MPS ’18- Proceedings of the 2nd International Workshop on Multimedia Privacy and Security
SESSION: Keynote Address
Artificial intelligence is increasingly employed in security-critical systems, such as autonomous cars and drones. Unfortunately, many machine learning techniques suffer from vulnerabilities that enable an adversary to thwart their successful application, either during the training or prediction phase. In this talk, we investigate this threat and discuss attacks against machine learning, such as ad- versarial perturbations and data poisoning. Surprisingly, several of the attacks are not entirely novel, and similar concepts have been developed independently for attacking digital watermarks in multimedia security. We review these similarities and provide links between the two research areas that may open new directions for improving both, machine learning and multimedia security.
SESSION: Internet of Things and Cloud-based Services
The internet-of-things (IoT) consists of embedded devices and their networks of communication as they form decentralized frameworks of ubiquitous computing services. Within such decentralized systems the potential for malicious actors to impact the system is significant, with far-reaching consequences. Hence this work addresses the challenge of providing IoT systems engineers with a framework to elicit privacy and security design considerations, specifically for indoor adaptive smart environments. It introduces a new ambient intelligence indoor adaptive environment framework (CORE) which leverages multiple forms of data, and aims to elicit the privacy and security needs of this representative system. This contributes both a new adaptive IoT framework, but also an approach to systematically derive privacy and security design requirements via a combined and modified OCTAVE-Allegro and Privacy-by-Design methodology. This process also informs the future developments and evaluations of the CORE system, toward engineering more secure and private IoT systems.
Personal Voice Assistants (PVAs) such as the Amazon Echo are commonplace and it is now likely to always be in range of at least one PVA. Although the devices are very helpful they are also continuously monitoring conversations. When a PVA detects a wake word, the immediately following conversation is recorded and transported to a cloud system for further analysis. In this paper we investigate an active protection mechanism against PVAs: reactive jamming. A Protection Jamming Device (PJD) is employed to observe conversations. Upon detection of a PVA wake word the PJD emits an acoustic jamming signal. The PJD must detect the wake word faster than the PVA such that the jamming signal still prevents wake word detection by the PVA. The paper presents an evaluation of the effectiveness of different jamming signals. We quantify the impact of jamming signal and wake word overlap on jamming success. Furthermore, we quantify the jamming false positive rate in dependence of the overlap. Our evaluation shows that a 100% jamming success can be achieved with an overlap of at least 60% with a negligible false positive rate. Thus, reactive jamming of PVAs is feasible without creating a system perceived as a noise nuisance.
Our devices and interactions in a world where physical and digital realities are more and more blended, generate a continuum of multimedia data that needs to be stored, shared and processed to provide services that enrich our daily lives. Cloud computing plays a key role in these tasks, dissolving resource allocation and computational boundaries, but it also requires advanced security mechanisms to protect the data and provide privacy guarantees. Therefore, security assurance must be evaluated before offloading tasks to a cloud provider, a process which is currently manual, complex and inadequate for dynamic scenarios. However, though there are many tools for evaluating cloud providers according to quality of service criteria, automated categorization and selection based on risk metrics is still challenging. To address this gap, we present FRiCS, a Framework for Risk-driven Cloud Selection, which contributes with: 1) a set of cloud security metrics and risk-based weighting policies, 2) distributed components for metric extraction and aggregation, and 3) decision-making plugins for ranking and selection. We have implemented the whole system and conducted a case-study validation based on public cloud providers’ security data, showing the benefits of the proposed approach.
SESSION: Steganography, Steganalysis, and Watermarking
This paper presents a new general framework of information hiding, in which the hidden information is embedded into a collection of activities conducted by selected human and computer entities (e.g., a number of online accounts of one or more online social networks) in a selected digital world. Different from other traditional schemes, where the hidden information is embedded into one or more selected or generated cover objects, in the new framework the hidden information is embedded in the fact that some particular digital activities with some particular attributes took place in some particular ways in the receiver-observable digital world. In the new framework the concept of “cover” almost disappears, or one can say that now the whole digital world selected becomes the cover. The new framework can find applications in both security (e.g., steganography) and non-security domains (e.g., gaming). For security applications we expect that the new framework calls for completely new steganalysis techniques, which are likely more complicated, less effective and less efficient than existing ones due to the need to monitor and analyze the whole digital world constantly and in real time. A proof-of-concept system was developed as a mobile app based on Twitter activities to demonstrate the information hiding framework works. We are developing a more hybrid system involving several online social networks.
The rise of social networks during the last 10 years has created a situation in which up to 100 million new images and photographs are uploaded and shared by users every day. This environment poses a ideal background for those who wish to communicate covertly by the use of steganography. It also creates a new set of challenges for steganalysts, who have to shift their field of work away from a purely scientific laboratory environment and into a diverse real-world scenario, while at the same time having to deal with entirely new problems, such as the detection of steganographic channels or the impact that even a low false positive rate has when investigating the millions of images which are shared every day on social networks. We evaluate how to address these challenges with traditional steganographic and statistical methods, rather then using high performance computing and machine learning. By the double embedding attack on the well-known F5 steganographic algorithm we achieve a false positive rate well below known attacks.
This paper proposes a reversible watermarking method that embeds binary bits into a digital image by gradient analysis, prediction value computation, two-step embedding process and difference expansion. The gradient analysis is introduced to detect whether a horizontal or vertical edge exists in the pixel context which would improve the accuracy of the prediction value. The two-step embedding process also aims at accurate prediction value computation. Since the prediction error is the key factor in the embedding process, the lower of the prediction error, the better the watermarked image quality. Experimental results show a higher percentage of zeros in the prediction error distribution histogram. Compared with other state-of-the-art reversible watermarking methods, better image quality can be realized by proposed method.
Mechanism and Implementation of Watermarked Sample Scanning Method for Speech Data Tampering Detection
The integrity and reliability of speech data have been important issues to probative use. Watermarking technologies supplies an alternative solution to guarantee the the authenticity of multiple data besides digital signature. This work proposes a novel digital watermarking based on a reversible compression algorithm with sample scanning to detect tampering in time domain. In order to detect tampering precisely, the digital speech data is divided into length-fixed frames and the content-based hash information of each frame is calculated and embedded into the speech data for verification. Huffman compression algorithm is applied to each four sampling bits from least significant bit in each sample after pulse-code modulation processing to achieve low distortion and high capacity for hiding payload. Experimental experiments on audio quality, detection precision and robustness towards attacks are taken, and the results show the effectiveness of tampering detection with a precision with an error around 0.032 s for a 10 s speech clip. Distortion is imperceptible with an average 22.068 dB for Huffman-based and 24.139 dB for intDCT-based method in terms of signal-to-noise, and with an average MOS 3.478 for Huffman-based and 4.378 for intDCT-based method. The bit error rate (BER) between stego data and attacked stego data in both of time-domain and frequency domain is approximate 28.6% in average, which indicates the robustness of the proposed hiding method.
SESSION: Communication and Data Privacy and Integrity
ETERNAL: Encrypted Transmission With an Error-correcting, Real-time, Noise-resilient Apparatus on Lightweight Devices
In this work, we describe the design and implementation of a private-key voice encryption system that is designed to encrypt and decrypt voice communications between two people using lightweight computational devices (such as a Raspberry Pi) that sits between the headset and the communication platform (computer, phone, etc.). The key challenge in this work is designing lightweight encryption algorithms in such a way that even before voice enters the phone/computer platform, voice is encrypted, yet such that modern audio communication channels such as popular VoIP applications (such as Skype, Google Voice, etc.), or mobile communications (GSM, etc.) or other (Radio, etc.) do not filter encrypted voice out as “noise” and voice quality is preserved. Thus, two people with two such devices can communicate securely even if their smart phones and/or computers are compromised. Unlike previous solutions, our proposed work does not rely on special-purpose hardware, nor does it rely on trusting the communication device. It is a standalone solution that can be readily deployed on lightweight commodity hardware. We have tested our solution on two Raspberry Pi models and over a variety of communication channels, where we were able to carry a real-time voice conversation.
Due to the significant advancements in image processing and machine learning algorithms, it is much easier to create, edit, and produce high quality images. However, attackers can maliciously use these tools to create legitimate looking but fake images to harm others, bypass image detection algorithms, or fool image recognition classifiers. In this work, we propose neural network based classifiers to detect fake human faces created by both 1) machines and 2) humans. We use ensemble methods to detect GANs-created fake images and employ pre-processing techniques to improve fake face image detection created by humans. Our approaches focus on image contents for classification and do not use meta-data of images. Our preliminary results show that we can effectively detect both GANs-created images, and human-created fake images with 94% and 74.9% AUROC score.
Are We There Yet?: Understanding the Challenges Faced in Complying with the General Data Protection Regulation (GDPR)
The EU General Data Protection Regulation (GDPR), enforced from 25\textsuperscriptth May 2018, aims to reform how organisations view and control the personal data of private EU citizens. The scope of GDPR is somewhat unprecedented: it regulates every aspect of personal data handling, includes hefty potential penalties for non-compliance, and can prosecute any company in the world that processes EU citizens’ data. In this paper, we look behind the scenes to investigate the real challenges faced by organisations in engaging with the GDPR. This considers issues in working with the regulation, the implementation process, and how compliance is verified. Our research approach relies on literature but, more importantly, draws on detailed interviews with several organisations. Key findings include the fact that large organisations generally found GDPR compliance to be reasonable and doable. The same was found for small-to-medium organisations (SMEs/SMBs) that were highly security-oriented. SMEs with less focus on data protection struggled to make what they felt was a satisfactory attempt at compliance. The main issues faced in their compliance attempts emerged from: the sheer breadth of the regulation; questions around how to enact the qualitative recommendations of the regulation; and the need to map out the entirety of their complex data networks.
The recent news of a large-scale online tracking campaign involving Facebook users, which gave way to systematic misuse of the collected user-related data, have left millions of people deeply concerned about the state of their online privacy as well as the state of the overall information security in the cyber world. While most to-date revelations pertaining to user tracking are related to websites and social media generally intended for adult online users, relatively little is known about the prevalence of online tracking in websites geared towards children and teens. In this paper, we first provide a brief overview of two laws that seek to protect the privacy of kids and teens online ? the US Children’s Online Privacy Act (COPPA) and the EU General Data Protection Regulation (GDPR). Subsequently, we present the results of our study which has looked for potential signs of user tracking in twenty select children-oriented websites in case of a user located in the USA (where COPPA is applicable) as well as a user located in the EU (where GDPR is applicable). The key findings of this study are alarming as they point to overwhelming evidence of widespread and highly covert user tracking in a range of different children-oriented websites. The majority of the discovered tracking is in direct conflict with both COPPA and GDPR, since it is performed without parental consent and by third-party advertising and tracking companies. The results also imply that, relative to their US counterparts, the children residing in the EU may be somewhat less subjected (but are still significantly exposed) to tracking by third-party companies.