

166 | Image coding based on mixture modeling of wavelet coefficients and a fast estimationquantization framework,” in
- LoPresto, Ramchandran, et al.
- 1997
(Show Context)
Citation Context ...ss the subbands is the embedded zerotree wavelet (EZW) algorithm proposed in [63] and its extended implementation in SPHIT [61]. The example of interscale coder is EQ-coder proposed by Lopresto et al =-=[41]-=- that utilizes the above stochastic image models for quantization scheme design. The aim of this section is to show the connection between image denoising and lossy compression with respect to the wat... |

165 |
Secure spread spectrum watermarking for images, audio, and video,”
- Cox, Killian, et al.
- 1996
(Show Context)
Citation Context ...ting image quality based on the Watson metric which overcomes the limitations of the PSNR. 1 Introduction Digital watermarking has emerged as an appropriate tool for the protection of author's rights =-=[12]-=-. It is now well accepted that an effective watermarking scheme must successfully deal with the triple requirement of imperceptibility (visibility) - robustness - capacity [58]. Imperceptibility requi... |

163 |
Rotation, scale and translation invariant spread spectrum digital image watermarking, "
- O'Ruanaidh, Pun
- 1998
(Show Context)
Citation Context ...ing methods survive after these attacks due to the usage of special synchronization techniques. Robustness to the global geometrical distortions rely on the use of either a transform invariant domain =-=[48]-=-, or an additional template [50,10,51], or an Autocorrelation Function (ACF) of the watermark itself [35,67]. If robustness to global affine transformations is more or less a solved issue, the local r... |

145 |
Signal Detection in Non-Gaussian Noise.
- Kassam
- 1988
(Show Context)
Citation Context ... where oe fi (i) is constant for the given codeword bit c i . The nonlinear structure of the matched filter is similar to local optimum detector nonlinearity that limits the outliers of the sGG model =-=[32]-=-. This model was considered for the DCT domain watermarking by Hern'andez et al [26] where the sGG model presented the distribution of the DCT coefficients in 64 equivalent channels of JPEG compressio... |

144 | A Fair Benchmark for Image Watermarking Systems”,
- Kutter, Petitcolas
- 1999
(Show Context)
Citation Context ...ut masking: PSNR=24.61dB, wPSNR=29.3dB, TPE=9.27, NB1=146, NB2=3 generation benchmark. We note that the benchmark we propose is not intended to replace the benchmark proposed by Kutter and Petitcolas =-=[36]-=-, but rather to complement it. While their benchmark heavily weights geometric transformations and contains non-adaptive attacks, the benchmark we propose includes models of the image and watermark in... |

138 | DCT-domain watermarking techniques for still images, detector performance analysis and a new structure.
- Hernandez, Amado, et al.
- 2000
(Show Context)
Citation Context ...e of the matched filter is similar to local optimum detector nonlinearity that limits the outliers of the sGG model [32]. This model was considered for the DCT domain watermarking by Hern'andez et al =-=[26]-=- where the sGG model presented the distribution of the DCT coefficients in 64 equivalent channels of JPEG compression. The authors considered this model assuming that all fading in the equivalent chan... |

123 | Watermarking resisting to translation, rotation and scaling.
- Kutter
- 1998
(Show Context)
Citation Context ... parallel channels. Moreover, the projection function can have a particular spatial structure with given correlation properties that can be used for the recovery of affine geometrical transformations =-=[34,67]-=-. The resulting watermark is obtained as the superposition w (j) = K X k=1 c k p k (j) M (j) (1) where j 2 Z. The watermark embedder performs the insertion of the watermark into the cover image in som... |

98 |
A DCT-domain system for robust image watermarking”,
- Barni, Bartolini, et al.
- 1998
(Show Context)
Citation Context ... attack can be obtained, when the embedding rule is known. We consider some generalized embedding rule (3). In practice, there are three mostly used embedding schemes: additive linear, multiplicative =-=[5,6,59]-=- and quantization index modulation (QIM) [9]. More generally, all of them could be considered from the point of view of some generalized linear additive scheme: y = x +H (x; wjM) (20) where H(:; :) is... |

96 | Some general methods for tampering with watermarks. Selected Areas in Communications,
- Cox, Linnartz
- 1998
(Show Context)
Citation Context ...on through an exhaustive search. Since many watermarking schemes use a secret key it is very important to use keys with a secure length. Another attack in this category is the so called Oracle attack =-=[54,14]-=- which can be used to create a non-watermarked image when a watermark detector device is available. 2.4 Protocol attacks The protocol attacks aim at attacking the concept of the watermarking applicati... |

95 | A stochastic approach to content adaptive digital image watermarking”, in
- Voloshynovskiy
- 1999
(Show Context)
Citation Context ...ame with the attacker consisting of making the watermark unpredictable. There are two basic concepts that focuses on the above issue: Power-Spectrum Condition (PSC) [64] and Noise Visibility Function =-=[68]-=-. 35 Power-Spectrum Condition (PSC): An idealized theoretical approach for analyzing estimation-based attacks treats the cover data and watermark as independent, zero-mean, stationary Gaussian random ... |

88 |
A method for signature casting on digital images,
- Pitas
- 1996
(Show Context)
Citation Context ... similar attack based on weighted mean prediction was proposed by Holliman at al [30] where authors report their success in removal of watermarks produced by the watermarking scheme proposed by Pitas =-=[57] and the c-=-ommercial software of Digimark [10]. Su and Girod [64] propose a "Wiener attack". The proposed attack consists of three steps: prediction of the watermark based on the Wiener filter, subtrac... |

87 | Can invisible watermarks resolve rightful ownerships?
- Craver, Memon, et al.
- 1996
(Show Context)
Citation Context ... watermark detector device is available. 2.4 Protocol attacks The protocol attacks aim at attacking the concept of the watermarking application. The first protocol attack was proposed by Craver et al =-=[16]-=-. They introduce the framework of invertible watermark and show that for copyright protection applications watermarks need to be non-invertible. The idea of inversion consists of the fact that an atta... |

83 |
Globally convergent edge-preserving regularized reconstruction: An application to limited-angle tomography,”
- Delaney, Bresler
- 1998
(Show Context)
Citation Context ... the above problem. Examples of such algorithms are the stochastic [23] and deterministic annealing (mean-field annealing) [17], graduated nonconvexity [2], ARTUR algorithm [49] or its generalization =-=[3]-=-. Of course, it is preferable to obtain the closed form solution for the analysis of the obtained estimate. To generalize the iterative approaches to the minimization of the non-convex function (23) w... |

82 | A blind watermarking scheme based on structured codeblocks”,
- Eggers, Su, et al.
(Show Context)
Citation Context ...nt that can lead to the missmatch between what was assumed for the design of the codebook and the real situation. To reduce the complexity of the encoder, different practical algorithms were proposed =-=[9,21,52]-=-. However, geometrical attacks and the attacks whose statistics are different from additive Gaussian remain an open issue. To relax the lack of the decoder adaptivity to the attacking channel state on... |

82 | Spread spectrum watermarking: Malicious attacks and counterattacks
- Hartung, Su, et al.
- 1999
(Show Context)
Citation Context ...o perform remodulation in such a way that the projection of the watermark on the space p in (12) will be on average a zero-mean vector. The particular cases of this generalized attack were studied in =-=[38,25]-=- assuming that the watermark is extracted from the stego image with some strength factor. These attacks have several drawbacks in the case of content-adaptive watermarking, where the strength of the w... |

80 |
Digital Pictures: Representation and Compression.
- Netravali, Haskeli
- 1988
(Show Context)
Citation Context ...tion of different weights for the perceptually different regions oppositely to the PSNR where all regions are treated with the same weight. Originally this idea was presented by Netravali and Haskell =-=[47]-=- with application to image compression. In application to watermarking quality evaluation it was reported in [69] using the NVF as a weighting matrix: wPSNR = 10log 10 max(x) 2 jjx 0 \Gamma xjj 2 NVF ... |

74 | The Watermark Copy Attack.
- Kutter, Voloshynocskiy, et al.
- 2000
(Show Context)
Citation Context ... generation benchmarking is discussed in Section 12. Section 13 concludes the paper. 2 State-of-art Watermarking attacks We will adopt the attack classification scheme presented in the previous paper =-=[37]-=-. The wide class of existing attacks can be divided into four main categories: removal attacks, geometrical attacks, cryptographic attacks and protocol attacks. Figure 1 summarizes the different attac... |

67 | Statistical analysis of watermarking schemes for copyright protection of images,”
- Hernandez, Perez-Gonzalez
- 1999
(Show Context)
Citation Context ...ta after affine transformation based on the above mentioned self-reference principle, and the opportunity to enhance the decoding performance by reducing the variance of the image considered as noise =-=[35,27]-=-, have motivated the development of key-independent watermark extraction methods. They could represent the main danger to linear additive watermarking technologies, as will be shown below. 14 Differen... |

56 | A new, fast, and e_cient image codec based on set partitioning in hierarchical trees. - Said, Pearlman - 1996 |

54 | Fast robust template matching for affine resistant image watermarks
- Pereira, Pun
- 1999
(Show Context)
Citation Context ...attacks due to the usage of special synchronization techniques. Robustness to the global geometrical distortions rely on the use of either a transform invariant domain [48], or an additional template =-=[50,10,51]-=-, or an Autocorrelation Function (ACF) of the watermark itself [35,67]. If robustness to global affine transformations is more or less a solved issue, the local random alterations integrated in Stirma... |

53 | Performance Analysis of a 2-D Multipulse Amplitude Modulation Scheme for Data Hiding and Watermarking of Still Images
- Hernandez, Perez-Gonzalez, et al.
- 1998
(Show Context)
Citation Context ...used for authentication purposes. To convert the message into a form efficient for communication, it is encoded using either error correction codes (ECC) or modulated using binary antipodal signaling =-=[29]-=- or M-ary modulation [35]. With respect to ECC, mostly Bose Chaudhuri (BCH) or convolutional codes are used [48,28]. Recent publications [52,67] report the successful results using novel Turbo codes a... |

48 | Template based recovery of Fourier-based watermarks using Log-polar and Log-log maps
- Pereira, Ruanaidh, et al.
- 1999
(Show Context)
Citation Context ...attacks due to the usage of special synchronization techniques. Robustness to the global geometrical distortions rely on the use of either a transform invariant domain [48], or an additional template =-=[50,10,51]-=-, or an Autocorrelation Function (ACF) of the watermark itself [35,67]. If robustness to global affine transformations is more or less a solved issue, the local random alterations integrated in Stirma... |

47 |
Dither Modulation: A New Approach to Digital Watermarking and Information Embedding,
- Chen
- 1999
(Show Context)
Citation Context ...nt that can lead to the missmatch between what was assumed for the design of the codebook and the real situation. To reduce the complexity of the encoder, different practical algorithms were proposed =-=[9,21,52]-=-. However, geometrical attacks and the attacks whose statistics are different from additive Gaussian remain an open issue. To relax the lack of the decoder adaptivity to the attacking channel state on... |

47 |
Filtering random noise from deterministic signals via data compression,”
- Natarajan
- 1995
(Show Context)
Citation Context ...ng applications. A detailed analysis of the denoising methods is given in Section 6.1. Lossy compression has recently been established to have roughly the same influence on noise removal as denoising =-=[62,46]-=-. This was reported with respect to watermarking applications in [69] and is considered in Section 6.2. Both the denoising and the lossy compression can significantly reduce the capacity of watermarki... |

40 |
Perceptual watermarking of still images
- Podilchuk, Zeng
- 1996
(Show Context)
Citation Context ...tion of author's rights [12]. It is now well accepted that an effective watermarking scheme must successfully deal with the triple requirement of imperceptibility (visibility) - robustness - capacity =-=[58]-=-. Imperceptibility requires that the Preprint submitted to Elsevier Science 20 June 2001 marked data and the original data should be perceptually undistinguishible. Robustness refers to the fact that ... |

39 |
A new decoder for the optimum recovery of nonadditive watermarks,”
- Barni, Bartolini, et al.
- 2001
(Show Context)
Citation Context ...ear additive techniques. The derivation of the optimal MAP estimate for multiplicative watermarks or generally non-additive techniques is required to cover methods like SysCop of MediaSec [42], Barni =-=[7]-=- and Pereira [52,53] that are mostly used in the transform domains. Although the above classification makes it possible to have a clear separation between the different classes of attacks, it is neces... |

34 | Complexity-regularized image denoising,” in
- Liu, Moulin
- 1997
(Show Context)
Citation Context ...the watermark removal problem. The idea of 28 using lossy compression for denoising has been proposed originally in [62,46]. There are two main recent points of view on this subject based on the work =-=[40,8]-=-. The first approach [40] refers to a theory of complexity regularization and the second one is the generalization of the shrinking principle to the case of quantized data [8]. In our formulation, los... |

34 |
The role of information theory in watermarking and its application to image watermarking,
- Moulin
- 2001
(Show Context)
Citation Context ...eriods that will increase the energy of watermark, the attacker has to increase the varaince of noise. Therefore, such an attack is useless from a practical point of view. In more recent paper Moulin =-=[43]-=- agrees with the conclusion that the optimal attack should consist of two cascades of the MMSE estimator of the cover data and a Gaussian noise. However, it is important to note that Moulin and O'Sull... |

34 | Power-spectrum condition for energy-efficient watermarking,
- Su, Girod
- 2002
(Show Context)
Citation Context ...y Holliman at al [30] where authors report their success in removal of watermarks produced by the watermarking scheme proposed by Pitas [57] and the commercial software of Digimark [10]. Su and Girod =-=[64] propose a-=- "Wiener attack". The proposed attack consists of three steps: prediction of the watermark based on the Wiener filter, subtraction of the estimated watermark from the stego with some strengt... |

34 |
Content adaptive watermarking based on a stochastic multiresolution image modeling
- Voloshynovskiy, Deguillaume, et al.
- 2000
(Show Context)
Citation Context ...s to the global geometrical distortions rely on the use of either a transform invariant domain [48], or an additional template [50,10,51], or an Autocorrelation Function (ACF) of the watermark itself =-=[35,67]-=-. If robustness to global affine transformations is more or less a solved issue, the local random alterations integrated in Stirmark still remains an open problem almost for all techniques. The so cal... |

29 |
Removing spatial spread spectrum watermarks.
- Langelaar, Lagendijk, et al.
- 1998
(Show Context)
Citation Context ... tool. On the other hand, attacks based on remodulation are a relatively new concept unique to the watermarking attack problem. An efficient remodulation attack was first presented by Langelaar et al =-=[38,39]-=-. 4 In this scheme the watermark was predicted via subtraction of the median filtered version of stego image from the stego image itself. The predicted watermark was additionally high-pass filtered, t... |

28 | Robustness of a Blind Image Watermarking Scheme”,
- Eggers, Su, et al.
- 2000
(Show Context)
Citation Context ...ecoder can only randomly guess whether the received signal sample was originally quantized by quantizer assigned to the codeword 1 or to codeword 2. Thus, the watermark information is completely lost =-=[22]-=-. 7.2 Watermark copy attack The other type of watermark estimation-based attacks is the copy attack [37]. The main idea of this attack is to copy a watermark from one image to another image without kn... |

27 |
Improved watermark detection reliability using filtering before correlation,
- Depovere, Kalker, et al.
- 1998
(Show Context)
Citation Context ...roblem is connected with the assumption that the stego image is not significantly altered after attack. This allows the perceptual mask used at embedding to be estimated from the attacked stego image =-=[35,27,19]-=-. However this assumption does not hold for attacks connected with histogram modification that could have a significant influence on models based on luminance masking, and lossy JPEG compression attac... |

24 | Capacity of the watermark-channel: How many bits can be hidden within a digital image
- Barni, Bartolini, et al.
- 1999
(Show Context)
Citation Context ... attack can be obtained, when the embedding rule is known. We consider some generalized embedding rule (3). In practice, there are three mostly used embedding schemes: additive linear, multiplicative =-=[5,6,59]-=- and quantization index modulation (QIM) [9]. More generally, all of them could be considered from the point of view of some generalized linear additive scheme: y = x +H (x; wjM) (20) where H(:; :) is... |

24 | Smart Images Using Digimarc's Watermarking Technology," - Alattar - 2000 |

20 |
The impact of channel coding on the performance of spatial watermarking for copyright protection
- Hernández, Pérez-González, et al.
- 1998
(Show Context)
Citation Context ...using either error correction codes (ECC) or modulated using binary antipodal signaling [29] or M-ary modulation [35]. With respect to ECC, mostly Bose Chaudhuri (BCH) or convolutional codes are used =-=[48,28]-=-. Recent publications [52,67] report the successful results using novel Turbo codes and low-density paritycheck (LDPC) codes in the DCT and wavelet domains. In the general case, the type of ECC and th... |

20 |
Improved robust watermarking through attack characterization
- Kundur, Hatzinakos
- 1998
(Show Context)
Citation Context ...r produces a soft output that can be important for further decoding. The scheme with the hard output assuming binary symmetric channels was first proposed by Kundur et al in watermarking applications =-=[33]-=-. This scheme is considerably simplified and 21 does not require the estimation of the channel non-stationary variances and parameters of the fading. It is modeled using only error probablities for ea... |

16 |
Digital image watermarking: hiding information in images
- Kutter
- 1999
(Show Context)
Citation Context ...s to the global geometrical distortions rely on the use of either a transform invariant domain [48], or an additional template [50,10,51], or an Autocorrelation Function (ACF) of the watermark itself =-=[35,67]-=-. If robustness to global affine transformations is more or less a solved issue, the local random alterations integrated in Stirmark still remains an open problem almost for all techniques. The so cal... |

14 |
A copyright protection environment for digital images, Diploma Dissertation, Ecole Polytechnique Federal de
- Perrig
- 1997
(Show Context)
Citation Context ...on through an exhaustive search. Since many watermarking schemes use a secret key it is very important to use keys with a secure length. Another attack in this category is the so called Oracle attack =-=[54,14]-=- which can be used to create a non-watermarked image when a watermark detector device is available. 2.4 Protocol attacks The protocol attacks aim at attacking the concept of the watermarking applicati... |

12 | Countermeasures for unintentional and intentional video watermarking attacks
- Deguillaume, Csurka, et al.
- 2000
(Show Context)
Citation Context ...nces of the same data are available, but this time the attacked data set is generated by taking only a small part of each data set and rebuilding a new attacked data set from these parts. Deguillaume =-=[18]-=- considers the averaging and collusion attacks in application to videos and proposes corresponding countermeasures. The other type of attack that impairs the detection and decoding of the watermark is... |

9 | Image watermarking-a spread spectrum application,” in - Tirkel, Osborne, et al. - 1996 |

8 | Schyndel, “Image watermarking – a spread spectrum application - Tirkel, Osborne, et al. - 1996 |

6 | A framework for optimal adaptive dct watermarks using linear programming - Pereira, Pun - 2000 |

6 |
A generalized watermark attack based on stochastic watermark estimation and perceptual remodulation
- Voloshynovskiy, Pereira, et al.
- 2000
(Show Context)
Citation Context ...n Section 6.1. Lossy compression has recently been established to have roughly the same influence on noise removal as denoising [62,46]. This was reported with respect to watermarking applications in =-=[69]-=- and is considered in Section 6.2. Both the denoising and the lossy compression can significantly reduce the capacity of watermarking channel setting to the zero the output of many equivalent channels... |

5 | PeH rez-GonzaH lez, Statistical analysis of watermarking schemes for copyright protection of images - ndez, F - 1999 |

4 | Adaptive weighted myriad filter optimization for robust signal processing
- Kalluri, Arce
- 1996
(Show Context)
Citation Context ...and MAP estimates which are the most frequently encountered in practice. In the case of the ML, the well-known denoising algorithms are local mean (or average), median, trimmed mean and myriad filter =-=[31]-=- which are the estimates for a Gaussian, Laplacian, ffl\Gamma contaminated (mixture model of Gaussian and Laplacian), and Cauchy watermark distributions respectively. The averaging and median filters ... |

4 |
E!ective channel coding for DCT watermarks
- Pereira, Voloshynovskiy, et al.
- 2000
(Show Context)
Citation Context ...hniques. The derivation of the optimal MAP estimate for multiplicative watermarks or generally non-additive techniques is required to cover methods like SysCop of MediaSec [42], Barni [7] and Pereira =-=[52,53]-=- that are mostly used in the transform domains. Although the above classification makes it possible to have a clear separation between the different classes of attacks, it is necessary to note that ve... |

4 | Digital image watermarking: Issues in resolving rightful ownership - Ratakonda, Dugad, et al. - 1998 |

4 |
Robust image restoration based on concept of m-estimation and parametric model of image spectrum
- Voloshynovskiy
- 1998
(Show Context)
Citation Context ... is available, one can estimate the watermark directly from the stego image as was discussed above. If there is some ambiguity with respect to these priors, a robust M-estimation approach can be used =-=[66]-=-. Assuming that the image and watermark are conditionally i.i.d. locally Gaussian, i.e. xsN (x; R x ) and wsN (0; Rw ) with covariance matrices R x and Rw , where Rw also includes the effect of percep... |

4 | Fast robust template matching for a$ne resistant watermarks - Pereira, Pun - 1999 |

3 |
Parallel and deterministic algorithms from MRFs surface reconstruction
- Deiger, Girosi
- 1984
(Show Context)
Citation Context ... for the sGG model. In practice, iterative algorithms are often used to solve the above problem. Examples of such algorithms are the stochastic [23] and deterministic annealing (mean-field annealing) =-=[17]-=-, graduated nonconvexity [2], ARTUR algorithm [49] or its generalization [3]. Of course, it is preferable to obtain the closed form solution for the analysis of the obtained estimate. To generalize th... |

3 |
Watermark estimation through local pixel correlation
- Holliman, Memon, et al.
- 1999
(Show Context)
Citation Context ...stics/spectrum are matched with the image to guarantee a visual imperceptability, this attack shows poor performance. A similar attack based on weighted mean prediction was proposed by Holliman at al =-=[30] wher-=-e authors report their success in removal of watermarks produced by the watermarking scheme proposed by Pitas [57] and the commercial software of Digimark [10]. Su and Girod [64] propose a "Wiene... |

3 |
http://www.cl.cam.ac.uk/ fapp2/watermarking/stirmark
- Petitcolas
- 1999
(Show Context)
Citation Context ...ally arises. A lack of systematic benchmarking of existing methods however creates confusion amongst content providers and watermarking technology suppliers. Existing benchmarking tools like Stirmark =-=[55]-=- or Unzign [1] integrate a number of image processing operations or geometrical transformations aimed at removing watermarks from a stego image. However, the quality of the processed image is often to... |

2 |
Simultaneous Noise Supression and Signal Compression using a Library of
- Saito
- 1995
(Show Context)
Citation Context ...ng applications. A detailed analysis of the denoising methods is given in Section 6.1. Lossy compression has recently been established to have roughly the same influence on noise removal as denoising =-=[62,46]-=-. This was reported with respect to watermarking applications in [69] and is considered in Section 6.2. Both the denoising and the lossy compression can significantly reduce the capacity of watermarki... |

2 | Performance analysis of a 2-D-multipulse amplitude modulation scheme for data hiding and watermarking of still images - ndez, lez, et al. - 1998 |

1 |
smart images" using digimarc's watermarking technology
- Alattar
- 2000
(Show Context)
Citation Context ...pical choice for the projection function p is a set of two dimensional orthogonal functions used for every codeword bit fc k g such that the empty set is formed by the intersection P k "P l ; 8k =-=6= l [35,29,4]. The proj-=-ection function performs a "spreading" of the data over the image area. It can be also considered as diversity communication problem with parallel channels. Moreover, the projection function... |

1 |
Watermark removal software, available under a Windows95 at http://www-it.et.tudelft.nl/gerhard/waterm.zip
- Langelaar
- 1998
(Show Context)
Citation Context ... tool. On the other hand, attacks based on remodulation are a relatively new concept unique to the watermarking attack problem. An efficient remodulation attack was first presented by Langelaar et al =-=[38,39]-=-. 4 In this scheme the watermark was predicted via subtraction of the median filtered version of stego image from the stego image itself. The predicted watermark was additionally high-pass filtered, t... |

1 | PeH rez-GonzaH lez, DCTdomain watermarking techniques for still images: detector performance analysis and a new structure - ndez, Amado, et al. |

1 | The impact of channel coding on the performance of spatial watermarking for copyright protection - ndez, lez, et al. - 1998 |

1 | Adaptive weighted myriad "lter optimization for robust signal processing - Kalluri, Arce - 1996 |

1 | Power-spectrun condition for energye$cient watermarking - Su, Girod - 1999 |