W. S. Lin

Digital image source coder forensics via intrinsic fingerprints

Recent development in multimedia processing and network technologies has facilitated the distribution and sharing of multimedia through networks, and increased the security demands of multimedia contents. Traditional image content protection schemes use extrinsic approaches, such as watermarking or fingerprinting. However, under many circumstances, extrinsic content protection is not possible. Therefore, there is great interest in developing forensic tools via intrinsic fingerprints to solve these problems. Source coding is a common step of natural image acquisition, so in this paper, we focus on the fundamental research on digital image source coder forensics via intrinsic fingerprints. First, we investigate the unique intrinsic fingerprint of many popular image source encoders, including transform-based coding (both discrete cosine transform and discrete wavelet transform based), subband coding, differential image coding, and also block processing as the traces of evidence. Based on the intrinsic fingerprint of image source encoders, we construct an image source coding forensic detector that identifies which source encoder is applied, what the coding parameters are along with confidence measures of the result. Our simulation results show that the proposed system provides trustworthy performance: for most test cases, the probability of detecting the correct source encoder is over 90%.

Behavior modeling and forensics for multimedia social networks

Factors influencing human behavior have seldom appeared in signal processing disciplines. Therefore, the goals of the article are to illustrate why human factors are important, identify emerging issues strongly related to signal processing, and to demonstrate that signal processing can be effectively used to model, analyze, and perform behavior forensics for multimedia social networks. Since media security and content protection is a major issue, the article illustrates various aspects of issues and problems in multimedia social networks via a case study of human behavior in traitor-tracing multimedia fingerprinting. We focus on the understanding of behavior forensics from signal processing perspective and present a framework to model and analyze user dynamics. The objective is to provide a broad overview of recent advances in behavior modeling and forensics for multimedia social networks.

Transform Coder Classification for Digital Image Forensics

The area of non-intrusive forensic analysis has found many applications in the area of digital imaging. One unexplored area is the identification of source coding in digital images. In other words, given a digital image, can we identify which compression scheme was used, if any? This paper focuses on the aspect of transform coder classification, where we wish to determine which transform was used during compression. This scheme analyzes the histograms of coefficient subbands to determine the nature of the transform method. By obtaining the distance between the obtained histogram and the estimate of the original histogram, we can determine if the image was compressed using the transform tested. Results show that this method can successfully classify compression by transform as well as detect whether any compression has occurred at all in an image.

Block Size Forensic Analysis in Digital Images

In non-intrusive forensic analysis, we wish to find information and properties about a piece of data without any reference to the original data prior to processing. An important first step to forensic analysis is the detection and estimation of block processing. Most existing work in block measurement uses strong assumptions on the data related to the block size or the method of compression. In this paper, we propose a new method to estimate the block size in digital images in a blind manner for use in a forensic context. We make no assumptions on the block size or the nature of any previous processing. Our scheme can accurately estimate block sizes in images up to a PSNR of 42 dB where block artifacts are perceptually invisible. We also offer a measure of detection accuracy which correctly classifies an image as block-processed with a probability of 95.0% while keeping the probability of false alarm at 7.4%.

Extrinsic Channel-Like Fingerprint Embedding for Authenticating MIMO Systems

A framework for introducing an extrinsic fingerprint signal to space-time coded transmissions at the physical layer is presented, where the fingerprint signal conveys a low capacity cryptographically secure authentication message of arbitrary length. The multi-bit digital fingerprint message conveyed by the fingerprint signal is available to all users within reception range and is used to authenticate the fingerprinted transmission. A novel approach is discussed where the fingerprint signaling mechanism mimics distortions similar to time-varying channel effects. Specifically, the fingerprint is detectable to receivers considering previous channel state information, but will be ignored by receivers equalizing according to current channel state information. Two example fingerprint signaling mechanisms and detection rules are presented based on pulse-amplitude keying and phase-shift keying approaches. The methods for obtaining the real (intrinsic) channel estimate, the extrinsic fingerprint message, and the primary transmission are analytically demonstrated using general pilot embedding schemes. The worst-case distortions caused by non-ideal equalization of a fingerprinted message are derived using the 2×2 Alamouti code. Simulation results including bit error rate (BER) and model mismatch error using a maximum-likelihood (ML) receiver are presented for both the primary and fingerprint signal, while authentication signal BERs lower than the primary signal are demonstrated.

A Proximity-Aware Hierarchical Clustering of Faces

In this paper, we propose an unsupervised face clustering algorithm called “Proximity-Aware Hierarchical Clustering” (PAHC) that exploits the local structure of deep representations. In the proposed method, a similarity measure between deep features is computed by evaluating linear SVM margins. SVMs are trained using nearest neighbors of sample data, and thus do not require any external training data. Clus- ters are then formed by thresholding the similarity scores. We evaluate the clustering performance using three challenging un- constrained face datasets, including Celebrity in Frontal-Profile (CFP), IARPA JANUS Benchmark A (IJB-A), and JANUS Challenge Set 3 (JANUS CS3) datasets. Experimental results demonstrate that the proposed approach can achieve significant improvements over state-of-the-art methods. Moreover, we also show that the proposed clustering algorithm can be applied to curate a set of large-scale and noisy training dataset while maintaining sufficient amount of images and their variations due to nuisance factors. The face verification performance on JANUS CS3 improves significantly by finetuning a DCNN model with the curated MS-Celeb-1M dataset which contains over three million face images.

Extrinsic Channel-Like Fingerprinting Overlays Using Subspace Embedding

We present a physical-layer fingerprint-embedding scheme for orthogonal frequency-division-multiplexing (OFDM) transmissions, where the fingerprint signal conveys a low capacity communication suitable for authenticating the transmission and further facilitating secure communications. Our system strives to embed the fingerprint message into the noise subspace of the channel estimates obtained by the receiver, using a number of signal spreading techniques. When side information of the channel state is known and leveraged by the transmitter, the performance of the fingerprint embedding can be improved. When channel state information is not known, blind spreading techniques are applied. The fingerprint message is only visible to aware receivers who explicitly preform detection of the signal, but is invisible to receivers employing typical channel equalization. A taxonomy of overlay designs is discussed and these designs are explored through experiment using time-varying channel-state information (CSI) recorded from IEEE802.16e Mobile WiMax base stations. The performance of the fingerprint signal as received by a WiMax subscriber is demonstrated using CSI measurements derived from the downlink signal. Detection performance for the digital fingerprint message in time-varying channel conditions is also presented via simulation.

A Game Theoretic Framework for Colluder-Detector Behavior Forensics

Digital fingerprinting is an emerging technology in media security to identify the source of illicit copies and trace traitors. Collusion is a powerful attack, in which a group of attacker collectively mount attacks against digital fingerprinting. In multimedia fingerprinting, there exists complex dynamics between the colluders and the fingerprint detector, who have conflicting objectives and influence each others performance and decisions. This paper proposes a game-theoretic framework to formulate and analyze the colluder-detector dynamics, in an effort to understand its impact on the traitor tracing performance of multimedia fingerprints. We investigate how colluders adjusts the collusion attacks to minimize their risk under the fairness constraint; and study how the fingerprint detector adapts his/her detection strategy accordingly to improve the collusion resistance, which is shown to be the min-max solution.

Best-Effort Cooperative Relaying

Traditional cooperative communications consider dedicated-relays, while often such relays may not be available. In this paper, we consider wireless transceivers that relay signals in addition to their own primary communication mission. We consider a best-effort delivery policy, where a node is not obligated to devote energy to cooperatively relay signals, nor does it provide a guarantee of signal quality on retransmissions. Instead the relay sacrifices energy at its own discretion, with priority given to the primary communication mission. We consider one best-effort delivery problem: a system that transmits an additional relay signal within its original transmission energy budget while inducing minimal degradation to the primary-users signal. To maintain this constraint, we consider the feasibility of reallocating energy from pilot signals used for channel estimation toward the relaying service, when channel conditions are stationary. We demonstrate that transmitter energy may be dynamically allocated between a relay component and a pilot component of the transmission using best-effort delivery. This power allocation is critical to system performance, since both the primary-user and the secondary-user may require pilot energy to correctly decode transmitted signals. Sub-optimal power allocation rules with respect to primary-user channel estimate mean-square error and pairwise error probability are derived.

Proximity-Aware Hierarchical Clustering of unconstrained faces

Abstract In this paper, we propose an unsupervised face clustering algorithm called “Proximity-Aware Hierarchical Clustering” (PAHC) that exploits the local structure of deep representations. In the proposed method, a similarity measure between deep features is computed by evaluating linear SVM margins, which are learned using nearest neighbors of sample data. Clusters are then formed by applying agglomerative hierarchical clustering (AHC). We evaluate the clustering performance using four unconstrained face datasets, including Celebrity in Frontal-Profile (CFP), Labeled Faces in the Wild (LFW), IARPA JANUS Benchmark A (IJB-A), and IARPA JANUS Benchmark B (IJB-B) datasets. Experimental results demonstrate that the proposed approach can achieve improved performance over state-of-the-art methods. Moreover, we show the proposed clustering algorithm has the potential to be applied to actively learn robust deep face representations by first harvesting sufficient number of unseen face images through curation of large-scale dataset, e.g. the MS-Celeb-1 M dataset. By training DCNNs on the curated MS-Celeb-1 M dataset which contains over three million face images, improved representation for face images are learned.