IEEE transactions on biometrics, behavior, and identity science最新文献

{"title":"Autoencoders for Open-Set Presentation Attack Detection","authors":"Igor Bastos;Anjith George;Sébastien Marcel;Anderson Rocha","doi":"10.1109/TBIOM.2026.3651671","DOIUrl":"https://doi.org/10.1109/TBIOM.2026.3651671","url":null,"abstract":"Face anti-spoofing (FAS) has gained significant attention due to its crucial role in ensuring the security of facial recognition systems. Despite numerous proposed methods for addressing various types of presentation attacks, there remains a gap in the existing literature, as most methods only tackle seen attack types and lack generalizability in real-world scenarios. This paper presents a novel method for face antispoofing, specifically focused on open-set scenarios. The proposed approach consists of an autoencoder exclusively trained to reconstruct bona fide data. The key idea is that features extracted using this model would exhibit poor quality for attack data, resulting in significantly higher reconstruction errors. Consequently, these low-quality features would differ from the well-reconstructed features of bona fide samples. Our method’s key novelty lies in the rich feature learning procedure that focuses on the properties of bona fide samples through an autoencoder formulation. The extracted features are used to train one-class classifiers using only bona fide samples. Experimental results on standard benchmarks such as RECOD-MPAD, WMCA, HQ-WMCA, Replay-Attack, OULU-NPU and MSU-MFSD datasets demonstrate the excellent performance of our proposed approach, placing it among state-of-the-art methods for metrics like Bona Fide Presentation Classification Error Rate (BPCER), Attack Presentation Classification Error Rate (APCER) and Average Classification Error Rate (ACER).","PeriodicalId":73307,"journal":{"name":"IEEE transactions on biometrics, behavior, and identity science","volume":"8 2","pages":"192-206"},"PeriodicalIF":5.0,"publicationDate":"2026-01-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146216654","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Dual Frequency-Guided Spatiotemporal Feature Learning for Face Forgery Detection","authors":"Junxian Duan;Siyu Liu;Yiming Hao;Huaibo Huang;Ran He","doi":"10.1109/TBIOM.2025.3646181","DOIUrl":"https://doi.org/10.1109/TBIOM.2025.3646181","url":null,"abstract":"Video forgery detection is challenging due to high computational demands and robustness issues in the era of generative AI. Existing methods struggle to capture temporal dynamics and subtle details, and may miss high-frequency features (e.g., textures, edges) critical for detecting forgery clues. To address this challenge, this paper proposes an innovative Dual Frequency-guided spatiotemporal feature learning (DFSL) network, which enhances forgery detection accuracy and robustness by incorporating frequency feature extraction alongside spatiotemporal feature capture. First, we introduce the Multi-Selective State-Space Module and Sequential Tri-frame Local Module to capture spatiotemporal information and temporal dynamics in videos more effectively. Then, we leverage the Dual-frequency Prompt Learning, which combines the Fourier Transform and the Wavelet Transform to extract global and local frequency information from the video. Finally, we explore a multimodal feature fusion scheme to enhance the model’s ability to capture both global dynamics and subtle traces. Experimental results demonstrate that the proposed method significantly improves forgery detection accuracy while maintaining low computational complexity and strong robustness.","PeriodicalId":73307,"journal":{"name":"IEEE transactions on biometrics, behavior, and identity science","volume":"8 2","pages":"179-191"},"PeriodicalIF":5.0,"publicationDate":"2025-12-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146223612","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Toward Certifiably Robust Face Recognition: Analyses and Improvements","authors":"Seunghun Paik;Dongsoo Kim;Chanwoo Hwang;Sunpill Kim;Jae Hong Seo","doi":"10.1109/TBIOM.2025.3644396","DOIUrl":"https://doi.org/10.1109/TBIOM.2025.3644396","url":null,"abstract":"Adversarial perturbations have been one of the most notable threats against the safe and trustworthy applications of deep learning. For security-critical applications, e.g., face recognition (FR), the importance of a theoretically robust defense against adversarial perturbations has been spotlighted. Certifiable robustness aims to defend against adversarial perturbations in a provable manner, and several studies have been conducted to achieve certifiable robustness in various domains. However, most existing studies for certifiable robustness are about classifiers, and adapting their techniques for FR is a non-trivial problem. In this study, we show that, similar to the image classifications, the 1-Lipschitz condition is sufficient for certifiable robustness of the face recognition system against any <inline-formula> <tex-math>$ell _{p}$ </tex-math></inline-formula> norm adversaries for <inline-formula> <tex-math>$p in mathbb {N} cup {infty }$ </tex-math></inline-formula>. In addition, we investigate the trade-off between accuracy drop and certifiable robustness in 1-Lipschitz FR models, and propose several techniques to reconcile such a trade-off. We conduct extensive theoretical and experimental analyses on our findings. Notably, our techniques improve the standard (certifiably robust, resp.) accuracy by 6.98% (at most 13.35%, resp.) in the LFW benchmark against <inline-formula> <tex-math>$ell _{2}$ </tex-math></inline-formula> norm adversaries compared to accuracies without them. To facilitate further study, we publicly release our source code on <uri>https://github.com/Cryptology-Algorithm-Lab/CertRobFR</uri>","PeriodicalId":73307,"journal":{"name":"IEEE transactions on biometrics, behavior, and identity science","volume":"8 2","pages":"255-269"},"PeriodicalIF":5.0,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146216655","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Transformer Network-Based Gait Identification Using WiFi","authors":"Oliver Custance;Saad Khan;Simon Parkinson","doi":"10.1109/TBIOM.2025.3643169","DOIUrl":"https://doi.org/10.1109/TBIOM.2025.3643169","url":null,"abstract":"Wireless sensing has become a prominent choice for human activity recognition, valued for its non-intrusive operation and privacy-conscious design. However, it grapples with environmental challenges, especially from static (e.g., furniture) and dynamic objects (e.g., people walking), as well as how demographic factors such as BMI and sensor quantity affect accuracy. This paper addresses these gaps through a focused experiment on gait-based participant identification. Our methods and techniques encompass variance exploration for movement detection, Channel Power Distribution (CPD) analysis, and polynomial fitting to determine walking direction. We present a novel model for gait recognition, using a hybrid architecture that combines convolutional layers, LSTM blocks with residual connections, and multi-head self-attention mechanisms. We conducted a comprehensive evaluation of our gait identification system using two datasets: MultiEnvironment and HWDD. Both consist of data from 10 volunteers. We achieved 96.1% and 96.66% accuracy for MultiEnvironment and HWDD, respectively. To benchmark our system, we selected four state-of-the-art pre-trained models: Transformer, LSTM, CNN and SVM. Finally, we benchmark our technique against a dataset that we collected for 25 individuals, demonstrating an accuracy better than other state-of-the-art techniques of 97.9%.","PeriodicalId":73307,"journal":{"name":"IEEE transactions on biometrics, behavior, and identity science","volume":"8 2","pages":"311-326"},"PeriodicalIF":5.0,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11298190","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146216638","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SpectraIrisPAD: Leveraging Vision Foundation Models for Spectrally Conditioned Multispectral Iris Presentation Attack Detection","authors":"Raghavendra Ramachandra;Sushma Venkatesh","doi":"10.1109/TBIOM.2025.3641831","DOIUrl":"https://doi.org/10.1109/TBIOM.2025.3641831","url":null,"abstract":"Iris recognition is widely recognized as one of the most accurate biometric modalities. However, its growing deployment in real-world applications raises significant concerns regarding its vulnerability to Presentation Attacks (PAs). Effective Presentation Attack Detection (PAD) is therefore critical to ensure the integrity and security of iris-based biometric systems. While conventional iris recognition systems predominantly operate in the near-infrared (NIR) spectrum, multispectral imaging across multiple NIR bands provides complementary reflectance information that can enhance the generalizability of PAD methods. In this work, we propose SpectraIrisPAD, a novel deep learning-based framework for robust multispectral iris PAD. The SpectraIrisPAD leverages a DINOv2 Vision Transformer (ViT) backbone equipped with learnable spectral positional encoding, token fusion, and contrastive learning to extract discriminative, band-specific features that effectively distinguish bona fide samples from various spoofing artifacts. Furthermore, we introduce a new comprehensive dataset Multispectral Iris PAD (MSIrPAD) with diverse PAIs, captured using a custom-designed multispectral iris sensor operating at five distinct NIR wavelengths (800 nm, 830 nm, 850 nm, 870 nm, and 980 nm). The dataset includes 18,848 iris images encompassing eight diverse PAI categories, including five textured contact lenses, print attacks, and display-based attacks. We conduct comprehensive experiments under unseen attack evaluation protocols to assess the generalization capability of the proposed method. SpectraIrisPAD consistently outperforms several state-of-the-art baselines across all performance metrics, demonstrating superior robustness and generalizability in detecting a wide range of presentation attacks.","PeriodicalId":73307,"journal":{"name":"IEEE transactions on biometrics, behavior, and identity science","volume":"8 2","pages":"295-310"},"PeriodicalIF":5.0,"publicationDate":"2025-12-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146216656","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"POANET: Parts-Based Occlusion Aware Network for Person Re-Identification","authors":"Pratyay Dutta;Padmaja Jonnalagedda;Bir Bhanu","doi":"10.1109/TBIOM.2025.3641048","DOIUrl":"https://doi.org/10.1109/TBIOM.2025.3641048","url":null,"abstract":"Occlusion remains a major challenge in person re-identification (ReID), as it corrupts feature representations and reduces retrieval accuracy. Most existing methods rely on proxy strategies such as random erasing or pose cues, but they lack explicit modeling of where and how occlusion occurs. To address this gap, we propose two key innovations. First, an Occlusion Detection Network (ODN) explicitly estimates the extent and location of occlusion, enabling the ReID backbone to downweight corrupted regions and emphasize features from visible regions. Second, a Part Aware Feature Extractor (PAFE) generates anatomically correct part embeddings (head, torso, legs), allowing the model to dynamically shift focus from occluded to unoccluded parts in a self-supervised manner while preserving global context. The paper also introduces SHHQ-occ, a large-scale dataset of 1.2M images with precise, part-level occlusion annotations, providing a robust foundation for occlusion modeling. Experiments on occluded benchmark datasets: Occluded-ReID, Occluded-Duke, P-DukeMTMC-ReID, PETHZ and holistic reid datasets: Market-1501, CUHK03, DukeMTMC, MSMT17 demonstrate that our approach achieves state-of-the-art accuracy with strong occlusion robustness and efficiency.","PeriodicalId":73307,"journal":{"name":"IEEE transactions on biometrics, behavior, and identity science","volume":"8 2","pages":"218-231"},"PeriodicalIF":5.0,"publicationDate":"2025-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146216660","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Cueless EEG Imagined Speech for Subject Identification: Dataset and Benchmarks","authors":"Ali Derakhshesh;Zahra Dehghanian;Reza Ebrahimpour;Hamid R. Rabiee","doi":"10.1109/TBIOM.2025.3634273","DOIUrl":"https://doi.org/10.1109/TBIOM.2025.3634273","url":null,"abstract":"Electroencephalogram (EEG) signals have emerged as a promising modality for biometric identification. While previous studies have explored the use of imagined speech with semantically meaningful words for subject identification, most have relied on additional visual or auditory cues to enhance accuracy. In this study, we introduce a cueless EEG-based imagined speech paradigm, where subjects imagine the pronunciation of semantically meaningful words without any external cues related to the target word. This innovative approach addresses the limitations of prior methods by requiring subjects to select and imagine words from a predefined list in a natural manner. The dataset comprises over 4,350 trials from 11 subjects across five sessions. We assess a variety of classification methods, including traditional machine learning techniques such as Support Vector Machines (SVM) and XGBoost, as well as time-series foundation models and deep learning architectures designed explicitly for EEG classification, such as EEG Conformer and Shallow ConvNet. A session-based hold-out validation strategy was employed to ensure reliable evaluation and prevent data leakage. Our results demonstrate outstanding classification accuracy, reaching 99.44%. These findings highlight the potential of cueless EEG paradigms for secure and reliable subject identification in real-world applications, such as brain-computer interfaces (BCIs).","PeriodicalId":73307,"journal":{"name":"IEEE transactions on biometrics, behavior, and identity science","volume":"8 2","pages":"207-217"},"PeriodicalIF":5.0,"publicationDate":"2025-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146216634","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Contactless Fingerprint Presentation Attack Detection Using Kolmogorov–Arnold Networks","authors":"Ashutosh Sharma;Tanuj;Ram Prakash Sharma","doi":"10.1109/TBIOM.2025.3632290","DOIUrl":"https://doi.org/10.1109/TBIOM.2025.3632290","url":null,"abstract":"Fingerprint recognition is a critical component of modern security systems which provides dependable and efficient way of authentication across many domains. However, the increasing prevalence of Presentation Attacks (PAs), where counterfeit fingerprints are exploited to trick recognition systems, offers a substantial security issue. To overcome these vulnerabilities, this study explores the use of Kolmogorov-Arnold Networks (KAN) in combination with Convolutional Neural Networks (CNNs) for improving contactless fingerprint Presentation Attack Detection (PAD). With the use of KAN, number of parameters can be reduced due to their spline functions instead of linear weights which makes it faster in terms of convergence and generalization. Minute fingerprint information and complex feature interactions (ridges, valleys, minutiae points, etc.) can effectively be captured by the proposed hybrid CNN+KAN, allowing the system to discriminate between bona fide and attack presentations under various evaluation protocols. Experiments are performed on varying evaluation settings to asses the performance mimicking real-world attack scenarios. Notably, our proposed model performs well in detecting presentation attacks in baseline and Leave One Out (LOO) evaluation protocols with very low Attack Presentation Classification Error Rate (APCER) at fixed Bona Fide Presentation Classification Error Rate (BPCER) of 1%. The proposed approach achieves minimum average APCER of 0.11% and Detection Equal Error Rate (D-EER) of 0.29% when tested under different evaluation protocol as demonstrated by experimental findings. To provide better insight into the model’s decision-making process, Gradient-weighted Class Activation Mapping (Grad-CAM) visualization was employed to underline the important regions influencing classification decision of the proposed model. The obtained findings underscore the adaptability of proposed Hybrid CNN+KAN in strengthening biometric security.","PeriodicalId":73307,"journal":{"name":"IEEE transactions on biometrics, behavior, and identity science","volume":"8 2","pages":"232-243"},"PeriodicalIF":5.0,"publicationDate":"2025-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146216661","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Novel Unified Body-Based Biometric for Clothing Invariant Human Recognition","authors":"Padmaja Jonnalagedda;Bir Bhanu","doi":"10.1109/TBIOM.2025.3631607","DOIUrl":"https://doi.org/10.1109/TBIOM.2025.3631607","url":null,"abstract":"Body-based human recognition faces increased complexity due to a myriad of challenging factors such as variations in imaging platform (ground- vs. aerial-based), diverse sensors, imaging distance, changes in clothing, arbitrary human poses, occlusion and articulation of the human body, and air turbulence. These conditions often make it difficult to discern facial features, particularly at long distances, where changes in attire also significantly contribute to recognition errors. The most commonly used RGB and shape-based features extracted from color video, suffer significant drawbacks under such conditions. To overcome these challenges, this paper introduces a novel biometric that integrates shape, appearance, and soft biometric features into a single, unified representation called the Outfit Regularizing Biometric (ORB). Through extensive experimentation, we demonstrate that the ORB representation effectively addresses some of the most significant obstacles in human recognition. We report our findings using the rigorous evaluation protocols of the BRIAR datasets (Phase 1 and 2), alongside widely used clothes-changing datasets such as PRCC, LTCC, Celeb-ReID, and VC-Clothes. Our analysis using the BRIAR dataset enables us to isolate and examine the impact of various artifacts, including changes in clothing, environmental conditions, and varying imaging distances. By employing the ORB representation, in combination with both appearance and shape-based features, and leveraging a transformer-based backbone (TransFuse), we achieve superior performance. Across different challenging scenarios, ORB consistently outperforms traditional RGB-based systems, delivering nearly an 18% improvement in accuracy on the standard BRIAR setup.","PeriodicalId":73307,"journal":{"name":"IEEE transactions on biometrics, behavior, and identity science","volume":"8 2","pages":"270-284"},"PeriodicalIF":5.0,"publicationDate":"2025-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146216530","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"PrIdentity: Generalizable Privacy-Preserving Adversarial Perturbations for Anonymizing Facial Identity","authors":"Saheb Chhabra;Kartik Thakral;Richa Singh;Mayank Vatsa","doi":"10.1109/TBIOM.2025.3625986","DOIUrl":"https://doi.org/10.1109/TBIOM.2025.3625986","url":null,"abstract":"With the rapid proliferation of face recognition systems, the risk of privacy leakage from facial images has become a pressing concern. Applications such as Find Face and Social Mapper can readily expose an individual’s identity without consent. Existing anonymization approaches partially address the problem: synthesis and fusion-based methods suppress identity but often distort facial attributes, reducing data utility, while adversarial perturbation methods improve generalizability across recognition models but rely on fixed <inline-formula> <tex-math>$mathcal {L}_{1}$ </tex-math></inline-formula> or <inline-formula> <tex-math>$mathcal {L}_{2}$ </tex-math></inline-formula> norms, leading to underfitting or overfitting. As a result, no single method jointly satisfies the three essential properties of effective anonymization: privacy, data utility, and generalizability. To address these limitations, we present a novel Privacy-Preserving Identity Anonymization (PrIdentity) algorithm that anonymizes the identity of a given image while preserving privacy. Our approach learns adversarial perturbations through an <inline-formula> <tex-math>$mathcal {L}_{p}$ </tex-math></inline-formula> norm based regularization technique, maintaining a balance between privacy and data utility. Furthermore, we ensure the anonymized images generalize effectively across different unseen face recognition models. To the best of our knowledge, this is the first work to introduce a learnable <inline-formula> <tex-math>$p$ </tex-math></inline-formula> parameter in the <inline-formula> <tex-math>$mathcal {L}_{p}$ </tex-math></inline-formula> norm for privacy-preservation. We evaluate PrIdentity on the LFW, CelebA, and CelebA-HQ datasets across multiple face recognition architectures, complemented by a user study on both original and anonymized images. The results demonstrate that our algorithm effectively conceals identities while preserving visual appearance, achieving state-of-the-art performance in identity anonymization. We also carry out bounding box distance prediction experiments to validate data utility, attaining a state-of-the-art Euclidean distance of 2.65, which is 1.17 lower than the second-best method.","PeriodicalId":73307,"journal":{"name":"IEEE transactions on biometrics, behavior, and identity science","volume":"8 1","pages":"137-151"},"PeriodicalIF":5.0,"publicationDate":"2025-10-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146045313","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}