Show all abstracts
View Session
- Front Matter: Volume 7254
- Media Fingerprinting
- Steganography and Steganalysis
- Theoretical Foundations
- Media Forensics I
- Media Forensics II
- Cryptographic Techniques for Content Protection
- Video
- Data Hiding
- Watermarking
- Authentication
- Miscellaneous
Front Matter: Volume 7254
Front Matter: Volume 7254
Show abstract
This PDF file contains the front matter associated with SPIE
Proceedings Volume 7254, including the Title Page, Copyright
information, Table of Contents, and the Conference Committee listing.
Media Fingerprinting
Video fingerprinting for copy identification: from research to industry applications
Show abstract
Research that began a decade ago in video copy detection has developed into a technology known as "video
fingerprinting". Today, video fingerprinting is an essential and enabling tool adopted by the industry for video content
identification and management in online video distribution. This paper provides a comprehensive review of video
fingerprinting technology and its applications in identifying, tracking, and managing copyrighted content on the Internet.
The review includes a survey on video fingerprinting algorithms and some fundamental design considerations, such as
robustness, discriminability, and compactness. It also discusses fingerprint matching algorithms, including complexity
analysis, and approximation and optimization for fast fingerprint matching. On the application side, it provides an
overview of a number of industry-driven applications that rely on video fingerprinting. Examples are given based on
real-world systems and workflows to demonstrate applications in detecting and managing copyrighted content, and in
monitoring and tracking video distribution on the Internet.
Video fingerprinting for live events
Show abstract
Multimedia fingerprinting (robust hashing) as a content identification technology is emerging as an effective tool
for preventing unauthorized distribution of commercial content through user generated content (UGC) sites.
Research in the field has mainly considered content types with slow distribution cycles, e.g. feature films, for
which reference fingerprint ingestion and database indexing can be performed offline. As a result, research focus
has been on improving the robustness and search speed.
Live events, such as live sports broadcasts, impose new challenges on a fingerprinting system. For instance,
highlights from a soccer match are often available-and viewed-on UGC sites well before the end of the match.
In this scenario, the fingerprinting system should be able to ingest and index live content online and offer
continuous search capability, where new material is identifiable within minutes of broadcast. In this paper, we
concentrate on algorithmic and architectural challenges we faced when developing a video fingerprinting solution
for live events. In particular, we discuss how to effectively utilize fast sorting algorithms and a master-slave
architecture for fast and continuous ingestion of live broadcasts.
Machine-assisted editing of user-generated content
Show abstract
Over recent years user-generated content has become ubiquitously available and an attractive entertainment source for
millions of end-users. Particularly for larger events, where many people use their devices to capture the action, a great
number of short video clips are made available through appropriate web services. The objective of this presentation is to
describe a way to combine these clips by analyzing them, and automatically reconstruct the time line in which the
individual video clips were captured. This will enable people to easily create a compelling multimedia experience by
leveraging multiple clips taken by different users from different angles, and across different time spans. The user will be
able to shift into the role of a movie director mastering a multi-camera recording of the event. To achieve this goal, the
audio portion of the video clips is analyzed, and waveform characteristics are computed with high temporal granularity
in order to facilitate precise time alignment and overlap computation of the user-generated clips. Special care has to be
given not only to the robustness of the selected audio features against ambient noise and various distortions, but also to
the matching algorithm used to align the user-generated clips properly.
Image and video fingerprinting: forensic applications
Show abstract
Fighting movie piracy often requires automatic content identification. The most common technique to achieve this uses
watermarking, but not all copyrighted content is watermarked. Video fingerprinting is an efficient alternative solution to
identify content, to manage multimedia files in UGC sites or P2P networks and to register pirated copies with master
content. When registering by matching copy fingerprints with master ones, a model of distortion can be estimated. In
case of in-theater piracy, the model of geometric distortion allows the estimation of the capture location. A step even
further is to determine, from passive image analysis only, whether different pirated versions were captured with the same
camcorder. In this paper we present three such fingerprinting-based forensic applications: UGC filtering, estimation of
capture location and source identification.
Robust efficient video fingerprinting
Show abstract
We have developed a video fingerprinting system with robustness and efficiency as the primary and secondary design
criteria. In extensive testing, the system has shown robustness to cropping, letter-boxing, sub-titling, blur, drastic
compression, frame rate changes, size changes and color changes, as well as to the geometric distortions often associated
with camcorder capture in cinema settings. Efficiency is afforded by a novel two-stage detection process in which a fast
matching process first computes a number of likely candidates, which are then passed to a second slower process that
computes the overall best match with minimal false alarm probability. One key component of the algorithm is a
maximally stable volume computation - a three-dimensional generalization of maximally stable extremal regions - that
provides a content-centric coordinate system for subsequent hash function computation, independent of any affine
transformation or extensive cropping. Other key features include an efficient bin-based polling strategy for initial
candidate selection, and a final SIFT feature-based computation for final verification. We describe the algorithm and its
performance, and then discuss additional modifications that can provide further improvement to efficiency and accuracy.
A review of video fingerprints invariant to geometric attacks
Show abstract
Video fingerprints can help us identify a large amount of video on the Internet and enable interesting services
to the end user. One of the main challenges for video fingerprints is for them to be robust against intentional/
unintentional geometric modifications on the content such as scaling, aspect ratio conversion, rotation
and cropping. In this paper, we review a number of fingerprinting methods proposed in literature that are particularly
designed to be robust against such modifications. We also present two approaches that we adopted. One
that is based on estimation of Singular Value Decomposition (SVD) bases from a window of past video frames
(Approach 1) and another that is based on extraction of moment invariant features from concentric circular
regions and doesn't require any specific transform (Approach 2). While both approaches provide the desired
robustness against geometric modifications, Approach 1 is computationally more intensive than Approach 2 as
the SVD bases are updated for every input frame at 12fps. It also requires a longer query clip than Approach 2
for reliable identification. We present results comparing the performance of both of these approaches for a 150hr
video database.
Steganography and Steganalysis
The square root law of steganographic capacity for Markov covers
Show abstract
It is a well-established result that steganographic capacity of perfectly secure stegosystems grows linearly with
the number of cover elements-secure steganography has a positive rate. In practice, however, neither the
Warden nor the Steganographer has perfect knowledge of the cover source and thus it is unlikely that perfectly
secure stegosystems for complex covers, such as digital media, will ever be constructed. This justifies study of
secure capacity of imperfect stegosystems. Recent theoretical results from batch steganography, supported by
experiments with blind steganalyzers, point to an emerging paradigm: whether steganography is performed in a
large batch of cover objects or a single large object, there is a wide range of practical situations in which secure
capacity rate is vanishing. In particular, the absolute size of secure payload appears to only grow with the square
root of the cover size. In this paper, we study the square root law of steganographic capacity and give a formal
proof of this law for imperfect stegosystems, assuming that the cover source is a stationary Markov chain and
the embedding changes are mutually independent.
The impact of information fusion in steganalysis on the example of audio steganalysis
Show abstract
Information fusion tries to determine the best set of experts in a given problem domain and devise an appropriate function that can optimally combine the decisions of the individual experts. Only few systematic approaches to information fusion exist so far in the signal processing field of steganalysis.
Under the basic assumption that steganalysis can be seen as a statistical pattern recognition process like biometrics, a state of the art five level information fusion model known from biometrics is transferred to steganalysis as well as statistical detectability evaluations for watermarking algorithms and its applicability is evaluated in practical testing.
The primary test goal for these evaluations is to measure the impact of fusion on the classification accuracy. Therefore a match and decision level fusion are performed here for three selected data hiding algorithms (one steganography and two watermarking), two feature extractors and five different classifiers. For the test heterogeneous audio test sets are used for content independent training and testing. The secondary test goal of this work is to consider the impact of the key selection assumption on the accuracy of the classification in steganalysis.
The results show for the test cases an increase of the classification accuracy for two of the three tested algorithms by match level fusions, no gain by decision level fusion and a considerably small impact of the key selection assumption on the statistical detectability.
Feature reduction and payload location with WAM steganalysis
Show abstract
WAM steganalysis is a feature-based classifier for detecting LSB matching steganography, presented in 2006 by
Goljan et al. and demonstrated to be sensitive even to small payloads. This paper makes three contributions
to the development of the WAM method. First, we benchmark some variants of WAM in a number of sets of
cover images, and we are able to quantify the significance of differences in results between different machine
learning algorithms based on WAM features. It turns out that, like many of its competitors, WAM is not
effective in certain types of cover, and furthermore it is hard to predict which types of cover are suitable for
WAM steganalysis. Second, we demonstrate that only a few the features used in WAM steganalysis do almost
all of the work, so that a simplified WAM steganalyser can be constructed in exchange for a little less detection
power. Finally, we demonstrate how the WAM method can be extended to provide forensic tools to identify the
location (and potentially content) of LSB matching payload, given a number of stego images with payload placed
in the same locations. Although easily evaded, this is a plausible situation if the same stego key is mistakenly
re-used for embedding in multiple images.
Model-based steganalysis using invariant features
Show abstract
One of the biggest challenges in universal steganalysis is the identification of reliable features that can be used
to detect stego images. In this paper, we present a steganalysis method using features calculated from a measure
that is invariant for cover images and is altered for stego images. We derive this measure, which is the ratio
of any two Fourier coefficients of the distribution of the DCT coefficients, by modeling the distribution of the
DCT coefficients as a Laplacian. We evaluate our steganalysis detector against three different pixel-domain
steganography techniques.
From blind to quantitative steganalysis
Show abstract
Quantitative steganalyzers are important in forensic steganalysis
as they can estimate the payload, or, more precisely, the number of
embedding changes in the stego image. This paper proposes a general
method for constructing quantitative steganalyzers from features used
in blind detectors. The method is based on support vector regression,
which is used to learn the mapping between a feature vector extracted
from the image and the relative embedding change rate. The performance is evaluated by constructing quantitative steganalyzers for eight steganographic methods for JPEG files, using a 275-dimensional feature set. Error distributions of within- and between-image errors are empirically estimated for Jsteg and nsF5. For Jsteg, the accuracy is compared to state-of-the-art quantitative steganalyzers.
Theoretical Foundations
Conception and limits of robust perceptual hashing: towards side information assisted hash functions
Show abstract
In this paper, we consider some basic concepts behind the design of existing robust perceptual hashing techniques for content identification. We show the limits of robust hashing from the communication perspectives as well as propose an approach that is able to overcome these shortcomings in certain setups. The consideration is based on both achievable rate and probability of error. We use the fact that most robust hashing algorithms are
based on dimensionality reduction using random projections and quantization. Therefore, we demonstrate the corresponding achievable rate and probability of error based on random projections and compare with the results for the direct domain. The effect of dimensionality reduction is studied and the corresponding approximations are provided based on the Johnson-Lindenstrauss lemma. Side-information assisted robust perceptual hashing is proposed as a solution to the above shortcomings.
Expectation maximization decoding of Tardos probabilistic fingerprinting code
Show abstract
This paper presents our recent works on multimedia fingerprinting, which consists in improving both the fingerprinting code and the watermarking scheme. The fingerprinting code is the well known Tardos code. Our contributions only focus on deriving a better accusation process.
It appears that Tardos orginal decoding is very conservative: its performances are guaranteed whatever the collusion strategy. Indeed, major improvements stems from the knowledge of the collusion strategy. Therefore, this paper investigates whether it is possible to learn and adapt to the collusion strategy. This done with an iterative algorithm a la EM, where a better estimation of their strategy yields a better tracing of the colluders, which in return yields a better estimation of their strategy etc.
The second part focuses on the multimedia watermarking scheme. In a previous paper, we already used the `Broken Arrows' technique as the watermarking layer for multimedia fingerprinting. However, a recent paper from A. Westfeld discloses a flaw in this technique. We just present a counter-measure which blocks this security hole while preserving the robustness of the original technique.
Joint detection-estimation games for sensitivity analysis attacks
Show abstract
Sensitivity analysis attacks aim at estimating a watermark from multiple observations of the detector's output.
Subsequently, the attacker removes the estimated watermark from the watermarked signal. In order to measure
the vulnerability of a detector against such attacks, we evaluate the fundamental performance limits for the
attacker's estimation problem. The inverse of the Fisher information matrix provides a bound on the covariance
matrix of the estimation error. A general strategy for the attacker is to select the distribution of auxiliary test
signals that minimizes the trace of the inverse Fisher information matrix. The watermark detector must trade off
two conflicting requirements: (1) reliability, and (2) security against sensitivity attacks. We explore this tradeoff
and design the detection function that maximizes the trace of the attacker's inverse Fisher information matrix
while simultaneously guaranteeing a bound on the error probability. Game theory is the natural framework to
study this problem, and considerable insights emerge from this analysis.
On security threats for robust perceptual hashing
Show abstract
Perceptual hashing has to deal with the constraints of robustness, accuracy and security. After modeling the
process of hash extraction and the properties involved in this process, two different security threats are studied,
namely the disclosure of the secret feature space and the tampering of the hash. Two different approaches for
performing robust hashing are presented: Random-Based Hash (RBH) where the security is achieved using a
random projection matrix and Content-Based Hash (CBH) were the security relies on the difficulty to tamper the
hash. As for digital watermarking, different security setups are also devised: the Batch Hash Attack, the Group
Hash Attack, the Unique Hash Attack and the Sensitivity Attack. A theoretical analysis of the information
leakage in the context of Random-Based Hash is proposed. Finally, practical attacks are presented: (1) Minor
Component Analysis is used to estimate the secret projection of Random-Based Hashes and (2) Salient point
tampering is used to tamper the hash of Content-Based Hashes systems.
Media Forensics I
Large scale test of sensor fingerprint camera identification
Show abstract
This paper presents a large scale test of camera identification from sensor fingerprints. To overcome the problem of
acquiring a large number of cameras and taking the images, we utilized Flickr, an existing on-line image sharing site. In
our experiment, we tested over one million images spanning 6896 individual cameras covering 150 models. The
gathered data provides practical estimates of false acceptance and false rejection rates, giving us the opportunity to
compare the experimental data with theoretical estimates. We also test images against a database of fingerprints,
simulating thus the situation when a forensic analyst wants to find if a given image belongs to a database of already
known cameras. The experimental results set a lower bound on the performance and reveal several interesting new facts
about camera fingerprints and their impact on error analysis in practice. We believe that this study will be a valuable
reference for forensic investigators in their effort to use this method in court.
Detection and description of geometrically transformed digital images
Show abstract
Geometric transformations such as scaling or rotation are common tools employed by forgery creators. These
procedures are typically based on a resampling and interpolation step. The interpolation process brings specific
periodic properties into the image. In this paper, we show how to detect these properties. Our aim is to detect
all possible geometric transformations in the image being investigated. Furthermore, as the proposed method, as
well as other existing detectors, is sensitive to noise, we also briefly show a simple method capable of detecting
image noise inconsistencies. Noise is a common tool used to conceal the traces of tampering.
Synthesis of color filter array pattern in digital images
Show abstract
We propose a method to synthetically create or restore typical color filter array (CFA) pattern in digital images.
This can be useful, inter alia, to conceal traces of manipulation from forensic techniques that analyze the CFA
structure of images. For continuous signals, our solution maintains optimal image quality, using a quadratic
cost function; and it can be computed efficiently. Our general approach allows to derive even more efficient
approximate solutions that achieve linear complexity in the number of pixels. The effectiveness of the CFA
synthesis as tamper-hiding technique and its superior image quality is backed with experimental evidence on
large image sets and against state-of-the-art forensic techniques. This exposition is confined to the most relevant
'Bayer'-grid, but the method can be generalized to other layouts as well.
Media Forensics II
Lighting analysis of diffusely illuminated tableaus in realist paintings: an application to detecting "compositing" in the portraits of Garth Herrick
Show abstract
The problem of estimating the direction of point-source illumination in digital photographs has been studied
extensively, and the cast-shadow and occluding-contour algorithms have been used to detect tampering and
compositing; differences between the lighting directions estimated from different objects indicate that at least
one of them was composited into the image. Such methods have also been applied to the analysis of realist
paintings to estimate the position of illuminants within a tableau and thereby test for artists' use of optical
aids. Recently, the occluding-contour algorithm has been enhanced to address the case of diffuse illumination,
for instance from light passing through several windows, from multiple lamps, and so forth. Here, the pattern
of lightness along the occluding contour of an object is expressed as a weighted sum of spherical harmonics.
Significant differences between the coefficients extracted from different objects indicates that they were recorded
under different illumination conditions, and thus that one or more was likely composited into the image. We
apply this technique to the analysis of diffuse lighting in realist paintings, focussing on the portraits of the
contemporary American realist Garth Herrick. Herrick often works with multiple photographs as referents, for
instance a photograph of the portrait subject and a different photograph of the background. There is no guarantee
that the two lighting conditions are the same, nor that Herrick can perceive or compensate for such lighting
discrepancies when executing his painting. We tested for lighting consistency throughout two of his paintings:
one based on a single photographic referent, and another "composited," i.e., based on two photographic referents.
Our algorithms found great illumination consistency in the first painting and significant inconsistencies in the
second painting-inconsistencies difficult to discern by eye. As such, our methods reveal this artist's working
methods. Our algorithms have broad applicability to the study of studio practice throughout the history of art.
A study of the robustness of PRNU-based camera identification
Show abstract
We investigate the robustness of
PRNU-based camera identification in cases where the test
images have been passed through common image processing
operations. We address the issue of whether current camera
identification systems remain effective in the presence of a
nontechnical, mildly evasive photographer who makes efforts
at circumvention using only standard and/or freely available
software. We study denoising, recompression, out-of-camera
demosaicing.
Cryptographic Techniques for Content Protection
Robust fingerprinting codes and applications
Show abstract
We discuss the notion of fingerprinting systems for digital objects. We focus on the two-layered concatenating approach in the design of such systems that relies on a suitable notion of equivalence between vector objects. We point to the fact that the primitive of robust fingerprinting codes put forth recently by Boneh and Naor
can be useful in the design of general fingerprinting systems as it can be employed over a relaxed notion of object equivalence. We then present a complete analysis that demonstrates that the original Tardos' codes distribution coupled with a suitable tracing algorithm possesses some level of robustness for sufficiently small
rates of adversarial signal loss.
Practical tracing traitors
Show abstract
In this paper we discuss tracing traitors systems, with a focus on a particular problem we have encountered in building a
commercial system. "Tracing traitors" is a cryptographic technology that determines, when an unauthorized copy of
copyrighted content is encountered, which user or users were the source of the copy by examining the copy itself.
In tracing traitors systems, it has been widely assumed that any two devices in the system should have as few
cryptographic keys in common as possible: then, when the variation the key decrypts is discovered in the unauthorized
copy, the number of devices that could have produced that variation is minimal. This assumption is so pervasive it often
is not even stated explicitly. However, tracing traitors schemes also often ignore the likely next step-once the
compromised device(s) are identified, their keys must be revoked so they cannot be further used in the system.
In this paper we will show that the traceability of any minimal-key-overlap system inevitably begins to degrade once
some of the keys are revoked. This has caused us to question the basic value of minimal key overlap. In fact, we think
that very revocation-efficient key schemes, like broadcast-encryption key trees, in which some keys are highly shared,
might actually provide better traceability over the life of a practical system with actual revocation.
Next generation DRM: cryptography or forensics?
Show abstract
Current content protection systems rely primarily on applied cryptographic techniques but there is an increased use of
forensic solutions in images, music and video distribution alike. The two approaches differ significantly, both in terms
of technology and in terms of strategy, and thus it begs the question: will one approach take over in the long run, and if
so which one? Discussing the evolution of both cryptographic and forensic solutions, we conclude that neither approach
is ideal for all constituents, and that in the video space at least they will continue to co-exist for the foreseeable future -
even if this may not be the case for other media types. We also analyze shortcomings of these approaches, and suggest
that new solutions are necessary in this still emerging marketplace.
Stream it to me! Why is it challenging?
Show abstract
We describe design and deployment infrastructure considerations for streamed delivery of multimedia. The
system addresses constraints introduced by variations in broadband network capacity, a rich variety of
playback devices, and content protection objectives. Streamed distribution, playback processing, and
conditions for resource consumption are described.
The constraints and design implications of streamed distribution and downloaded distribution of media are
described and compared. Server resources, client resources, and broadband connectivity considerations are
described, with the objective of supporting computer platforms and low-cost consumer electronics.
For playback of multimedia content we describe the use cases for full and partial media views, and the
digital rights considerations for authorized access from multiple playback devices. We describe client
device and server state interactions to support the resumed playback use case.
For time-based conditions and limitations for streamed media, we describe design considerations for
calculating resource availability, rights grants, and the customer experience for media interaction as
authorization windows reach their conclusion.
Broadcast encryption: paving the road to practical content protection systems
Show abstract
Broadcast encryption is a well established alternative to public key encryption for use in content protection systems. It
offers significant performance benefits, as well as useful features such a one-to-many delivery, dynamic membership in
the authorized receivers group, and provides anonymous access to content, permitting content protection systems to
preserve privacy for consumers. Broadcast encryption has been successfully deployed to users for protection of
commercial content on digital media such as flash memory devices and optical media for both standard-definition and
high-definition content. In this paper, we present the Advanced Secure Content Cluster Technology which applies
broadcast encryption to content protection for home media networks
Video
Exhibition QIM-based watermarking for digital cinema
Show abstract
The copyright protection of Digital Cinema requires the insertion of forensic watermarks during exhibition playback.
This paper presents a low-complexity exhibition watermarking method based on quantization index modulation (QIM)
and embedded in the DCI compliant decoder. Watermark embedding is proposed to fit in the JPEG2000 decoding
process, prior to the inverse wavelet transform and such as it has a minimal impact on the image quality, guarantying a
strong link between decompression and watermarking. The watermark is embedded by using an adaptive-Spread
Transform Dither Modulation (STDM) method, based on a new multi-resolution perceptual masking to adapt watermark
strength. Watermark detection is thereafter performed over the wavelet transformation of the recovered images. The
proposed approach offers a wide range of channel capacities according to robustness to several kinds of distortions while
maintaining a low computational complexity. Watermarking detection performance on Digital Cinema pictures captured
with a video camera from a viewing room has been preliminary assessed, showing very promising results. The proposed
approach provides high levels of imperceptibility, yet good robustness to degradations resulting from camcorder
exhibition capture, to common signal processing operations such as filtering or re-sampling, and to very high
compression.
Digital watermarking for digital cinema
Show abstract
Changing the distribution channel of movies from analogue to digital provides new perspectives and possibilities for
applying digital watermarking as security mechanisms. Digital watermarking provides its best security options when
used for individual marking of copies to trace back distribution leaks. For analogue copies, protecting movies with
individual watermarks was only cost-effective for small sets of copies as in promotional copies as duplication is done in
an automated photo-optical way. Modifications of the analogue copies only provide relatively crude approaches of
individual marking. Digital copies on the other hand can easily be modified before, during or even after distribution to
cinemas. It is also possible to embed watermarks at all of these stages to secure the complete distribution chain from
studio to cinema. In our paper we discuss two watermarking strategies suitable for different applications in the digital
cinema domain, video and still image watermarking. We provide technical background and also discuss the interference
of applying both algorithms at the same time.
Data Hiding
Protocols for data hiding in pseudo-random state
Show abstract
An emerging form of steganographic communication uses ciphertext to replace the output of a random or strong
pseudo-random number generator. PRNG-driven media, for example computer animated backdrops in video-conferencing
channels, can then be used as a covert channel, if the PRNG bits that generated a piece of content
can be estimated by the recipient.
However, all bits sent over such a channel must be computationally indistinguishable from i.i.d. coin flips. Ciphertext
messages and even key exchange datagrams are easily shaped to match this distribution; however, when
placing these messages into a continous stream of PRNG bits, the sender is unable to provide synchronization
markers, metadata, or error correction to ensure the message's location and proper decoding.
In this paper we explore methods for message transmission and steganographic key exchange in such a "coin
flip" channel. We establish that key exchange is generally not possible in this channel if an adversary possesses
even a modest noise budget. If the warden is not vigilant in adding noise, however, communication is very simple.
High-capacity data hiding in text documents
Show abstract
In today's digital world securing different forms of content is very important in terms of protecting copyright
and verifying authenticity. One example is watermarking of digital audio and images. We believe that a marking
scheme analogous to digital watermarking but for documents is very important. In this paper we describe the
use of laser amplitude modulation in electrophotographic printers to embed information in a text document.
In particular we describe an embedding and detection process which has the capability to embed 14 bits into
characters that have a left vertical edge. For a typical 12 point document this translates to approximately 12000
bits per page.
Watermarking
Digital watermarking opportunities enabled by mobile media proliferation
Show abstract
Consumer usages of mobile devices and electronic media are changing. Mobile devices now include increased
computational capabilities, mobile broadband access, better integrated sensors, and higher resolution screens. These
enhanced features are driving increased consumption of media such as images, maps, e-books, audio, video, and games.
As users become more accustomed to using mobile devices for media, opportunities arise for new digital watermarking
usage models. For example, transient media, like images being displayed on screens, could be watermarked to provide a
link between mobile devices. Applications based on these emerging usage models utilizing watermarking can provide
richer user experiences and drive increased media consumption. We describe the enabling factors and highlight a few of
the usage models and new opportunities. We also outline how the new opportunities are driving further innovation in
watermarking technologies. We discuss challenges in market adoption of applications based on these usage models.
H.264/AVC substitution watermarking: a CAVLC example
Show abstract
This work addresses the watermarking of an entropy coded H.264/AVC video stream. The phrase Substitution
Watermarking is used to imply that the application of the watermark to the stream is accomplished by substituting an
original block of bits in the entropy-encoded stream with an alternative block of bits. This substitution is done for many
different blocks of bits to embed the watermark. This can be a particularly powerful technique for applications in which
the embedder must be very simple (substitution is a very light operation) and a computationally complex, pre-embedding
analysis is practical. The pre-embedding analysis can generate a substitution table and the embedder can simply select
entries from the table based on the payload. This paper presents the framework along with an example for H.264/AVC
streams that use CAVLC for entropy coding. A separate paper addresses the CABAC entropy coding case.
Real-time application of digital watermarking to embed tactical metadata into full motion video captured from unmanned aerial systems
Show abstract
A persistent challenge with imagery captured from Unmanned Aerial Systems (UAS), is the loss of
critical information such as associated sensor and geospatial data, and prioritized routing information
(i.e., metadata) required to use the imagery effectively. Often, there is a loss of synchronization between
data and imagery. The losses usually arise due to the use of separate channels for metadata, or due to
multiple imagery formats employed in the processing and distribution workflows that do not preserve
the data. To contend with these issues and provide another layer of authentication, digital watermarks
were inserted at point of capture within a tactical UAS. Implementation challenges included
traditional requirements surrounding, image fidelity, performance, payload size, robustness and
application requirements such as power consumption, digital to analog conversion and a fixed
bandwidth downlink, as well as a standard-based approach to geospatial exploitation through a serviceoriented-
architecture (SOA) for extracting and mapping mission critical metadata from the video
stream. The authors capture the application requirements, implementation trade-offs and ultimately
analysis of selected algorithms. A brief summary of results is provided from multiple test flights onboard
the SkySeer test UAS in support of Command, Control, Communications, Computers,
Intelligence, Surveillance and Reconnaissance applications within Network Centric Warfare and
Future Combat Systems doctrine.
Binary forensic code for multimedia signals: resisting minority collusion attacks
Show abstract
Digital forensic marking is a technology to discourage unauthorized redistribution of multimedia signals by embedding a
unique mark into each user's copy of the content. A powerful class of attacks on forensic marking is the collusion attack
by a group of users. Recently, a new collusion attack, called the minority attack, has been proposed against forensic
marking schemes with correlation-based detectors. Although this attack is not very effective on Gaussian-based forensic
marking, it is quite powerful on removing the traces of users when the forensic marking is binary. In this paper, we first
study the performance of an ECC-based binary forensic code under the minority attack and we model the additional
processing, such as compression, applied on colluded copy as a binary symmetric channel. We confirm that the system
can be defeated by a minority attack from only 3 colluders. To resist the minority attack, we propose a row-permuted
binary orthogonal code to serve as the inner code for ECC-based forensic code, coupled with an adaptive detector.
Experimental results show that the proposed scheme has a significantly improved resistance to a minority attack.
Authentication
Authenticating cropped and resized images using distributed source coding and expectation maximization
Show abstract
Media authentication is important in content delivery via untrusted intermediaries, such as peer-to-peer (P2P) file sharing. Many differently encoded versions of a media file might exist. Our previous work applied distributed source coding not only to distinguish the legitimate diversity of encoded images from tampering but also localize the tampered regions in an image already deemed to be inauthentic. The authentication data supplied to the
decoder consisted of a Slepian-Wolf encoded image projection.
We extend our scheme to authenticate cropped and resized images using an Expectation Maximization algorithm. Experimental results demonstrate that the proposed algorithm can distinguish legitimate encodings of authentic cropped and resized images from illegitimately modified versions using authentication data of less than 250 bytes.
Random projections based item authentication
Show abstract
In this paper, we consider an item authentication using unclonable forensic features of item surface microstructure images (a.k.a. fingerprints). The advocated authentication approach is based on the source coding jointly with the random projections. The source coding ensures the source reconstruction at the decoder based on the authentication data. The random projections are used to cope with the security, privacy, robustness and complexity issues. Finally, the authentication is accomplished as a binary hypothesis testing for both direct and random projections domains. The asymptotic performance approximation is derived and compared with the exact solutions.
Perception-based authentication watermarking for digital audio data
Show abstract
Current systems and protocols based on cryptographic methods for integrity and authenticity verification of media
data do not distinguish between legitimate signal transformation and malicious tampering that manipulates the
content. Furthermore, they usually provide no localization or assessment of the relevance of such manipulations
with respect to human perception or semantics. We present an algorithm for a authentication audio watermarking
that uses a perception-based robust hash function in combination with robust watermarking to verify the integrity
of audio recordings. Experimental results show that the proposed system provides both a high level of distinction
between perceptually different audio data and a high robustness against signal transformations that do not change
the perceived information.
Improvements on image authentication and recovery using distributed source coding
Show abstract
This paper investigates the performance and proposes modifications to earlier methods for image authentication
using distributed source coding. This approach works well on images that have undergone affine geometric
transformations such as rotation and resizing and intensity transformations such as contrast and brightness
adjustment. The results show that the improvements proposed here can be used to make the original scheme for
image authentication robust to affine geometric and intensity transformations. The modifications are of much
lesser computational complexity when compared with other schemes for estimation of channel parameters.
Miscellaneous
Nested object watermarking
Show abstract
Annotation watermarking is an application of watermarking where information about a cover or a part thereof are
embedded in the cover itself to link both directly together. In earlier work we introduced Nested Object Annotation
Watermarking as a special case, where the semantic, shape and hierarchical relations between the depicted nested objects
are embedded in the area of each object only. As these regions can be anywhere and may be composed of any shape
there is very limited a-priori knowledge for synchronization, which results in a higher complexity and ultimately in a
higher error-proneness. In general an exhaustive search strategy for the proper blocks to reconstruct the shape suffers
from the intrinsic combinatorial explosion of this process. Therefore in earlier work, at first we focused on rectangular
embedding schemes with a block luminance algorithm, a steganographic WetPaperCode algorithm, a rectangular and
finally a polygonal Dual-Domain DFT algorithm.
In this paper we review and compare these algorithms in terms of their transparency, capacity, shape fidelity and
robustness against the selected aspects JPEG compression and cropping. For the DFT algorithm we also show the
influence of several parameters, present our idea of a method to reduce the combinatorial explosion by collating the
paths in the search tree, and show that our new synchronization approach surpasses our former rectangular method in
terms of correct retrievals, despite its higher complexity.
High capacity color barcodes using dot orientation and color separability
Show abstract
Barcodes are widely utilized for embedding data in printed format to provide automated identification and
tracking capabilities in a number of applications. In these applications, it is desirable to maximize the number
of bits embedded per unit print area in order to either reduce the area requirements of the barcodes or to
offer an increased payload, which in turn enlarges the class of applications for these barcodes. In this paper,
we present a new high capacity color barcode. Our method operates by embedding independent data in two
different printer colorant channels via halftone-dot orientation modulation. In the print, the dots of the two
colorants occupy the same spatial region. At the detector, however, by using the complementary sensor channels
to estimate the colorant channels we can recover the data in each individual colorant channel. The method
therefore (approximately) doubles the capacity of encoding methods based on a single colorant channel and
provides an embedding rate that is higher than other known barcode alternatives. The effectiveness of the
proposed technique is demonstrated by experiments conducted on Xerographic printers. Data embedded at
a high density by using the two cyan and yellow colorant channels for halftone dot orientation modulation is
successfully recovered by using the red and blue channels for the detection, with an overall symbol error rate
that is quite small.
Enabling search over encrypted multimedia databases
Show abstract
Performing information retrieval tasks while preserving data confidentiality is a desirable capability when a
database is stored on a server maintained by a third-party service provider. This paper addresses the problem
of enabling content-based retrieval over encrypted multimedia databases. Search indexes, along with multimedia
documents, are first encrypted by the content owner and then stored onto the server. Through jointly applying
cryptographic techniques, such as order preserving encryption and randomized hash functions, with image
processing and information retrieval techniques, secure indexing schemes are designed to provide both privacy
protection and rank-ordered search capability. Retrieval results on an encrypted color image database and security
analysis of the secure indexing schemes under different attack models show that data confidentiality can
be preserved while retaining very good retrieval performance. This work has promising applications in secure
multimedia management.
RF device forensics using passband filter analysis
Show abstract
Given the wide use of Radio Frequency (RF) devices for applications ranging from data networks to wireless
sensors, it is of interest to be able to characterize individual devices to verify compliance with FCC Part 15 rules.
In an effort to characterize these types of devices we have developed a system that utilizes specially designed
probe signals to elicit a response from the device from which unique characteristics can be extracted. The
features that uniquely characterize a device are referred to as device signatures or device fingerprints.
We apply this approach to RF devices which employ different bandpass filters, and construct training based
classifiers which are highly accurate. We also introduce a model-based framework for optimal detection that can
be employed to obtain performance limits, and to study model mismatch and probe optimization.