Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

Digital watermarking is a key ingredient to copyright protection. It provides a solution to illegal copying of digital material and has many other useful applications such as broadcast monitoring and the recording of electronic transactions. Now, for the first time, there is a book that focuses exclusively on this exciting technology. Digital Watermarking covers the crucial research findings in the field: it explains the principles underlying digital watermarking technologies, describes the requirements that have given rise to them, and discusses the diverse ends to which these technologies are being applied. As a result, additional groundwork is laid for future developments in this field, helping the reader understand and anticipate new approaches and applications.

Digital Watermarking

Reversible data embedding has drawn lots of interest recently Being reversible, the original digital content can be completely restored We present a novel reversible data-embedding method for digital images We explore the redundancy in digital images to achieve very high embedding capacity, and keep the distortion low

Reversible data embedding using a difference expansion

Information-hiding techniques have recently become important in a number of application areas. Digital audio, video, and pictures are increasingly furnished with distinguishing but imperceptible marks, which may contain a hidden copyright notice or serial number or even help to prevent unauthorized copying directly. Military communications systems make increasing use of traffic security techniques which, rather than merely concealing the content of a message using encryption, seek to conceal its sender, its receiver, or its very existence. Similar techniques are used in some mobile phone systems and schemes proposed for digital elections. Criminals try to use whatever traffic security properties are provided intentionally or otherwise in the available communications systems, and police forces try to restrict their use. However, many of the techniques proposed in this young and rapidly evolving field can trace their history back to antiquity, and many of them are surprisingly easy to circumvent. In this article, we try to give an overview of the field, of what we know, what works, what does not, and what are the interesting topics for research.

/pdf/information-hiding-a-survey-wve0ipfu9n.pdf

Information hiding-a survey

A novel reversible data hiding algorithm, which can recover the original image without any distortion from the marked image after the hidden data have been extracted, is presented in this paper. This algorithm utilizes the zero or the minimum points of the histogram of an image and slightly modifies the pixel grayscale values to embed data into the image. It can embed more data than many of the existing reversible data hiding algorithms. It is proved analytically and shown experimentally that the peak signal-to-noise ratio (PSNR) of the marked image generated by this method versus the original image is guaranteed to be above 48 dB. This lower bound of PSNR is much higher than that of all reversible data hiding techniques reported in the literature. The computational complexity of our proposed technique is low and the execution time is short. The algorithm has been successfully applied to a wide range of images, including commonly used images, medical images, texture images, aerial images and all of the 1096 images in CorelDraw database. Experimental results and performance comparison with other reversible data hiding schemes are presented to demonstrate the validity of the proposed algorithm.

Reversible data hiding

Digital audio, video, images, and documents are flying through cyberspace to their respective owners. Unfortunately, along the way, individuals may choose to intervene and take this content for themselves. Digital watermarking and steganography technology greatly reduces the instances of this by limiting or eliminating the ability of third parties to decipher the content that he has taken. The many techiniques of digital watermarking (embedding a code) and steganography (hiding information) continue to evolve as applications that necessitate them do the same. The authors of this second edition provide an update on the framework for applying these techniques that they provided researchers and professionals in the first well-received edition. Steganography and steganalysis (the art of detecting hidden information) have been added to a robust treatment of digital watermarking, as many in each field research and deal with the other. New material includes watermarking with side information, QIM, and dirty-paper codes. The revision and inclusion of new material by these influential authors has created a must-own book for anyone in this profession.

*This new edition now contains essential information on steganalysis and steganography
*New concepts and new applications including QIM introduced
*Digital watermark embedding is given a complete update with new processes and applications

Digital Watermarking and Steganography

We describe a novel general strategy for building steganography detectors for digital images. The process starts with assembling a rich model of the noise component as a union of many diverse submodels formed by joint distributions of neighboring samples from quantized image noise residuals obtained using linear and nonlinear high-pass filters. In contrast to previous approaches, we make the model assembly a part of the training process driven by samples drawn from the corresponding cover- and stego-sources. Ensemble classifiers are used to assemble the model as well as the final steganalyzer due to their low computational complexity and ability to efficiently work with high-dimensional feature spaces and large training sets. We demonstrate the proposed framework on three steganographic algorithms designed to hide messages in images represented in the spatial domain: HUGO, edge-adaptive algorithm by Luo , and optimally coded ternary ±1 embedding. For each algorithm, we apply a simple submodel-selection technique to increase the detection accuracy per model dimensionality and show how the detection saturates with increasing complexity of the rich model. By observing the differences between how different submodels engage in detection, an interesting interplay between the embedding and detection is revealed. Steganalysis built around rich image models combined with ensemble classifiers is a promising direction towards automatizing steganalysis for a wide spectrum of steganographic schemes.

/pdf/rich-models-for-steganalysis-of-digital-images-4osa224324.pdf

Rich Models for Steganalysis of Digital Images

In this paper, we propose a new method for the problem of digital camera identification from its images based on the sensor's pattern noise. For each camera under investigation, we first determine its reference pattern noise, which serves as a unique identification fingerprint. This is achieved by averaging the noise obtained from multiple images using a denoising filter. To identify the camera from a given image, we consider the reference pattern noise as a spread-spectrum watermark, whose presence in the image is established by using a correlation detector. Experiments on approximately 320 images taken with nine consumer digital cameras are used to estimate false alarm rates and false rejection rates. Additionally, we study how the error rates change with common image processing, such as JPEG compression or gamma correction.

/pdf/digital-camera-identification-from-sensor-pattern-noise-2zx31ufisg.pdf

Digital camera identification from sensor pattern noise

Digital images are easy to manipulate and edit due to availability of powerful image processing and editing software. Nowadays, it is possible to add or remove important features from an image without leaving any obvious traces of tampering. As digital cameras and video cameras replace their analog counterparts, the need for authenticating digital images, validating their content, and detecting forgeries will only increase. Detection of malicious manipulation with digital images (digital forgeries) is the topic of this paper. In particular, we focus on detection of a special type of digital forgery – the copy-move attack in which a part of the image is copied and pasted somewhere else in the image with the intent to cover an important image feature. In this paper, we investigate the problem of detecting the copy-move forgery and describe an efficient and reliable detection method. The method may successfully detect the forged part even when the copied area is enhanced/retouched to merge it with the background and when the forged image is saved in a lossy format, such as JPEG. The performance of the proposed method is demonstrated on several forged images. 1. The Need for Detection of Digital Forgeries The availability of powerful digital image processing programs, such as PhotoShop, makes it relatively easy to create digital forgeries from one or multiple images. An example of a digital forgery is shown in Figure 1. As the newspaper cutout shows, three different photographs were used in creating the composite image: Image of the White House, Bill Clinton, and Saddam Hussein. The White House was rescaled and blurred to create an illusion of an out-of-focus background. Then, Bill Clinton and Saddam were cut off from two different images and pasted on the White House image. Care was taken to bring in the speaker stands with microphones while preserving the correct shadows and lighting. Figure 1 is, in fact, an example of a very realisticlooking forgery. Another example of digital forgeries was given in the plenary talk by Dr. Tomaso A. Poggio at Electronic Imaging 2003 in Santa Clara. In his talk, Dr. Poggio showed how engineers can learn the lip movements of any person from a short video clip and then digitally manipulate the lips to arbitrarily alter the spoken content. In a nice example, a video segment showing a TV anchor announcing evening news was altered to make the anchor appear singing a popular song instead, while preserving the match between the sound and lip movement. The fact that one can use sophisticated tools to digitally manipulate images and video to create non-existing situations threatens to diminish the credibility and value of video tapes and images presented as evidence in court independently of the fact whether the video is in a digital or analog form. To tamper an analogue video, one can easily digitize the analog video stream, upload it into a computer, perform the forgeries, and then save the result in the NTSC format on an ordinary videotape. As one can expect, the situation will only get worse as the tools needed to perform the forgeries will move from research labs to commercial software. Figure 1 Example of a digital forgery. Despite the fact that the need for detection of digital forgeries has been recognized by the research community, very few publications are currently available. Digital watermarks have been proposed as a means for fragile authentication, content authentication, detection of tampering, localization of changes, and recovery of original content [1]. While digital watermarks can provide useful information about the image integrity and its processing history, the watermark must be present in the image before the tampering occurs. This limits their application to controlled environments that include military systems or surveillance cameras. Unless all digital acquisition devices are equipped with a watermarking chip, it will be unlikely that a forgery-inthe-wild will be detectable using a watermark. It might be possible, but very difficult, to use unintentional camera “fingerprints” related to sensor noise, its color gamut, and/or its dynamic range to discover tampered areas in images. Another possibility for blind forgery detection is to classify textures that occur in natural images using statistical measures and find discrepancies in those statistics between different portions of the image ([2], [3]). At this point, however, it appears that such approaches will produce a large number of missed detections as well as false positives. In the next section, we introduce one common type of digital forgeries – the copy-move forgery – and show a few examples. Possible approaches to designing a detector are discussed in Section 3. In Section 4, we describe the detection method based on approximate block matching. This approach proved to be by far the most reliable and efficient. The method is tested in the last Section 5 on a few forgeries. In the same section, we summarize the paper and outline future research directions. 2. Copy-Move Forgery Because of the extraordinary difficulty of the problem and its largely unexplored character, the authors believe that the research should start with categorizing forgeries by their mechanism, starting with the simple ones, and analyzing each forgery type separately. In doing so, one will build a diverse Forensic Tool Set (FTS). Even though each tool considered separately may not be reliable enough to provide sufficient evidence for a digital forgery, when the complete set of tools is used, a human expert can fuse the collective evidence and hopefully provide a decisive answer. In this paper, the first step towards building the FTS is taken by identifying one very common class of forgeries, the Copy-Move forgery, and developing efficient algorithms for its detection. In a Copy-Move forgery, a part of the image itself is copied and pasted into another part of the same image. This is usually performed with the intention to make an object “disappear” from the image by covering it with a segment copied from another part of the image. Textured areas, such as grass, foliage, gravel, or fabric with irregular patterns, are ideal for this purpose because the copied areas will likely blend with the background and the human eye cannot easily discern any suspicious artifacts. Because the copied parts come from the same image, its noise component, color palette, dynamic range, and most other important properties will be compatible with the rest of the image and thus will not be detectable using methods that look for incompatibilities in statistical measures in different parts of the image. To make the forgery even harder to detect, one can use the feathered crop or the retouch tool to further mask any traces of the copied-and-moved segments. Examples of the Copy-Move forgery are given in Figures 2–4. Figure 2 is an obvious forgery that was created solely for testing purposes. In Figure 3, you can see a less obvious forgery in which a truck was covered with a portion of the foliage left of the truck (compare the forged image with its original). It is still not too difficult to identify the forged area visually because the original and copied parts of the foliage bear a suspicious similarity. Figure 4 shows another Copy-Move forgery that is much harder to identify visually. This image has been sent to the authors by a third party who did not disclose the nature or extent of the forgery. We used this image as a real-life test for evaluating our detection tools. A visual inspection of the image did not reveal the presence of anything suspicious. Figure 2 Test image “Hats”. Figure 3 Forged test image “Jeep” (above) and its original version (below). Figure 4 Test image “Golf” with an unknown original. 3. Detection of Copy-Move Forgery Any Copy-Move forgery introduces a correlation between the original image segment and the pasted one. This correlation can be used as a basis for a successful detection of this type of forgery. Because the forgery will likely be saved in the lossy JPEG format and because of a possible use of the retouch tool or other localized image processing tools, the segments may not match exactly but only approximately. Thus, we can formulate the following requirements for the detection algorithm: 1. The detection algorithm must allow for an approximate match of small image segments 2. It must work in a reasonable time while introducing few false positives (i.e., detecting incorrect matching areas). 3. Another natural assumption that should be accepted is that the forged segment will likely be a connected component rather than a collection of very small patches or individual pixels. In this section, two algorithms for detection of the Copy-Move forgery are developed – one that uses an exact match for detection and one that is based on an approximate match. Before describing the best approach based on approximate block matching that produced the best balance between performance and complexity, two other approaches were investigated – Exhaustive search and Autocorrelation. 3.1 Exhaustive search This is the simplest (in priciple) and most obvious approach. In this method, the image and its circularly shifted version (see Figure 5) are overlaid looking for closely matching image segments. Let us assume that xij is the pixel value of a grayscale image of size M×N at the position i, j. In the exhaustive search, the following differences are examined: | xij – xi+k mod(M) j+l mod(N) |, k = 0, 1, ..., M–1, l = 0, 1, ..., N–1 for all i and j. It is easy to see that comparing xij with its cyclical shift [k,l] is the same as comparing xij with its cyclical shift [k’,l’], where k’=M–k and l’=N–l. Thus, it suffices to inspect only those shifts [k,l] with 1≤ k ≤ M/2, 1≤ l ≤ N/2, thus cutting the computational complexity by a factor of 4. Figure 5 Test image “Lenna” and its circular shift. For each shift [k,l], the differences ∆xij = | xij – xi+k mod

/pdf/detection-of-copy-move-forgery-in-digital-images-4wm3z2a9ua.pdf

Detection of Copy-Move Forgery in Digital Images

Today, the most accurate steganalysis methods for digital media are built as supervised classifiers on feature vectors extracted from the media. The tool of choice for the machine learning seems to be the support vector machine (SVM). In this paper, we propose an alternative and well-known machine learning tool-ensemble classifiers implemented as random forests-and argue that they are ideally suited for steganalysis. Ensemble classifiers scale much more favorably w.r.t. the number of training examples and the feature dimensionality with performance comparable to the much more complex SVMs. The significantly lower training complexity opens up the possibility for the steganalyst to work with rich (high-dimensional) cover models and train on larger training sets-two key elements that appear necessary to reliably detect modern steganographic algorithms. Ensemble classification is portrayed here as a powerful developer tool that allows fast construction of steganography detectors with markedly improved detection accuracy across a wide range of embedding methods. The power of the proposed framework is demonstrated on three steganographic methods that hide messages in JPEG images.

/pdf/ensemble-classifiers-for-steganalysis-of-digital-media-14urqbpk2p.pdf

Jessica Fridrich

Papers

Digital Watermarking and Steganography

Rich Models for Steganalysis of Digital Images

Digital camera identification from sensor pattern noise

Detection of Copy-Move Forgery in Digital Images

Ensemble Classifiers for Steganalysis of Digital Media