This study investigates the trade-offs between fairness, privacy, and utility in image classification using machine learning (ML). Recent research suggests that generalization techniques can improve the balance between privacy and utility. One focus of this work is sharpness-aware training (SAT) and its integration with differential privacy (DP-SAT) to further improve this balance. Additionally, we examine fairness in both private and non-private learning models trained on datasets with synthetic and real-world biases. We also measure the privacy risks involved in these scenarios by performing membership inference attacks (MIAs) and explore the consequences of eliminating high-privacy risk samples, termed outliers. Moreover, we introduce a new metric, named harmonic score, which combines accuracy, privacy, and fairness into a single measure. Through empirical analysis using generalization techniques, we achieve an accuracy of 81.11% under (8, 10^-5)-DP on CIFAR-10, surpassing the 79.5% reported by De et al. (2022). Moreover, our experiments show that memorization of training samples can begin before the overfitting point, and generalization techniques do not guarantee the prevention of this memorization. Our analysis of synthetic biases shows that generalization techniques can amplify model bias in both private and non-private models. Additionally, our results indicate that increased bias in training data leads to reduced accuracy, greater vulnerability to privacy attacks, and higher model bias. We validate these findings with the CelebA dataset, demonstrating that similar trends persist with real-world attribute imbalances. Finally, our experiments show that removing outlier data decreases accuracy and further amplifies model bias.

This paper explores the application of large language models (LLMs), like ChatGPT, for biometric tasks. We specifically examine the capabilities of ChatGPT in performing biometric-related tasks, with an emphasis on face recognition, gender detection, and age estimation. Since biometrics are considered as sensitive information, ChatGPT avoids answering direct prompts, and thus we crafted a prompting strategy to bypass its safeguard and evaluate the capabilities for biometrics tasks. Our study reveals that ChatGPT recognizes facial identities and differentiates between two facial images with considerable accuracy. Additionally, experimental results demonstrate remarkable performance in gender detection and reasonable accuracy for the age estimation tasks. Our findings shed light on the promising potentials in the application of LLMs and foundation models for biometrics.

Face inpainting, the technique of restoring missing or damaged regions in facial images, is pivotal for applications like face recognition in occluded scenarios and image analysis with poor-quality captures. This process not only needs to produce realistic visuals but also preserve individual identity characteristics. The aim of this paper is to inpaint a face given periocular region (eyes-to-face) through a proposed new Generative Adversarial Network (GAN)-based model called Eyes-to-Face Network (E2F-Net). The proposed approach extracts identity and non-identity features from the periocular region using two dedicated encoders have been used. The extracted features are then mapped to the latent space of a pre-trained StyleGAN generator to benefit from its state-of-the-art performance and its rich, diverse and expressive latent space without any additional training. We further improve the StyleGAN's output to find the optimal code in the latent space using a new optimization for GAN inversion technique. Our E2F-Net requires a minimum training process reducing the computational complexity as a secondary benefit. Through extensive experiments, we show that our method successfully reconstructs the whole face with high quality, surpassing current techniques, despite significantly less training and supervision efforts. We have generated seven eyes-to-face datasets based on well-known public face datasets for training and verifying our proposed methods. The code and datasets are publicly available1.

Online Social Networks are responsible for disclosing a large amount of sensitive information. Often, users unknowingly disclose vast amounts of sensitive and potentially (un)related data, oblivious to the associated privacy risks. Our research provides a comprehensive evaluation of the linkability between user profiles and shared content across various OSNs, a factor that has considerable implications for privacy leakage. We introduce a novel method for quantifying the linkability between profiles across multiple networks, based on key features and metrics that capture profile similarities. We applied this methodology to a dataset of user profiles across three online social networks named Flickr, Facebook, and Twitter. Our approach includes examining both structured and unstructured data related to user profiles, enabling us to offer a valuable understanding of linkability trends and identify potential privacy risks. Through our findings, we aim to inform the development of privacy-enhancing technologies and contribute to improving the current privacy landscape within OSNs. Our research underscores the critical need for robust privacy measures in the face of the growing interconnectedness of user data across different social networks.

This study aims to investigate the relationship between stress levels among hospital staff and their risky cybersecurity practices. A web-based survey was conducted with a sample of 353 hospital staff from Ghana, Norway, and Indonesia. The results indicate a statistically significant positive correlation between the stress levels of hospital staff and their engagement in unsafe cybersecurity practices (r = 0.201, p < 0.01). Specifically, the study finds that staff members’ inclination to click on links from unknown sources is the cybersecurity practice most strongly influenced by stress levels. The study did not observe any significant differences in cybersecurity practices based on gender, age, job, position level, or work experience. However, it does highlight notable differences in cybersecurity practices across countries, with Norwegian hospital staff exhibiting better cybersecurity practices than their counterparts from Ghana and Indonesia.

This study introduces a new technique for generating synthetic faces using eyes-to-face inpainting methods. The proposed method can synthesize a face image using a combination of the eyes of two different individuals and use it as an input for inpainting, demonstrating its vast potential for various applications in biometrics. Despite minor biases in age and gender, our method proved effective in training reliable age- and gender-detection models using the generated datasets. We also addressed the challenge of training face recognition models using synthetic datasets, and the results demonstrated satisfactory accuracy across four benchmark face recognition datasets. This method could be particularly beneficial for underrepresented groups, for whom there is a scarcity of face samples in biometric datasets.

This paper presents the summary of the Efficient Face Recognition Competition (EFaR) held at the 2023 International Joint Conference on Biometrics (IJCB 2023). The competition received 17 submissions from 6 different teams. To drive further development of efficient face recognition models, the submitted solutions are ranked based on a weighted score of the achieved verification accuracies on a diverse set of benchmarks, as well as the deployability given by the number of floating-point operations and model size. The evaluation of submissions is extended to bias, cross-quality, and large-scale recognition benchmarks. Overall, the paper gives an overview of the achieved performance values of the submitted solutions as well as a diverse set of baselines. The submitted solutions use small, efficient network architectures to reduce the computational cost, some solutions apply model quantization. An outlook on possible techniques that are underrepresented in current solutions is given as well.

Online Social Networks are responsible for disclosing a large amount of sensitive information. Users unintentionally reveal their sensitive information and are unaware of the privacy risks involved. But the users should be well informed about their privacy quotient and should know where they stand on the privacy measuring scale. In this paper, we proposed an adaptive privacy measuring framework called PriMe that can measure the privacy leakage score for each action of a user in an OSN and subsequently adjust the privacy settings based on the preferred privacy scopes and boundaries. Various types of data, actions, and personal characteristics of each user have been considered to ensure the calculated privacy leakage score is accurate.

Since healthcare information systems have many important data that can attract many adversaries, it is important to take the right steps to prevent data breaches. Recent studies suggested that 85% of breaches involved a human element and the frequent patterns used are social engineerings. Therefore, many studies focus on making a better understanding of human behavior in cybersecurity and the factors that affect cybersecurity practices. However, there are only a few peer-reviewed studies that focus on the link between stress level and cybersecurity practices. In this study, we examined the link between stress level and cybersecurity practices among hospital employees in Indonesia by surveying 99 hospital workers. Perceived Stress Scale (PSS) was used to measure the employees’ stress level and a new scale to measure hospital staff’s risky cybersecurity practices was proposed. This study showed that both PSS and proposed cybersecurity practices scales are reliable with Cronbach’s α value of more than 0.7. The survey results also revealed that hospital worker’s higher stress levels correlate significantly with riskier cybersecurity practices (rs = 0.305, p < 0.01). Besides, a higher stress level is also significantly linked to certain cybersecurity practices, such as clicking on a link in an email from an unknown sender, not preventing colleagues from viewing patients’ information for a non-therapeutic purpose, posting patient information on social media, ignoring colleagues who engage in negative information security practices, and failing to create strong passwords.

Because of the population aging, homecare monitoring systems and assisted living technologies have been promoted by researchers and industries to help patients and the elderly at home to get medical help in time. Nevertheless, these systems and technologies usually come up with ethics problems and involve patient's privacy concerns. Even if privacy-enhancing technologies have been introduced to the monitoring systems to help protect patient's privacy, it remains a problem for researchers to figure out individual's privacy attitudes and behaviors when being monitored. Since patients have different privacy attitudes and preferences, not only technical staff but also health care providers should take patient's privacy concerns into account when making medical decisions and provide clinical help. In this paper, we are going to discuss a preliminary study about patient privacy decision-making we carried out recently. After analyzing the results, we conducted a follow-up study to figure out the reasons for the results.

Electronic health records (EHR) consists of broad, numerous and erratic accesses through self-authorizations and "brake the glass" scenarios. This is to fulfil the availability aspect of the the CIA (confidentiality, integrity) due to the time sensitive nature in healthcare especially during health emergency situations. Adversaries can use this as opportunity to illegitimately access patients records, thereby, compromising the entire EHR system.To avert this, a comparative analysis of machine learning classification methods was conducted with simulated EHR logs. The methods which were compared are Multinomial Naive Bayes(multnb), Bernoulli Naive Bayes (bernnb), Support Vector Machine (svm), Neural Network (nn), K-Nearest Neighbours(knn), Logistic Regression (lr), Random Forest (rf), and Decision Tree (dt).The experiment results show that all of the machine learning models used in this work performed very well for the role classification task but, Decision Tree (dt) and Random Forrest (rf) obtained the best result among all of the methods with the same accuracy value of 0.889 on all three datasets. For the anomaly detection task, generally, our proposed approach obtained a high recall and accuracy but low precision and F1-score. Soft Classification approach performed better than the Hard Classification approach. The best performance was achieved with Bernoulli Naive Bayes with none normalised data, with an F1-score of 0.893.

The legacy-UI honeywords generation approach is more favored due to its high usability compared to the modified-UI approach that sometimes becomes unusable in practice. However, several prior arts on legacy-UI based honeywords generation methods often fail to obtain the security standard, especially the flatness criterion. In this work, we propose two legacy-UI honeywords generation strategies based on two password guessing methods: PassGAN and Probabilistic Context-Free Grammar (PCFG). Besides, we also introduce two hybrid strategies by combining PassGAN, PCFG, and random-based methods. We empirically examine the flatness of the proposed honeywords generation strategy against Top Password (Top-PW) attack using real-world datasets, instead of only providing heuristic security arguments. The experiment results show that three of the proposed methods (the PassGAN-based and the two hybrid methods) have lower flatness value than all previous legacy-UI methods and able to meet the "perfectly flat" criterion.

Healthcare professionals are often the weakest link in the security chain, which is contributing to data breaches in the healthcare sector. A number of reasons account for this. Technological countermeasures for cyber defenses have been heightened and the adversaries tend to exploit easy entry points. Besides, healthcare staffs are usually occupied by their core duty of healthcare provison with little experience in information security.With a Design Science Approach (DSA), observational measures for effective profiling of healthcare staffs were developed. Regulations and security standards such as the Code of Conduct, General Data Protection Regulation (GDPR) of European Union (EU), ISO 7799, and other Norwegian Acts and regulations for personal data protection, were reviewed for the observational measures. A comprehensive Healthcare Security Practice Analysis, Modeling and Incentivization (HSPAMI) framework was proposed for analyzing healthcare staffs' security practices in a comprehensive way.

The healthcare sector is characterized with variant situations and services such as emergency services, collaborations in patient care and patient referrals. These activities require erratic accesses and electronic exchange of personal health information (PHI) between health professionals and healthcare organizations. Also, healthcare information is deemed to be among the most confidential of all types of personal data. Analyzing and modeling the security threats emanating from healthcare staffs' security practices therefore need an efficient approach. There is a need for tailored measures to be adopted in assessing healthcare personnel security practices in relation to Confidentiality, Integrity and Availability (CIA) threats. Standards and technical security implementations, required by regulatory bodies, have resulted in tracking healthcare staffs' security practices in various data sources which can be explored for security countermeasures. A literature survey was adopted to obtain the most appropriate observational measures that can be used to empirically study healthcare staffs' security practice analysis, modeling and incentivization (HSPAMI). The survey was conducted in journal and conference articles, healthcare security breaches reports and AI tools for detecting anomalous healthcare staff security practices. The survey results did not find a comprehensive and tailed observational measures suitable for the HSPAMI project. A comprehensive and tailored observational measures were therefore developed from healthcare standards, legal, regulatory aspects, and the code of conduct. Observational measures relating to healthcare security practices such as self-authorization, inter-organizational accesses to PHI and ICT readiness were found to be unique and have not been factored in existing observational measures for efficient profiling of healthcare staffs.

Identity fraud is a serious problem which can be used to conduct crimes such as economic fraud, human trafficking and terrorism. Many organizations have pointed out challenges in performing identity control, and a more operable framework for identity proofing and verification is desired in many practical applications. Although there are several guidelines and international standardization activities available on identity proofing and verification routines, complexity and variation among these frameworks can make them complicated to interpret and understand and thus little operable, particularly for smaller organizations performing identity control. Against the increasing trend of identity fraud worldwide, this paper proposes an Evidence of Identity (EoI) evaluation system design aiming at operationalizing requirements to evidence of identities in ID proofing and verification processes. The suggested system is designed to be included in a computer application, allowing easy use by front-desk officers.

Presents the introductory welcome message from the conference proceedings. May include the conference officers' congratulations to all involved with the conference event and publication of the proceedings record.

Biometric binarization is the feature-type transformation that converts a specific feature representation into a binary representation. It is a fundamental issue to transform the real-valued feature vectors to the binary vectors in biometric template protection schemes. The transformed binary vectors should be high for both discriminability and privacy protection when they are employed as the input data for biometric cryptosystems. In this paper, we propose a novel binarization scheme based on random projection and random Support Vector Machine (SVM) to further enhance the security and privacy of biometric binary vectors. The proposed scheme can generate a binary vector of any given length as an ideal input for biometric cryptosystems. In addition, the proposed scheme is independent of the biometric feature data distribution. Several comparative experiments are conducted on multiple biometric databases to show the feasibility and efficiency of the proposed scheme.

Using one password for all web services is not secure because the leakage of the password compromises all the web services accounts, while using independent passwords for different web services is inconvenient for the identity claimant to memorize. A password manager is used to address this security-convenience dilemma by storing and retrieving multiple existing passwords using one master password. On the other hand, a password manager liberates human brain by enabling people to generate strong passwords without worry about memorizing them. While a password manager provides a convenient and secure way to managing multiple passwords, it centralizes the passwords storage and shifts the risk of passwords leakage from distributed service providers to a software or token authenticated by a single master password. Concerned about this one master password based security, biometrics could be used as a second factor for authentication by verifying the ownership of the master password. However, biometrics based authentication is more privacy concerned than a non-biometric password manager. In this paper we propose a cloud password manager scheme exploiting privacy enhanced biometrics, which achieves both security and convenience in a privacy-enhanced way. The proposed password manager scheme relies on a cloud service to synchronize all local password manager clients in an encrypted form, which is efficient to deploy the updates and secure against untrusted cloud service providers.

The face quality assessment from video must quantitatively measure the applicability of the face images that are typically captured over multiple frames with various degradations. In this work, we address the face quality assessment from the video captured using Automatic Border Control (ABC) system. To this extent, we employed MorphoWayTM ABC system as a data capture device to construct a new database by simulating real-life scenario. We then propose a new scheme for face quality estimation that can be viewed in three steps: (1) Pose estimation by detecting face parts (eyes and nose) to separate frontal from non-frontal faces. (2) We then consider the frontal face and evaluate its corresponding image quality by analyzing its texture components using Grey Level Co-occurrence Matrix (GLCM). (3) Finally, we quantify the quality of the given face image using likelihood values obtained using Gaussian Mixture Model (GMM). Extensive experiments are carried out on our new database that exhibits various quality degradations due to head pose variations, change in illumination, expression, motion blur, etc. The experimental results have indicated that the proposed face quality assessment algorithm can effectively classify the input image into relevant quality bins that in turn can be employed for the improved face verification.

This paper proposes an approach to qualifying fingerprint samples captured by smartphone cameras under real-life scenarios, foreseeing the future application using such general purposed cameras as fingerprint sensors. In this approach, a sample image is first divided into non-overlapping blocks. Then a 7-dimensional feature vector will be formed from the proposed 7 quality features. We use a support vector machine to produce a binary indication for each image block on its quality. Finally a quality score is generated to indicate the whole fingerprint sample's quality by counting the number of qualified blocks in a sample. Experiments demonstrate the proposed approach's capability of qualifying such quality-challenging fingerprint samples - the Spearman's rank correlation coefficient ρ between the proposed quality metric and samples' normalized comparison scores reaches as high as 0.53 in our experiment.

In this paper, we present an empirical study on exploring the Light Field Camera (LFC) for identifying multiple faces present at different distance. Since LFC can render multiple focus images in single exposure, one can combine these multiple images to obtain single all-in-focus image. Thus the constructed all-in-focus image will have all regions in focus and hence allows one to capture more information about the subject present even at a far distance. At the same time one can also construct the super resolution image to further improve the face recognition at a distance. Thus, in this work, we explore both all-in-focus and super resolution schemes to evaluate the multiple face recognition at a distance using LFC. We carry out extensive experiments on light field face dataset and present both qualitative and quantitative results.

A local-area based fingerprint minutiae vicinity can be represented in a self-aligned way and achieves better robustness for recognition than a global minutiae template whose global geometric references (e.g., core or delta) are usually unstable to locate. However, local comparison based on vicinities ignores the global topology and thus still has potential to improve in performance if some stable global information can be fused in vicinities. We construct in this paper an augmented minutia vicinity, which incorporates more contextual minutiae information. Application of the proposed augmented vicinity to minutiae template protection is tested and demonstrates desirable biometric performance (e.g., FRR = 0.04-0.06 and FAR = 0.001 on the FVC2002DB2_A database) with roughly 70-bit complexity against reversing a binary protected augmented vicinity.

A renewable fingerprint minutiae template generation scheme is proposed to utilize random projection for template diversification in a security enhanced way. The scheme first achieves absolute pre-alignment over local minutiae quadruplets in the original template and results in a fix-length feature vector; and then encrypts the feature vector by projecting it to multiple random matrices and quantizing the projected result; and finally post-process the resultant binary vector in a size and security tunable way to obtain the final protected minutia vicinity. Experiments on the fingerprint database FVC2002DB2_A demonstrate the desirable biometric performance of the proposed scheme.

Often in the development of a biometric product an evaluator of the system is the same entity who developed the algorithm. Moreover, usually the test data employed in such evaluation is also collected by the same developer/evaluator. In most cases such database will not be made public and consequently test results cannot be verified by independent institutions. This paper presents an independent report on fingerprint performance evaluation that has been conducted in the context of the TURBINE project. In this study the algorithm developer and system evaluator are represented by separate and independent entities. In addition, the algorithm developer does not have access to the primary test database. All these provide pre-conditions to unbiased and trustworthy performance reports. Furthermore, this paper introduces biometric performance testing on a level of biometric references, which is complementary to image- or minutiae-based references. Biometric references in the TURBINE project are pseudonymous identifiers that have been generated by the template protection algorithms. The results of the performance evaluation in this paper are generated by applying the algorithm developers (binary) algorithms at the minutiae (traditional) and pseudonymous identifier levels. The test data set consists of almost 72000 fingerprint images from 100 subjects acquired by several fingerprint scanners to which the algorithm developers did not have access.

A robust audio hashing algorithm for AAC audio is proposed in this paper. The robust hash is calculated from MDCT-coefficients which are derived by partial decoding of AAC bitstream. There is no complicated transformation in the proposed algorithm, therefore, it is of low computational complexity. The proposed method is highly robust to MDCT-based audio compression. Experimental results also show its high discrimination power and the robustness against content preserved operations.

In this paper, a perceptual audio hashing method in compressed domain is proposed for content identification, in which MDCT coefficients as the intermediate decoding result are selected for perceptual feature extraction and hash generation. The perceptual feature extraction is based on psychoacoustic model and exhibits good discrimination ability for different audio contents but robustness against common audio signal processing operations. Via feature extraction in the compressed domain, the MDCT-based compressed audios, such as MP3, AAC, etc., could be efficiently identified without complete decoding which facilitates those practical applications with strict requirements of memory and computational complexity, such as online audio retrieval, indexing of massive compressed audio data, audio identification by mobile phone, etc. The algorithm is highly robust against MDCT compression which is widely used in audio coding. Experiments demonstrate the effectiveness of the proposed scheme.

We present a reversible watermarking scheme which achieves perfect restoration of both the embedded watermark and the original image during extraction. The proposed scheme embeds data by modifying those integer DCT coefficients with peak amplitudes in each coefficient histogram. The integer DCT performed over the original image is a lossless 8/spl times/8 block transform with high energy concentrating ability, which guarantees reversibility and high capacity/distortion ratio for the proposed watermarking scheme. In addition, this scheme provides a wide quality (PSNR) range from around 40 dB to 60 dB for the watermarked image, and an inherent fine adjustment capability for the quality (PSNR). Some experimental results are presented to demonstrate the high performance of our scheme in terms of capacity and the quality of the watermarked image.