Aug 2026· Journal of Intelligent Decision Making and Information Science· Vol 3, pp. 63-80· 0 citations· 21 references
TL;DR
EquiAI is proposed, a robust fairness-aware remote identity verification framework that integrates a pretrained Vision Foundation Model (DINOv2), fairness-aware representation learning, adaptive feature alignment, presentation attack detection, and explainable artificial intelligence into a unified architecture.
Abstract
Remote identity verification has become a critical component of autonomous agents deployed in smart healthcare, intelligent transportation, digital finance, and secure access control systems. Recent advances in Vision Foundation Models (VFMs) have significantly improved facial representation learning; however, their deployment remains challenged by demographic bias, which leads to inconsistent verification performance across different age groups, genders, ethnicities, and skin tones. Such disparities reduce the reliability, fairness, and trustworthiness of AI-enabled biometric systems, particularly in safety-critical applications where accurate identity verification is essential. This study proposes EquiAI, a robust fairness-aware remote identity verification framework that integrates a pretrained Vision Foundation Model (DINOv2), fairness-aware representation learning, adaptive feature alignment, presentation attack detection, and explainable artificial intelligence into a unified architecture. The proposed framework employs transformer-based feature extraction to learn generalized facial embeddings while minimizing demographic disparities and enhancing resistance against spoofing attacks. Experimental evaluation demonstrates that EquiAI achieves a verification accuracy of 98.21%, an F1-score of 97.94%, an ROC-AUC of 0.991, and an Equal Error Rate of 1.82%, while maintaining balanced performance across diverse demographic groups and supporting real-time inference. These findings demonstrate that integrating Vision Foundation Models with fairness-aware optimization and explainable AI substantially improves the robustness, equity, and practical deployment of remote identity verification systems, providing a scalable foundation for trustworthy autonomous agent authentication in real-world environments.
Generative diffusion models have revolutionized facial image synthesis, yet robust identity preservation in high resolution outputs remains a critical challenge. This issue is especially vital for security systems, biometric authentication, and privacy sensitive applications, where any drift in identity integrity can undermine trust and functionality. We introduce Diff-ID, a diffusion based framework that enforces identity consistency while delivering photorealistic quality. Central to our approach is a custom 210K image dataset synthesized from CelebA-HQ, FFHQ, and LAION-Face and captioned via a fine tuned BLIP model to bolster identity awareness during training. Diff-ID integrates ArcFace and CLIP embeddings through a dual cross attention adapter within a fine tuned Stable Diffusion UNet. To further reinforce identity fidelity, we propose a pseudo discriminator loss based on ArcFace cosine similarity with exponential timestep weighting. Experiments on held out and unseen faces show that Diff-ID does not exceed InstantID in raw ArcFace Face Similarity, but achieves substantially lower FID and the strongest FIQ based identity--realism trade off among the evaluated methods. We also present a unified DDIM based morphing pipeline that enables qualitative facial interpolation without per identity fine tuning. We further argue that identity preservation and photorealism should be evaluated jointly rather than in isolation, as high identity similarity alone does not guarantee realistic outputs. To make this trade off explicit, we report Face Image Quality (FIQ) as a complementary ratio based score that combines identity similarity and perceptual realism while keeping FS and FID as the primary metrics.
T. Rizwan, Sara Atito, Muhammad Awais et al.· 0 citations
Biometric authentication systems employed in Nigeria’s electoral process continue to encounter significant challenges, particularly in addressing impersonation, facial occlusion, and presentation attacks. Although technologies such as the Bimodal Voter Accreditation System (BVAS) have enhanced electoral transparency and credibility, concerns regarding system reliability and susceptibility to spoofing attacks persist. To mitigate these limitations, this study proposes a hybrid framework that integrates mask-resilient face recognition with anti-spoofing mechanisms for secure voter authentication. The framework is based on a deep convolutional neural network (CNN) trained on a dataset comprising 1,478 facial images categorized into real, masked, and spoof classes. Standard preprocessing procedures, including face detection, image resizing to 128 × 128 pixels, and normalization, were applied to ensure input consistency and optimize model performance. Experimental results indicate that the proposed model achieves 97.3% classification accuracy, demonstrating its ability to effectively distinguish among real, masked, and spoof facial inputs. Furthermore, biometric evaluation metrics, namely the Attack Presentation Classification Error Rate (APCER), Bona fide Presentation Classification Error Rate (BPCER), and Average Classification Error Rate (ACER), confirm the model’s effectiveness in detecting spoofing attempts while maintaining acceptable performance for legitimate users. In conclusion, the findings suggest that the proposed framework offers strong potential for real-world biometric authentication applications. It provides an efficient, integrated solution for enhancing security in electoral systems and can be extended to other security-critical domains.
Opeyemi Lateef Usman, Khadijah Opeyemi Owodunni· The Scientific World Journal· 0 citations
Face analysis systems are widely used in security, authentication, and public-sector applications; however, demographic bias and the statistical reliability of reported performance remain key concerns. Many studies rely on aggregate accuracy without quantifying subgroup disparities or uncertainty, potentially overstating model fairness. This study presents a statistically grounded evaluation of demographic bias in face attribute classification across three representative architectures, ResNet50, MobileNetV3, and a vision transformer (DeiT), using the FairFace and UTKFace datasets. Subgroup analysis is conducted across race and gender, incorporating disparity indices, bootstrap confidence intervals, and inferential statistical testing with effect size analysis. The evaluation uses an embedding-based nearest-neighbor approach to examine representation-level behavior consistently across models. Results show that race-based disparities are substantially larger than gender-based disparities across both datasets. On FairFace, race disparity gaps range from 0.1124 to 0.1266, while on UTKFace they increase significantly to 0.4726–0.4944, with large effect sizes (Cohen's d>1). In contrast, gender disparities remain smaller, with gaps between 0.0280 and 0.0582 on FairFace and 0.0194–0.0326 on UTKFace, and correspondingly small effect sizes (d < 0.13). Despite modest differences in overall accuracy across models, subgroup disparities remain statistically significant across all architectures. These findings emphasize the importance of subgroup-level evaluation, uncertainty quantification, and statistical validation for reliable fairness assessment in face analysis systems.
Andisani Nemavhola, Serestina Viriri, C. Chibaya· Frontiers in Artificial Inte...· 0 citations
Facial recognition technology is an artificial intelligence-based biometric technology that stands at the forefront of global forensic investigations as a leading and widely adopted biometric modality. In comparison to alternative parameters such as voice, fingerprint, iris, retina, eye scan, gait, ear, and hand geometry; facial recognition emerges as the most popular and effective tool for personal identification and verification. Beyond its role in research, access control, user authentication, and border security, the technology plays a dynamic role in law enforcement and surveillance. Despite its extensive applications, the technology sparks privacy and ethical debates. Contemporary concerns revolve around its potential to implicate innocent individuals, raising issues of civil liberties, human rights, and privacy infringement. In light of these considerations, this article critically examines the reliability and consistency of facial recognition technology and incorporates case studies highlighting instances of wrongful detentions. It also explores the principles and functioning of facial recognition technology. The face recognition outcomes rely heavily on features that are extracted to reflect the face pattern and classification techniques used to distinguish between faces. However, there are a few limitations associated with the facial recognition technology. This is prone to errors in detecting some facial features and skin tones, which raises potential risks and consequences for innocent persons and communities. Therefore, these mistaken identity cases emphasize the urgent need for robust oversight, transparency and safeguards to ensure fairness, preventing misuse and upholding the dignity and privacy of individuals in the growing era of artificial intelligence. The present communication concludes with actionable recommendations to address the ethical and privacy challenges associated with facial recognition in the forensic context.
Ankita Guleria, Nandini Chitara, Damini Siwan et al.· Medicine, Science and the La...· 0 citations
Digital identity verification has become essential across online banking, e-governance, healthcare, education, and e-commerce. While biometric systems such as facial recognition, fingerprint scanning, and iris detection enhance convenience and scalability, they remain vulnerable to spoofing attacks including photo/video replays, deepfakes, silicone masks, and display-based presentation attacks. Liveness detection has therefore emerged as a critical security layer to distinguish genuine biometric traits from fraudulent representations. This paper presents a comprehensive study of liveness detection models, covering architecture, algorithms, and performance evaluation. It examines active and passive techniques, including texture analysis, motion cues, physiological signal extraction, and deep learning approaches. Traditional handcrafted features are compared with CNNs, RNNs, and transformer-based models. Emphasis is placed on multimodal biometrics and challenge-response mechanisms to counter advanced GAN-based deepfake attacks. A systematic framework is proposed, encompassing data acquisition, preprocessing, feature extraction, model training, and decision fusion, supported by mathematical formulations for classification, loss optimization, and evaluation metrics. Experimental results on benchmark datasets demonstrate improvements in accuracy, FAR, and APCER, with hybrid deep learning models integrating temporal and physiological cues outperforming single-modality methods. The paper concludes by addressing deployment challenges, ethical considerations, and future directions such as privacy-preserving learning, federated identity systems, and explainable AI security, offering a scalable approach for secure digital identity verification.
Salma El-Sayed· International Journal of App...· 0 citations
With the continuous development of intelligent vision technology, facial image anonymization plays an increasingly important role in public privacy protection and data compliance sharing. However, traditional methods have a contradiction between identity feature removal and image naturalness preservation, and are prone to distortion when dealing with semantic consistency issues such as posture. To this end, a structure identity separation guided facial anonymous generation method is constructed on the framework of generative adversarial networks. A structural perturbation module guided by differential privacy mechanisms is designed, and an attention guided feature weakening mechanism is introduced. Additionally, a style guided and structure preserving conditional generation network is integrated. In performance testing, the proposed method achieves identity consistency scores of 0.814 and 0.759 for frontal and lateral postures, respectively, with corresponding natural scores of 4.08 and 3.95. The inference delay, number of generated graphs per unit energy consumption, and peak memory of this method are 114.8 ms, 19.2 IpJ, and 1286 MB, respectively. The experimental results show that this method balances visual consistency and structural rationality while enhancing identity concealment, providing a feasible path and technical support for the secure generation and compliant application of private images.
Zhiyong Sun, Li-Hsuan Li· Information Technology and C...· 0 citations