This PDF file contains the front matter associated with SPIE Proceedings Volume 9409 including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

We describe a photo forensic technique based on detecting inconsistencies in lighting. This technique explicitly measures the 3-D lighting properties for individual people, objects, or surfaces in a single image. We show that with minimal training, an analyst can accurately specify 3-D shape in a single image from which 3-D lighting can be automatically estimated. A perturbation analysis on the estimated lighting is performed to yield a probabilistic measure of the location of the illuminating light. Inconsistencies in lighting within an image evidence photo tampering.

Detection of copy{move forgeries is one of the most actively researched topics in image forensics. It has been shown that so-called block-based methods give the highest pixel-wise accuracy for detecting copy{move forgeries. However, matching of block-based features can be computationally extremely demanding. Hence, the current predominant line of thought is that block-based algorithms are too slow to be applicable in practice. In this paper, we revisit the matching stage of block-based copy{move forgery detection methods. We propose an efficient approach for finding duplicate patterns of a given size in integer-valued input data. By design, we focus on the spatial relation of potentially duplicated elements. This allows us to locate copy{move forgeries via bit-wise operations, without expensive block comparisons in the feature space. Experimental investigation of different matching strategies shows that the proposed method has its benefits. However, on a broader scale, our experiments demonstrate that the performance of matching by lexicographic sorting might have been underestimated in previous work, despite its remarkable speed benefit on large images. In fact, in a practical setting, where accuracy and computational efficiency have to be balanced, lexicographic sorting may be considered the method of choice.

Biometric detectors for speaker identification commonly employ a statistical model for a subject’s voice, such as a Gaussian Mixture Model, that combines multiple means to improve detector performance. This allows a malicious insider to amend or append a component of a subject’s statistical model so that a detector behaves normally except under a carefully engineered circumstance. This allows an attacker to force a misclassification of his or her voice only when desired, by smuggling data into a database far in advance of an attack. Note that the attack is possible if attacker has access to database even for a limited time to modify victim’s model. We exhibit such an attack on a speaker identification, in which an attacker can force a misclassification by speaking in an unusual voice, and replacing the least weighted component of victim’s model by the most weighted competent of the unusual voice of the attacker’s model. The reason attacker make his or her voice unusual during the attack is because his or her normal voice model can be in database, and by attacking with unusual voice, the attacker has the option to be recognized as himself or herself when talking normally or as the victim when talking in the unusual manner. By attaching an appropriately weighted vector to a victim’s model, we can impersonate all users in our simulations, while avoiding unwanted false rejections.

The Digimarc® Barcode is a digital watermark applied to packages and variable data labels that carries GS1 standard GTIN-14 data traditionally carried by a 1-D barcode. The Digimarc Barcode can be read with smartphones and imaging-based barcode readers commonly used in grocery and retail environments. Using smartphones, consumers can engage with products and retailers can materially increase the speed of check-out, increasing store margins and providing a better experience for shoppers. Internal testing has shown an average of 53% increase in scanning throughput, enabling 100’s of millions of dollars in cost savings [1] for retailers when deployed at scale. To get to scale, the process of embedding a digital watermark must be automated and integrated within existing workflows. Creating the tools and processes to do so represents a new challenge for the watermarking community. This paper presents a description and an analysis of the workflow implemented by Digimarc to deploy the Digimarc Barcode at scale. An overview of the tools created and lessons learned during the introduction of technology to the market are provided.

In January 2014, Digimarc announced Digimarc® Barcode for the packaging industry to improve the check-out efficiency and customer experience for retailers. Digimarc Barcode is a machine readable code that carries the same information as a traditional Universal Product Code (UPC) and is introduced by adding a robust digital watermark to the package design. It is imperceptible to the human eye but can be read by a modern barcode scanner at the Point of Sale (POS) station. Compared to a traditional linear barcode, Digimarc Barcode covers the whole package with minimal impact on the graphic design. This significantly improves the Items per Minute (IPM) metric, which retailers use to track the checkout efficiency since it closely relates to their profitability. Increasing IPM by a few percent could lead to potential savings of millions of dollars for retailers, giving them a strong incentive to add the Digimarc Barcode to their packages. Testing performed by Digimarc showed increases in IPM of at least 33% using the Digimarc Barcode, compared to using a traditional barcode.

A method of watermarking print ready image data used in the commercial packaging industry is described. A significant proportion of packages are printed using spot colors, therefore spot colors needs to be supported by an embedder for Digimarc Barcode. Digimarc Barcode supports the PANTONE spot color system, which is commonly used in the packaging industry. The Digimarc Barcode embedder allows a user to insert the UPC code in an image while minimizing perceptibility to the Human Visual System (HVS). The Digimarc Barcode is inserted in the printing ink domain, using an Adobe Photoshop plug-in as the last step before printing. Since Photoshop is an industry standard widely used by pre-press shops in the packaging industry, a Digimarc Barcode can be easily inserted and proofed.

This paper presents a speed comparison between the use of Digimarc® Barcodes and the Universal Product Code (UPC) for customer checkout at point of sale (POS). The recently introduced Digimarc Barcode promises to increase the speed of scanning packaged goods at POS. When this increase is exploited by workforce optimization systems, the retail industry could potentially save billions of dollars. The Digimarc Barcode is based on Digimarc’s watermarking technology, and it is imperceptible, very robust, and does not require any special ink, material, or printing processes. Using an image-based scanner, a checker can quickly scan consumer packaged goods (CPG) embedded with the Digimarc Barcode without the need to reorient the packages with respect to the scanner. Faster scanning of packages saves money and enhances customer satisfaction. It reduces the length of the queues at checkout, reduces the cost of cashier labor, and makes self-checkout more convenient. This paper quantifies the increase in POS scanning rates resulting from the use of the Digimarc Barcode versus the traditional UPC. It explains the testing methodology, describes the experimental setup, and analyzes the obtained results. It concludes that the Digimarc Barcode increases number of items per minute (IPM) scanned at least 50% over traditional UPC.

This paper reports on the implementation of the Digimarc® Discover platform on Google Glass, enabling the reading of a watermark embedded in a printed material or audio. The embedded watermark typically contains a unique code that identifies the containing media or object and a synchronization signal that allows the watermark to be read robustly. The Digimarc Discover smartphone application can read the watermark from a small portion of printed image presented at any orientation or reasonable distance. Likewise, Discover can read the recently introduced Digimarc Barcode to identify and manage consumer packaged goods in the retail channel. The Digimarc Barcode has several advantages over the traditional barcode and is expected to save the retail industry millions of dollars when deployed at scale. Discover can also read an audio watermark from ambient audio captured using a microphone. The Digimarc Discover platform has been widely deployed on the iPad, iPhone and many Android-based devices, but it has not yet been implemented on a head-worn wearable device, such as Google Glass. Implementing Discover on Google Glass is a challenging task due to the current hardware and software limitations of the device. This paper identifies the challenges encountered in porting Discover to the Google Glass and reports on the solutions created to deliver a prototype implementation.

The possibility of forging latent fingerprints at crime scenes is known for a long time. Ever since it has been stated that an expert is capable of recognizing the presence of multiple identical latent prints as an indicator towards forgeries. With the possibility of printing fingerprint patterns to arbitrary surfaces using affordable ink- jet printers equipped with artificial sweat, it is rather simple to create a multitude of fingerprints with slight variations to avoid raising any suspicion. Such artificially printed fingerprints are often hard to detect during the analysis procedure. Moreover, the visibility of particular detection properties might be decreased depending on the utilized enhancement and acquisition technique. In previous work primarily such detection properties are used in combination with non-destructive high resolution sensory and pattern recognition techniques to detect fingerprint forgeries. In this paper we apply Benford's Law in the spatial domain to differentiate between real latent fingerprints and printed fingerprints. This technique has been successfully applied in media forensics to detect image manipulations. We use the differences between Benford's Law and the distribution of the most significant digit of the intensity and topography data from a confocal laser scanning microscope as features for a pattern recognition based detection of printed fingerprints. Our evaluation based on 3000 printed and 3000 latent print samples shows a very good detection performance of up to 98.85% using WEKA's Bagging classifier in a 10-fold stratified cross-validation.

In digital crime scene forensics, contactless non-destructive detection and acquisition of latent fingerprints by means of optical devices such as a high-resolution digital camera, confocal microscope, or chromatic white-light sensor is the initial step prior to destructive chemical development. The applicability of an optical sensor to digitalize latent fingerprints primarily depends on reflection properties of a substrate. Metallic painted surfaces, for instance, pose a problem for conventional sensors which make use of visible light. Since metallic paint is a semi-transparent layer on top of the surface, visible light penetrates it and is reflected off of the metallic flakes randomly disposed in the paint. Fingerprint residues do not impede light beams making ridges invisible. Latent fingerprints can be revealed, however, using ultraviolet light which does not penetrate the paint. We apply a UV-VIS spectroscope that is capable of capturing images within the range from 163 to 844 nm using 2048 discrete levels. We empirically show that latent fingerprints left behind on metallic painted surfaces become clearly visible within the range from 205 to 385 nm. Our proposed streakiness score feature determining the proportion of a ridge-valley pattern in an image is applied for automatic assessment of a fingerprint’s visibility and distinguishing between fingerprint and empty regions. The experiments are carried out with 100 fingerprint and 100 non-fingerprint samples.

We address the selection of fingerprint minutiae given a fingerprint ISO template. Minutiae selection plays a very important role when a secure element (i.e. a smart-card) is used. Because of the limited capability of computation and memory, the number of minutiae of a stored reference in the secure element is limited. We propose in this paper a comparative study of 6 minutiae selection methods including 2 methods from the literature and 1 like reference (No Selection). Experimental results on 3 fingerprint databases from the Fingerprint Verification Competition show their relative efficiency in terms of performance and computation time.

In digitized forensics the support of investigators in any manner is one of the main goals. Using conservative lifting methods, the detection of traces is done manually. For non-destructive contactless methods, the necessity for detecting traces is obvious for further biometric analysis. High resolutional 3D confocal laser scanning microscopy (CLSM) grants the possibility for a detection by segmentation approach with improved detection results. Optimal scan results with CLSM are achieved on surfaces orthogonal to the sensor, which is not always possible due to environmental circumstances or the surface's shape. This introduces additional noise, outliers and a lack of contrast, making a detection of traces even harder. Prior work showed the possibility of determining angle-independent classification models for the detection of latent fingerprints (LFP). Enhancing this approach, we introduce a larger feature space containing a variety of statistical-, roughness-, color-, edge-directivity-, histogram-, Gabor-, gradient- and Tamura features based on raw data and gray-level co-occurrence matrices (GLCM) using high resolutional data. Our test set consists of eight different surfaces for the detection of LFP in four different acquisition angles with a total of 1920 single scans. For each surface and angles in steps of 10, we capture samples from five donors to introduce variance by a variety of sweat compositions and application influences such as pressure or differences in ridge thickness. By analyzing the present test set with our approach, we intend to determine angle- and substrate-dependent classification models to determine optimal surface specific acquisition setups and also classification models for a general detection purpose for both, angles and substrates. The results on overall models with classification rates up to 75.15% (kappa 0.50) already show a positive tendency regarding the usability of the proposed methods for LFP detection on varying surfaces in non-planar scenarios.

Optical, nano-meter range, contactless, non-destructive sensor devices are promising acquisition techniques in crime scene trace forensics, e.g. for digitizing latent fingerprint traces. Before new approaches are introduced in crime investigations, innovations need to be positively tested and quality ensured. In this paper we investigate sensor reproducibility by studying different scans from four sensors: two chromatic white light sensors (CWL600/CWL1mm), one confocal laser scanning microscope, and one NIR/VIS/UV reflection spectrometer. Firstly, we perform an intra-sensor reproducibility testing for CWL600 with a privacy conform test set of artificial-sweat printed, computer generated fingerprints. We use 24 different fingerprint patterns as original samples (printing samples/templates) for printing with artificial sweat (physical trace samples) and their acquisition with contactless sensory resulting in 96 sensor images, called scan or acquired samples. The second test set for inter-sensor reproducibility assessment consists of the first three patterns from the first test set, acquired in two consecutive scans using each device. We suggest using a simple feature space set in spatial and frequency domain known from signal processing and test its suitability for six different classifiers classifying scan data into small differences (reproducible) and large differences (non-reproducible). Furthermore, we suggest comparing the classification results with biometric verification scores (calculated with NBIS, with threshold of 40) as biometric reproducibility score. The Bagging classifier is nearly for all cases the most reliable classifier in our experiments and the results are also confirmed with the biometric matching rates.

A steganographic file system is a secure file system whose very existence on a disk is concealed. Customarily, these systems hide an encrypted volume within unused disk blocks, slack space, or atop conventional encrypted volumes. These file systems are far from undetectable, however: aside from their ciphertext footprint, they require a software or driver installation whose presence can attract attention and then targeted surveillance. We describe a new steganographic operating environment that requires no visible software installation, launching instead from a concealed bootstrap program that can be extracted and invoked with a chain of common Unix commands. Our system conceals its payload within innocuous files that typically contain high-entropy data, producing a footprint that is far less conspicuous than existing methods. The system uses a local web server to provide a file system, user interface and applications through a web architecture.

The vast majority of steganographic schemes for digital images stored in the raster format limit the amplitude of embedding changes to the smallest possible value. In this paper, we investigate the possibility to further improve the empirical security by allowing the embedding changes in highly textured areas to have a larger amplitude and thus embedding there a larger payload. Our approach is entirely model driven in the sense that the probabilities with which the cover pixels should be changed by a certain amount are derived from the cover model to minimize the power of an optimal statistical test. The embedding consists of two steps. First, the sender estimates the cover model parameters, the pixel variances, when modeling the pixels as a sequence of independent but not identically distributed generalized Gaussian random variables. Then, the embedding change probabilities for changing each pixel by 1 or 2, which can be transformed to costs for practical embedding using syndrome-trellis codes, are computed by solving a pair of non-linear algebraic equations. Using rich models and selection-channel-aware features, we compare the security of our scheme based on the generalized Gaussian model with pentary versions of two popular embedding algorithms: HILL and S-UNIWARD.

This paper considers the research goal of dependable steganalysis: where false positives occur once in a million or less, and this rate is known with high precision. Despite its importance for real-world application, there has been almost no study of steganalysis which produces very low false positives. We test existing and novel classifiers for their low false-positive performance, using millions of images from Flickr. Experiments on such a scale require considerable engineering. Standard steganalysis classifiers do not perform well in a low false-positive regime, and we make new proposals to penalize false positives more than false negatives.

Current work on steganalysis for digital images is focused on the construction of complex handcrafted features. This paper proposes a new paradigm for steganalysis to learn features automatically via deep learning models. We novelly propose a customized Convolutional Neural Network for steganalysis. The proposed model can capture the complex dependencies that are useful for steganalysis. Compared with existing schemes, this model can automatically learn feature representations with several convolutional layers. The feature extraction and classification steps are unified under a single architecture, which means the guidance of classification can be used during the feature extraction step. We demonstrate the effectiveness of the proposed model on three state-of-theart spatial domain steganographic algorithms - HUGO, WOW, and S-UNIWARD. Compared to the Spatial Rich Model (SRM), our model achieves comparable performance on BOSSbase and the realistic and large ImageNet database.

With the powerful image editing tools available today, it is very easy to create forgeries without leaving visible traces. Boundaries between host image and forgery can be concealed, illumination changed, and so on, in a naive form of counter-forensics. For this reason, most modern techniques for forgery detection rely on the statistical distribution of micro-patterns, enhanced through high-level filtering, and summarized in some image descriptor used for the final classification. In this work we propose a strategy to modify the forged image at the level of micro-patterns to fool a state-of-the-art forgery detector. Then, we investigate on the effectiveness of the proposed strategy as a function of the level of knowledge on the forgery detection algorithm. Experiments show this approach to be quite effective especially if a good prior knowledge on the detector is available.

Over the past decade, a number of information forensic techniques have been developed to identify digital image manipulation and falsification. Recent research has shown, however, that an intelligent forger can use anti-forensic countermeasures to disguise their forgeries. In this paper, an anti-forensic technique is proposed to falsify the lateral chromatic aberration present in a digital image. Lateral chromatic aberration corresponds to the relative contraction or expansion between an image's color channels that occurs due to a lens's inability to focus all wavelengths of light on the same point. Previous work has used localized inconsistencies in an image's chromatic aberration to expose cut-and-paste image forgeries. The anti-forensic technique presented in this paper operates by estimating the expected lateral chromatic aberration at an image location, then removing deviations from this estimate caused by tampering or falsification. Experimental results are presented that demonstrate that our anti-forensic technique can be used to effectively disguise evidence of an image forgery.

Photo response noise uniformity (PRNU) based source attribution has proven to be a powerful technique in multimedia forensics. The increasing prominence of this technique, combined with its introduction as evidence in the court, brought with it the need for it to withstand anti-forensics. Although robustness under common signal processing operations and geometrical transformations have been considered as potential attacks on this technique, new adversarial settings that curtail the performance of this technique are constantly being introduced. Starting with an overview of proposed approaches to counter PRNU based source attribution, this work introduces photographic panoramas as one such approach and discusses how to defend against it.

In the context of stereo video, disparity-coherent watermarking has been introduced to provide superior robustness against virtual view synthesis, as well as to improve perceived fidelity. Still, a number of practical considerations have been overlooked and in particular the role of the underlying depth estimation tool on performances. In this article, we explore the interplay between various stereo video processing primitives and highlight a few take away lessons that should be accounted for to improve performances of future disparity-coherent watermarking systems. In particular, we highlight how lost correspondences during the stereo warping process impact watermark detection, thereby calling for innovative designs.

To read a watermark from printed images requires that the watermarking system read correctly after affine distortions. One way to recover from affine distortions is to add a synchronization signal in the Fourier frequency domain and use this synchronization signal to estimate the applied affine distortion. Using the Fourier Magnitudes one can estimate the linear portion of the affine distortion. To estimate the translation one must first estimate the phase of the synchronization signal and then use phase correlation to estimate the translation. In this paper we provide a new method to measure the phase of the synchronization signal using only the data from the complex Fourier domain. This data is used to compute the linear portion, so it is quite convenient to estimate the phase without further data manipulation. The phase estimation proposed in this paper is computationally simple and provides a significant computational advantage over previous methods while maintaining similar accuracy. In addition, the phase estimation formula gives a general way to interpolate images in the complex frequency domain.

Mobile object identification based on its visual features find many applications in the interaction with physical objects and security. Discriminative and robust content representation plays a central role in object and content identification. Complex post-processing methods are used to compress descriptors and their geometrical information, aggregate them into more compact and discriminative representations and finally re-rank the results based on the similarity geometries of descriptors. Unfortunately, most of the existing descriptors are not very robust and discriminative once applied to the various contend such as real images, text or noise-like microstructures next to requiring at least 500-1'000 descriptors per image for reliable identification. At the same time, the geometric re-ranking procedures are still too complex to be applied to the numerous candidates obtained from the feature similarity based search only. This restricts that list of candidates to be less than 1'000 which obviously causes a higher probability of miss. In addition, the security and privacy of content representation has become a hot research topic in multimedia and security communities. In this paper, we introduce a new framework for non- local content representation based on SketchPrint descriptors. It extends the properties of local descriptors to a more informative and discriminative, yet geometrically invariant content representation. In particular it allows images to be compactly represented by 100 SketchPrint descriptors without being fully dependent on re-ranking methods. We consider several use cases, applying SketchPrint descriptors to natural images, text documents, packages and micro-structures and compare them with the traditional local descriptors.

Diffractive Optically Variable Image Devices (DOVIDs), sometimes loosely referred to as holograms, are popular security features for protecting banknotes, ID cards, or other security documents. Inspection, authentication, as well as forensic analysis of these security features are still demanding tasks requiring special hardware tools and expert knowledge. Existing equipment for such analyses is based either on a microscopic analysis of the grating structure or a point-wise projection and recording of the diffraction patterns. We investigated approaches for an examination of DOVID security features based on sampling the Bidirectional Reflectance Distribution Function (BRDF) of DOVIDs using photometric stereo- and light-field-based methods. Our approach is demonstrated on the practical task of automated discrimination between genuine and counterfeited DOVIDs on banknotes. For this purpose, we propose a tailored feature descriptor which is robust against several expected sources of inaccuracy but still specific enough for the given task. The suggested approach is analyzed from both theoretical as well as practical viewpoints and w.r.t. analysis based on photometric stereo and light fields. We show that especially the photometric method provides a reliable and robust tool for revealing DOVID behavior and authenticity.

State-of-the-art JPEG steganographic algorithms, such as J-UNIWARD, are currently better detected in the spatial domain rather than the JPEG domain. Rich models built from pixel residuals seem to better capture the impact of embedding than features constructed as co-occurrences of quantized JPEG coefficients. However, when steganalyzing JPEG steganographic algorithms in the spatial domain, the pixels’ statistical properties vary because of the underlying 8 × 8 pixel grid imposed by the compression. In order to detect JPEG steganography more accurately, we split the statistics of noise residuals based on their phase w.r.t. the 8 × 8 grid. Because of the heterogeneity of pixels in a decompressed image, it also makes sense to keep the kernel size of pixel predictors small as larger kernels mix up qualitatively different statistics more, losing thus on the detection power. Based on these observations, we propose a novel feature set called PHase Aware pRojection Model (PHARM) in which residuals obtained using a small number of small-support kernels are represented using first-order statistics of their random projections as in the projection spatial rich model PSRM. The benefit of making the features “phase-aware” is shown experimentally on selected modern JPEG steganographic algorithms with the biggest improvement seen for J-UNIWARD. Additionally, the PHARM feature vector can be computed at a fraction of computational costs of existing projection rich models.

The cover source mismatch is a common problem in steganalysis, which may result in the degradation of detection accuracy. In this paper, we present a novel method to mitigate the problem of JPEG quantization table mismatch, named as Robust Discriminative Feature Transformation (RDFT). RDFT transforms original features to new feature representations based on a non-linear transformation matrix. It can improve the statistical consistency of the training samples and testing samples and learn new matched feature representations from original features by minimizing feature distribution difference while preserving the classification ability of training data. The comparison to prior arts reveals that the detection accuracy of the proposed RDFT algorithm can significantly outperform traditional steganalyzers under mismatched conditions and it is close to that of matched scenario. RDFT has several appealing advantages: 1) it can improve the statistical consistency of the training and testing data; 2) it can reduce the distribution difference between the training features and testing features; 3) it can preserve the classification ability of the training data; 4) it is robust to parameters and can achieve a good performance under a wide range of parameter values.

Color interpolation is a form of upsampling, which introduces constraints on the relationship between neighboring pixels in a color image. These constraints can be utilized to substantially boost the accuracy of steganography detectors. In this paper, we introduce a rich model formed by 3D co-occurrences of color noise residuals split according to the structure of the Bayer color filter array to further improve detection. Some color interpolation algorithms, AHD and PPG, impose pixel constraints so tight that extremely accurate detection becomes possible with merely eight features eliminating the need for model richification. We carry out experiments on non-adaptive LSB matching and the content-adaptive algorithm WOW on five different color interpolation algorithms. In contrast to grayscale images, in color images that exhibit traces of color interpolation the security of WOW is significantly lower and, depending on the interpolation algorithm, may even be lower than non-adaptive LSB matching.

We examine whether steganographic images can be detected more reliably when there exist other images, taken with the same camera under the same conditions, of the same scene. We argue that such a circumstance is realistic and likely in practice. In `laboratory conditions' mimicking circumstances favourable to the analyst, and with a custom set of digital images which capture the same scenes with controlled amounts of overlap, we use an overlapping reference image to calibrate steganographic features of the image under analysis. Experimental results show that the analysed image can be classified as cover or stego with much greater reliability than traditional steganalysis not exploiting overlapping content, and the improvement in reliability depends on the amount of overlap. These results are curious because two different photographs of exactly the same scene, taken only a few seconds apart with a fixed camera and settings, typically have steganographic features that differ by considerably more than a cover and stego image.