This PDF file contains the front matter associated with SPIE Proceedings Volume 8303, including the Title Page, Copyright information, Table of Contents, and the Conference Committee listing.

We present two approaches to robust image obfuscation based on permutation of image regions and channel intensity modulation. The proposed concept of robust image obfuscation is a step towards end-to-end security in Web 2.0 applications. It helps to protect the privacy of the users against threats caused by internet bots and web applications that extract biometric and other features from images for data-linkage purposes. The approaches described in this paper consider that images uploaded to Web 2.0 applications pass several transformations, such as scaling and JPEG compression, until the receiver downloads them. In contrast to existing approaches, our focus is on usability, therefore the primary goal is not a maximum of security but an acceptable trade-off between security and resulting image quality.

Robustness against distortions caused by common image processing is one of the essential properties for image watermarking to be applicable in real-world applications. Typical distortions include lossy JPEG compression, filtering, cropping, scaling, rotation, and so on, among which geometric distortion is more challenging. Even slight geometric distortion can totally fail the watermark detection through de-synchronization. Another important property is the watermark payload. Although one-bit watermark is widely used in research work for algorithm testing and evaluation, only checking whether a specific watermark exists does not meet the requirement of many practical applications. This paper presents a practical robust image watermarking algorithm which combines template embedding and patchwork watermarking in Fourier domain. The embedded template enables the necessary robustness against geometric distortions and the patchwork approach provides a reasonable watermark payload which can meet the requirement of most applications. A spatial perceptual mask is used to reshape the embedded energy after it is inverted to the spatial domain, which significantly improves the image quality and enhances the robustness of both template and watermark. Implementation issues and solutions, e.g. fine-tuning of embedding energy of individual pixels, are also discussed. Experimental results demonstrate the effectiveness and practicability of the proposed algorithm.

Digital transaction watermarking today is a widely accepted mechanism to discourage illegal distribution of multimedia. The transaction watermark is a user-specific message that is embedded in all copies of one content and thus makes it individual. Therewith it allows to trace back copyright infringements. One major threat on transaction watermarking are collusion attacks. Here, multiple individualized copies of the work are compared and/or combined to attack the integrity or availability of the embedded watermark message. One solution to counter such attacks are mathematical codes called collusion secure fingerprinting codes. Problems arise when applying such codes to multimedia files with small payload, e.g. short audio tracks or images. Therefore the code length has to be shortened which increases the error rates and/or the effort of the tracing algorithm. In this work we propose an approach whether to use as an addition to probabilistic fingerprinting codes for a reduction of the effort and increment of security, as well as a new separate method providing shorter codes at a very fast and high accurate tracing algorithm.

Depth-image-based rendering (DIBR) is a method to represent a stereoscopic content. The DIBR consists of a monoscopic center view and an associated per-pixel depth map. Using these two components and given depth condition from a user, the DIBR renders left and right views. The advantages of DIBR are numerous. The user can choose not only the monoscopic or stereoscopic view selectively, but also the depth condition what he prefers when he watches a stereoscopic content. However, in the view of copyright protection, since not only the center view but also each left or right view can be used as a monoscopic content when they are illegally distributed, the watermark signal which is embedded in the center view must have an ability to protect the respective three views. In this study, we solve this problem by exploiting the horizontal noise mean shifting (HNMS) technique. We exploit the fact that the objects in the view are shifted only to horizontal way when the center view renders to the left and right views. Using this fact, the proposed stereoscopic watermarking scheme moves the mean of horizontal noise histogram which is invariant to horizontal shifting, and we achieve good performance as shown in the experimental results.

In this paper a new q-ary reversible high-capacity lossless scheme with a controllable prediction error is presented. The proposed scheme holds the advantages of embedding of more than one bit per pixel in a single run of an algorithm due to the utilization of secret data from Galois field, and avoidance of the redundant data, by deriving the special conditions. The new histogram shifting approach was elaborated to achieve higher capacity versus quality of the image compared to introduced by authors¹ data hiding counterpart based on difference expansion method. In order to reach high computation efficiency we introduce a new weighted simplified predictor. The comparison with the existing predictors in terms of computation efficiency and the amount of embedded payload is presented. Experimental part produces the comparison of the proposed scheme on different test images. The new technique is compared with a number of reversible methods, including our previous scheme¹ based on difference expansion, where it produces the better embedding capacity versus image quality performance. We also demonstrate the behavior of the schemes for various q via data rate versus quality curves. The proposed scheme not only holds the advantages of the location map free data embedding, but also enables high payload capacity.

A standard way to design steganalysis features for digital images is to choose a pixel predictor, use it to compute a noise residual, and then form joint statistics of neighboring residual samples (co-occurrence matrices). This paper proposes a general data-driven approach to optimizing predictors for steganalysis. First, a local pixel predictor is parametrized and then its parameters are determined by solving an optimization problem for a given sample of cover and stego images and a given cover source. Our research shows that predictors optimized to detect a specific case of steganography may be vastly different than predictors optimized for the cover source only. The results indicate that optimized predictors may improve steganalysis by a rather non-negligible margin. Furthermore, we construct the predictors sequentially - having optimized k predictors, design the k + 1st one with respect to the combined feature set built from all k predictors. In other words, given a feature space (image model) extend (diversify) the model in a selected direction (functional form of the predictor) in a way that maximally boosts detection accuracy.

In this paper, we propose a rich model of DCT coefficients in a JPEG file for the purpose of detecting steganographic embedding changes. The model is built systematically as a union of smaller submodels formed as joint distributions of DCT coefficients from their frequency and spatial neighborhoods covering a wide range of statistical dependencies. Due to its high dimensionality, we combine the rich model with ensemble classifiers and construct detectors for six modern JPEG domain steganographic schemes: nsF5, model-based steganography, YASS, and schemes that use side information at the embedder in the form of the uncompressed image: MME, BCH, and BCHopt. The resulting performance is contrasted with previously proposed feature sets of both low and high dimensionality. We also investigate the performance of individual submodels when grouped by their type as well as the effect of Cartesian calibration. The proposed rich model delivers superior performance across all tested algorithms and payloads.

The state of the art steganalytic features for spatial domain, and to some extent for transfer domains (CDT) as well, are based on histogram of co-occurances of neighboring elements. The rationale behind is that neighboring pixels in digital images are correlated, which is caused by the smoothness of our world and by the usual image processing. The limitation of the histogram-based features is that they do not scale well with respect to the number of modeled neighboring elements, since the number of histogram bins (hence number of features) depends exponentially on this quanitity. The remedy adopted by the prior art is to sum values of neighboring bins together, which can be seen as a vector quantization controlled by the position of the quantization centers. So far the quantization centers has been determined manually according to the intuition of the staganalyst. Heere we proposedto use Linde, Buso, and Gray algorithm in order to automatically find quantization centers maximizing the detection accuracy of resulting features. The quantization centers found by the proposed algorithm are experimentally compared to the ones used by the prior art on the steganalysis of Hugo algorithm. Tbhe results show a non-negligible improvements in the accuracy, especially when more complicated filters and higher-order histograms are used.

In the field of imaging device identification, the unique property of Photo-Response Non-Uniformity (PRNU) is widely employed. One of disadvantages of the PRNU based methods is sensitive to de-synchronization attacks. In this paper, we propose an improved PRNU based camcorder identification method which performs well with simultaneously cropped and scaled videos. The proposed method solves the out-of-sync problems by achieving downscale-tolerance of Minimum Average Correlation Energy Mellin Radial Harmonic (MACE-MRH) filter. The experimental results demonstrate that the proposed method identifies source devices faster and more accurate than the existing method.

Image display technology has greatly developed over the past few decades, which make it possible to recapture high-quality images from the display medium, such as a liquid crystal display(LCD) screen or a printed paper. The recaptured images are not regarded as a separate image class in the current research of digital image forensics, while the content of the recaptured images may have been tempered. In this paper, two sets of features based on the noise and the traces of double JPEG compression are proposed to identify these recaptured images. Experimental results showed that our proposed features perform well for detecting photographic copying.

Digital audio watermarking detection is often computational complex and requires at least as much audio information as required to embed a complete watermark. In some applications, especially real-time monitoring, this is an important drawback. The reason for this is the usage of sync sequences at the beginning of the watermark, allowing a decision about the presence only if at least the sync has been found and retrieved. We propose an alternative method for detecting the presence of a watermark. Based on the knowledge of the secret key used for embedding, we create a mark for all potential marking stages and then use a sliding window to test a given audio file on the presence of statistical characteristics caused by embedding. In this way we can detect a watermark in less than 1 second of audio.

Computational photography is quickly making its way from research labs to the market. Recently, camera manufacturers started using in-camera lens-distortion correction of the captured image to give users more powerful range of zoom in compact and affordable cameras. Since the distortion correction (barrel/pincushion) depends on the zoom, it desynchronizes the pixel-to-pixel correspondence between images taken at two different focal lengths. This poses a serious problem for digital forensic methods that utilize the concept of sensor fingerprint (photo-response non-uniformity), such as "image ballistic" techniques that can match an image to a specific camera. Such techniques may completely fail. This paper presents an extension of sensor-based camera identification to images corrected for lens distortion. To reestablish synchronization between an image and the fingerprint, we adopt a barrel distortion model and search for its parameter to maximize the detection statistic, which is the peak to correlation energy ratio. The proposed method is tested on hundreds of images from three compact cameras to prove the viability of the approach and demonstrate its efficiency.

We introduce an approach for verifying the integrity of digital audio recording by means of content-based integrity watermarking. Here an audio fingerprint is extracted from the Fourier domain and embedded as a digital watermark in the same domain. The design of the feature extraction allows a fine temporal resolution of the verification of the integrity. Experimental results show a good distinction between authentic and tampered audio content.

Fingerprints are used for the identification of individuals for over a century in crime scene forensics. Here, often physical or chemical preprocessing techniques are used to render a latent fingerprint visible. For quality assurance purposes of those development techniques, Schwarz1 introduces a technique for the reproducible generation of latent fingerprints using ink-jet printers and artificial amino acid sweat. However, this technique allows for printing latent fingerprints at crime scenes to leave false traces, too. Hence, Kiltz et al.2 introduce a first framework for the detection of printed fingerprints. However, the utilized printers have a maximum resolution of 2400×1200 dpi. In this paper, we use a Canon PIXMA iP46003 printer with a much higher resolution of 9600×400 dpi, which does not produce the kind of visible dot patterns reported in Kiltz et al.2 We show that an acquisition with a resolution of 12700 to 25400 ppi is necessary to extract microstuctures, which perspectively allows for an automated detection of printed fingerprint traces fabricated with high-resolution printers. Using our first test set with 20 printed and 20 real, natural fingerprint patterns from the human the evaluation results indicate a very positive tendency towards the detectability of such traces using the method proposed in this paper.

In this paper a methodology for digital image forgery detection by means of an unconventional use of image quality assessment is addressed. In particular, the presence of differences in quality degradations impairing the images is adopted to reveal the mixture of different source patches. The ratio behind this work is in the hypothesis that any image may be affected by artifacts, visible or not, caused by the processing steps: acquisition (i.e., lens distortion, acquisition sensors imperfections, analog to digital conversion, single sensor to color pattern interpolation), processing (i.e., quantization, storing, jpeg compression, sharpening, deblurring, enhancement), and rendering (i.e., image decoding, color/size adjustment). These defects are generally spatially localized and their strength strictly depends on the content. For these reasons they can be considered as a fingerprint of each digital image. The proposed approach relies on a combination of image quality assessment systems. The adopted no-reference metric does not require any information about the original image, thus allowing an efficient and stand-alone blind system for image forgery detection. The experimental results show the effectiveness of the proposed scheme.

With most image steganalysis traditionally based on supervised machine learning methods, the size of training data has remained static at up to 20000 training examples. This potentially leads to the classifier being undertrained for larger feature sets and it may be too narrowly focused on characteristics of a source of cover images, resulting in degradation in performance when the testing source is mismatched or heterogeneous. However it is not difficult to obtain larger training sets for steganalysis through simply taking more photos or downloading additional images. Here, we investigate possibilities for creating steganalysis classifiers trained on large data sets using large feature vectors. With up to 1.6 million examples, naturally simpler classification engines must be used and we examine the hypothesis that simpler classifiers avoid overtraining and so perform better on heterogeneous data. We highlight the possibilities of online learners, showing that, when given sufficient training data, they can match or exceed the performance of complex classifiers such as Support Vector Machines. This applies to both their accuracy and training time. We include some experiments, not previously reported in the literature, which provide benchmarks of some known feature sets and classifier combinations.

We consider the problem of universal pooled steganalysis, in which we aim to identify a steganographer who sends many images (some of them innocent) in a network of many other innocent users. The detector must deal with multiple users and multiple images per user, and particularly the differences between cover sources used by different users. Despite being posed for five years, this problem has only previously been addressed by our 2011 paper. We extend our prior work in two ways. First, we present experiments in a new, highly realistic, domain: up to 4000 actors each transmitting up to 200 images, real-world data downloaded from a social networking site. Second, we replace hierarchical clustering by the method called local outlier factor (LOF), giving greater accuracy of detection, and allowing a guilty actor sending moderate payloads to be detected, even amongst thousands of other actors sending hundreds of thousands of images.

Forensic analysis of image sets today is most often done with the help of cryptographic hashes due to their efficiency, their integration in forensic tools and their excellent reliability in the domain of false detection alarms. A drawback of these hash methods is their fragility to any image processing operation. Even a simple re-compression with JPEG results in an image not detectable. A different approach is to apply image identification methods, allowing identifying illegal images by e.g. semantic models or facing detection algorithms. Their common drawback is a high computational complexity and significant false alarm rates. Robust hashing is a well-known approach sharing characteristics of both cryptographic hashes and image identification methods. It is fast, robust to common image processing and features low false alarm rates. To verify its usability in forensic evaluation, in this work we discuss and evaluate the behavior of an optimized block-based hash.

This paper presents a traitor tracing method dedicated to video content distribution. It is based on a two-level approach with probabilistic traitor tracing codes. Codes are concatenated and decoded successively, the first one is used to decrease the decoding complexity and the second to accuse users. We use the well-known Tardos fingerprinting code for the accusation process and a Boneh-Shaw code with replication scheme to reduce the search space of users. This method ensures a decrease of the computational time compared to classical Tardos codes decoding. We present a method to select suspect groups of users and compare it to a more complex two-level Tardos code which follows the same construction.

Payload location using residuals is a successful approach to identify load-carrying pixels provided a large number of stego images are available. Furthermore, each image must have the payload embedded at the same locations. The success of payload location is therefore limited if different keys are used or an adaptive embedding algorithm is used. Given these limitations, the focus of this paper is to locate modified pixels in a single stego image. Given a sufficiently large set of independent binary decision functions, each determines whether a pixel has been modified better than guessing, we show that it is possible to locate modified pixels in a single stego image with low error rate. We construct these functions using existing cover estimators and provide experimental results to support our analysis.

Digital camcording in the premises of cinema theaters is the main source of pirate copies of newly released movies. To trace such recordings, watermarking systems are exploited in order for each projection to be unique and thus identifiable. The forensic analysis to recover these marks is different for digital and legacy cinemas. To avoid running both detectors, a reliable oracle discriminating between cams originating from analog or digital projections is required. This article details a classification framework relying on three complementary features : the spatial uniformity of the screen illumination, the vertical (in)stability of the projected image, and the luminance artifacts due to the interplay between the display and acquisition devices. The system has been tuned with cams captured in a controlled environment and benchmarked against a medium-sized dataset (61 samples) composed of real-life pirate cams. Reported experimental results demonstrate that such a framework yields over 80% classification accuracy.

In this paper, we extend an existing context model for statistical pattern recognition based microphone forensics by: first, generating a generalized model for this process and second, using this general model to construct a complex new application scenario model for microphone forensic investigations on the detection of playback recordings (a.k.a. replays, re-recordings, double-recordings). Thereby, we build the theoretical basis for answering the question whether an audio recording was made to record a playback or natural sound. The results of our investigations on the research question of playback detection imply that it is possible with our approach on our evaluation set of six microphones. If the recorded sound is not modified prior to playback, we achieve in our tests 89.00% positive indications on the correct two microphones involved. If the sound is post-processed (here, by normalization) this figure decreases (in our normalization example to 36.00%, while another 50.67% of the tests still indicate two microphones, of which one has actually not been involved in the recording and playback recording process).

This paper presents an approach to evaluate the acoustic path transmission of watermarked audio tracks through large scale simulations. The multitude of signal alterations performed implicitly via acoustic path transmission are aggregated through the measurement of impulse responses. These impulse responses are integrated in a test suite in order to be able to perform large scale automated tests as a replacement of the time intensive and expensive individual measurements. The reliability of the approach is demonstrated by the comparison of measurements and simulations.

Cryptographic techniques are used to secure confidential data from unauthorized access but these techniques are very sensitive to noise. A single bit change in encrypted data can have catastrophic impact over the decrypted data. This paper addresses the problem of removing bit error in visual data which are encrypted using AES algorithm in the CBC mode. In order to remove the noise, a method is proposed which is based on the statistical analysis of each block during the decryption. The proposed method exploits local statistics of the visual data and confusion/diffusion properties of the encryption algorithm to remove the errors. Experimental results show that the proposed method can be used at the receiving end for the possible solution for noise removing in visual data in encrypted domain.