This paper proposes DPPN-Entropy, a dual-path prototypical pretraining network (DPPN) for few-shot EEG–based person identification using precomputed multiscale entropy features. Unlike existing approaches that rely on raw EEG signals, the proposed system operates on compact 180-dimensional entropy feature vectors (30 channels × 6 entropy types), enabling efficient template storage and fast matching in real-world biometric enrollment with limited data. The architecture introduces two key innovations: (1) a DPPN that leverages supervised contrastive learning and mutual distillation between parallel encoder branches to learn robust discriminative embeddings from scarce training data; and (2) a global–local prototype fusion (GLPF) module that adaptively combines class-level global prototypes with sample-level local prototypes via variance-based weighting, with a fallback to global-only prototypes when K = 1 to prevent variance-collapse failure. Under a strict subject-independent evaluation protocol with 16 subjects for meta-training and 10 unseen subjects for meta-testing, DPPN achieves 86.6% accuracy in the 5-way 5-shot setting, significantly outperforming ProtoNet (85.4%) and traditional classifiers (logistic regression 95.1%). Comprehensive downstream analyses—including channel selection, cognitive state comparison (eyes-open (EO) significantly higher than eyes-closed (EC)), N-way scalability, and biometric metrics (EER 21.2%, AUC 0.843)—establish practical benchmarks for few-shot EEG person identification. Although the framework is termed MPAN-entropy, ablation studies identify DPPN as the primary driver of performance; we thus adopt DPPN for all downstream analyses.
The outermost layer of the skin, stratum corneum, functions as a critical barrier. Its hydration levels are being actively affected by sweat gland activity, which is a significant mechanism for thermoregulation during physical exertion. Despite the known relationship, a direct, noninvasive method for monitoring dynamic changes in skin hydration with common wearable devices has remained elusive. Prior studies rely on specialized biosensors or preinduced hydration states to formulate the problem as a supervised classification, which lacks generalizability for continuous, free-living assessment. Our study utilized the photoplethysmography (PPG)-DaLiA publicly available dataset to develop a mathematical model that computes a continuous, relative skin hydration proxy by integrating electrodermal activity (EDA) and skin temperature signals from a common multimodal wrist-worn device (Empatica E4). Despite the lack of direct hydration ground truth, the model’s output was validated by its physiological responses across various activities, exhibiting stable and interpretable trends. This study has significant practical implications, as it demonstrates a feasible approach to adapting affordable, consumer-grade wristbands into effective tools for daily hydration monitoring, utilizing their current multimodal sensors without necessitating specialized hardware or incurring additional costs, thus fostering proactive personal health management.
Human walking balance relies on proprioception, vision, and the vestibular sense. Dysfunction in any of these systems can lead to unstable gait. Accurately identifying sensory disturbances is key to understanding the causes of sensory dysfunction. Most existing studies focus on disturbances in a single sensory system and lack methods to identify multiple disturbance types. This paper proposes a recognition method based on Newton-Raphson-based optimizer (NRBO)-convolutional neural network (CNN)-squeeze-and-excitation (SE)-long short-term memory (LSTM). The original feature set is filtered through multiple rounds of hybrid filtering-wrapping feature selection algorithm (HLOA-HFWA) to obtain a feature subset. The union of these features forms an integrated feature (IF) set. This set is used as input to a CNN-SE-LSTM model designed for multiclassification. The Newton-Raphson method (NRBO) is then used to optimize key hyperparameters, including learning rate (LR), LSTM hidden node number, and L2 regularization (L2R) coefficient. This improves the network's feature extraction and pattern recognition capabilities. Experiments on public IF sets and gait-IF sets show that NRBO-CNN-SE-LSTM outperforms other models in both sensory disturbance recognition and overall performance.
Accurate speaker identification is critical for forensic investigations and biometric authentication. This study introduces Lightweight squeeze-and-excitation network (LSE-Net), a compact, robust deep learning architecture specifically tailored for speaker identification in challenging, noisy environments. Our approach integrates four core design principles: RMS-energy-based voice activity detection (VAD) for signal refinement, standardized Log-Mel spectrogram extraction, noise-aware regularization with spectral data augmentation to enhance generalization, and channel-wise squeeze-excitation (SE) blocks that dynamically recalibrate feature maps to emphasize speaker-discriminative information. Finally, an utterance-level probability aggregation strategy ensures robust classification of entire audio recordings. We evaluated LSE-Net on two datasets: the newly constructed PERSID33 (933 formal Persian news-style utterances from 33 speakers) and the benchmark TIMIT dataset (6300 utterances from 630 speakers). Experiments were conducted across diverse conditions, including clean audio and various real-world and synthetic noise scenarios at multiple signal-to-noise ratios. LSE-Net achieved 100% accuracy on PERSID33 and 95.07% on TIMIT under noise-free conditions, consistently outperforming baseline methods. In noisy settings, LSE-Net maintained near-perfect accuracy on the PERSID33 dataset and demonstrated superior robustness on the larger TIMIT dataset compared to state-of-the-art (SOTA) approaches. Ablation studies confirmed that spectral augmentation and SE attention are pivotal in preserving identification accuracy under acoustic stress. These results demonstrate that LSE-Net is a highly efficient, forensic-grade solution for speaker identification, particularly suitable for constrained or noisy investigative contexts.
Human activity recognition (HAR) is a core technology in fields such as smart healthcare and human-computer interaction, which aims to classify daily activities (e.g., walking and running) based on sensor data automatically. While existing approaches achieve high accuracy in controlled laboratory settings, they often perform poorly in real-world applications due to limited multiscale feature extraction and poor modeling of interchannel sensor correlations. These shortcomings lead to high sensitivity to noise and irrelevant time segments. An innovative model integrating multiscale convolution, bidirectional long short-term memory (BiLSTM), and a spatiotemporal attention mechanism is proposed to address these issues in this research. The model employs a multiscale parallel convolutional structure with filter sizes of 3, 5, 7, and 9 that enable it to capture both short-term local dependencies and long-term global patterns simultaneously. The introduction of a spatial-temporal dual attention mechanism dynamically focuses on key sensor channels and time segments, significantly improving the accuracy of feature selection. In addition, the Swish activation function is used to optimize the feature extraction process. Its smooth characteristics and self-gating mechanism avoid the gradient vanishing problem of the traditional ReLU effectively. The experimental results show that the model achieves an accuracy of 95.39% on the UCI-HAR public dataset and an excellent performance of 99.58% on the custom dataset.
The Internet of Medical Things (IoMT) has emerged as a transformative force in healthcare, enabling the development of innovative and noninvasive diagnostic tools that leverage connected technologies. This research focuses on the application of IoMT in conjunction with deep learning techniques to detect and classify heart diseases through the analysis of heart sounds, a method that promises to enhance the accuracy and accessibility of diagnostics. We proposed frameworks that utilize advanced signal processing algorithms to capture and analyze heart sound data, that is, phonocardiograms (PCGs) collected from District Headquarters Hospital, Bahawalnagar, Pakistan. Deep learning models, specifically convolutional neural networks (CNN) and long short-term memory (LSTM), and a combination of these are employed to identify patterns associated with various cardiac conditions. This study involves a comprehensive dataset comprising annotated heart sounds from diverse demographics, ensuring the robustness and generalizability of the proposed models. The proposed methodology involves using three models: one-dimensional CNN (1D-CNN), LSTM, and a combination of 1D-CNN + LSTM to classify heart sounds into normal, murmur, extra-systole, and artifact categories. Results demonstrate high accuracy in the classification of common heart diseases, including normal, murmur, extra-systole, and artifact, surpassing traditional diagnostic methods. The model achieved the highest accuracy for normal cases as of 95%, for murmur cases as of 94%, for extra-systole cases as of 95%, and for artifact cases as of 96%. This approach holds significant potential for improving patient outcomes through timely and accurate diagnostics, thus exemplifying the power of digital health innovations in modern medicine.
Throughout the history of automated personal identification, now called “biometrics” in some communities, there has been controversy over its implications for personal privacy and human dignity. This controversy has been deepened by equivocation regarding the philosophical concept of personal identity, the social concept of recognition of persons, and the forensic concept of person identification. These concepts of personal identity, recognition, and identification are central to Western literature, shared traditions, and our notions of “self,” so it is not surprising that technologies dealing with these issues would elicit controversy. The assumption of this paper is that greater understanding of the cultural meanings and nuances of these concepts is necessary to illuminate some critical aspects of the current ethical and policy debate surrounding biometric technology. Accordingly, we attempt to deconfound these core concepts by differentiating persons from attributes and labels from notions of self. We explicate two different but interrelated forms of identity and discuss their connection to recognition, identifiers, and identification. Clear definitions are presented for each of these concepts. With the intent of enlightening the complicated relationship society and individuals have with biometric technology, we discuss social risks and suggest that the complexities of identity and recognition have always been important in Western tradition, as illustrated in the Biblical story of Abraham and the Greek story of Odysseus. We conclude that biometric technologies, although not without risks, have a legitimate place in modern society for assigning identifiers to persons based on body characteristics and should not be understood as extending to the philosophical and religious foundations of personal concepts of self.
Face recognition in unconstrained environments remains a challenging problem due to significant variations in facial appearance. In this article, a systematic investigation is conducted into the effectiveness of integrating attention mechanisms within a ResNet34 backbone. ResNet34 is adopted as a compact backbone compared to commonly used deeper ResNet variants, offering a good balance between representational capacity and computational cost. Specifically, three variants are examined: self-attention (SA), contextual transformer (CoT) attention, and the proposed dilated CoT (DCoT) attention. The primary contribution of this study is the demonstration that DCoT, by enlarging the receptive field without substantially increasing the number of parameters, constitutes a more suitable architectural choice for this backbone. This integration enables the network to capture broader contextual information and attend more effectively to discriminative facial features. The proposed models are trained on the CASIA-WebFace dataset and evaluated on standard benchmarks, including labled faces in the wild (LFW), CPLFW, and AgeDB_30. Extensive experiments show that the DCoT-enhanced model achieves consistent improvements over the baseline and other attention variants. Notably, the best-performing model attains an accuracy of 99.60% on the LFW dataset. These findings provide meaningful insights into the interaction between attention mechanisms and convolutional architectures, illustrating the efficacy of the proposed approach for face recognition in unconstrained, real-world conditions, while using a moderately sized backbone that balances capacity and parameter count.
Cross-domain fingerprint recognition remains difficult when latent impressions and smartphone finger photos are matched against contact-based sensor templates. We study this problem in the latent fingerprints in the wild (LFIW) database. We benchmark five representative matchers on eight latent- and finger-photo-to-sensor protocols. The results confirm a clear gap: no single matcher is robust across all domains. We then introduce two domain-aware components. First, domain-normalized quality score (DNQS) standardizes existing quality metrics within each domain. This reduces domain bias and improves the agreement between quality and verification error. Second, quality-weighted score fusion (QWSF) uses DNQS to fuse two complementary matchers, MCC and Neurotechnology. QWSF reduces errors across all protocols. It achieves a 15% relative equal error rate (EER) reduction on iPad vs. Opt and a 7% reduction on average. Paired bootstrap confidence intervals support the reliability of the gains. Improvements also hold at a low-FMR operating point (FNMR at FMR = 10-3). These findings show that quality normalization and quality-aware fusion can improve cross-domain robustness without retraining matchers.
Unified face attack detection (UAD) systems, which utilize vision-language models to simultaneously discriminate between physical and digital attacks, remain challenging due to the difficulty in effectively identifying live information and the biological cues forged by a variety of distinct technologies. The difficulty stems from two primary aspects: (1) Single spatial information is insufficient for capturing comprehensive cues regarding both genuine and forged identities. (2) Textual prompts struggle to acquire low-redundancy and complementary information to facilitate the discrimination between live and fake cues. To address these issues, we propose a novel frequency-aware and ensembled text prompt UAD model built upon the contrastive language-image pretraining (CLIP) framework. Our model adaptively fuses spatial and frequency information to enhance the representation of genuine faces and all types of attacks, while simultaneously introducing an ensemble learning strategy to acquire low-redundancy textual prompts. Specifically, the ensemble prompt module generates general live and fake prompts from spatial and frequency information on the language branch, thus guiding the model to learn a unified feature space to deal with different attacks. Meanwhile, this module optimizes the redundancy and complementarity between prompts through an ensemble strategy and a designed information diversity constraint. Furthermore, we design a layer-wise cross-attention module in the vision branch to fuse frequency information from different layers. A designed redundancy minimization module is employed on the fused image features, thereby compelling the spatial and frequency feature extraction modules to generate maximally exclusive features. Extensive experiments on multiple benchmarks demonstrate that our model achieves state-of-the-art performance across most protocols of the datasets.
Despite major advancements in the sphere of the public-key authentication specifically in the instances of the newly established standards like WebAuthn and the FIDO2, the practical implementation of the passwordless login systems is still hindered by the usability factors, platform-related requirements, and the very nature of the deployment process is predetermined by its complex character. We formulate in this contribution a practical, push-based authentication model, which takes active advantage of the utilization of the public-key cryptography, combined with the traditional set of capabilities in mobile devices, to provide a secure and consumer-friendly way of logging into a system. The architecture has been implemented using the Laravel web application framework, the Flutter cross-platform mobile development system, and Firebase Cloud Messaging (FCM), thus enabling authentication using smartphone-initiated approvals and digital signatures. A within-subject usability experiment with 160 participants was the measure of effectiveness of the proposed approach. The participants were requested to fulfill the tasks of logins with the postulated push-based mechanism and an experimental password-based one. System logs provided objective metrics such as the latency to log in, rate of errors, and the success rate of the first attempt. At the same time, the subjective ratings were gathered through the means of structured questionnaires where the ease of use, speed, security, and preferences were measured. Empirical data revealed that despite the insignificant increase in time of login, the passwordless system reported a significant decrease in error, increase in success rate, and a significant preference among the subjects in all the dimensions of usability that were being tested. Each of the observed differences was statistically significant and with large effect sizes. This research provides a deplorable, platform adaptable authentication paradigm that preempts realistic coping, yet holds core security standards. Emphasizing the need to balance system design with human-centered assessment the work suggests a possible solution to more complex or more hardware-resources limited passwordless usage.
With the rise of sophisticated image manipulation techniques, image detection forgeries have become increasingly challenging. This paper presents a novel deep learning framework, the multibranch collaboration and segmented training network (MBC-STN), for robust image forgery detection. MBC-STN addresses the challenges posed by diverse forgery types and complex noisy environments through a multibranch collaborative architecture that integrates three specialized branches: the forgery edge-aware branch (FEB), the noise-aware branch (NB), and the color-aware branch (CB). These branches capture image forgery features from multiple dimensions, significantly enhancing the precision and robustness of forgery detection. Additionally, MBC-STN employs a segmented training strategy to optimize the performance of different branches in stages, improving training efficiency, and model adaptability. Three loss functions (balanced binary cross-entropy loss, balanced dice loss, and binary cross-entropy loss) are employed to guide MBC-STN in learning manipulated traces and identifying forgery regions. Extensive experiments on multiple datasets demonstrate that MBC-STN outperforms state-of-the-art methods maintain robustness under different postprocessing conditions.
Face antispoofing (FAS) technologies play a pivotal role in safeguarding face recognition (FR) systems against potential security loopholes. The biometric community has witnessed significant advancements lately, largely due to the exceptional performance of deep learning architectures and the abundance of substantial datasets. Despite these progress, FR systems remain susceptible to both physical and digital forgery attacks. However, most existing detection methods do not address both types of threats concurrently. To bridge this gap and foster the development of a comprehensive detection framework, we have compiled a unified dataset named UniAttackData. This dataset incorporates both physical and digital spoofing attacks while maintaining identity consistency, encompassing 1800 participants each subjected to two different physical attacks (PAs) and 12 different digital attacks (DAs), respectively. This effort has resulted in a comprehensive collection of 29,706 video samples. We organized the Chalearn FAS face attack detection challenge based on this novel resource to boost research aiming to promote joint antispoofing efforts. The Chalearn unified antispoofing attack detection challenge drew 136 teams during the development phase, with 13 teams advancing to the final round. The organizing team revalidated and re-executed the submitted code to determine the final rankings. This paper provides a summary of the challenge, covering the dataset used, the protocol definition, the evaluation metrics, and the competition results. Additionally, we discuss the top-ranked algorithms and the research insights offered by the competition for attack detection.
Facial expression recognition is vital in pattern recognition and affective computing. With the advancement of deep learning, its performance has improved, yet challenges remain in nonlaboratory environments due to occlusion, poor lighting, and varying head poses. This study explores a robust facial expression recognition approach using a CNN-based model integrated with key point localization techniques. Instead of relying on a dense set of landmarks, the proposed method focuses on fewer but more informative expression key points. Each point is analyzed for local shape features, and contour consistency is verified using indexing along the normal direction. This strategy enhances robustness while reducing computational complexity. Specifically, the hybrid active shape model (ASM) + structure method significantly lowers the processing load compared to the traditional ASM approach. Experimental results demonstrate a 3.02% improvement in recognition accuracy over one-to-many SVM classifiers when dealing with clear facial images. Furthermore, the system shows strong resilience to partial occlusions and maintains real-time performance, making it suitable for real-world applications. The proposed framework highlights the importance of selecting effective key points and optimizing feature extraction to enhance both accuracy and efficiency in facial expression recognition tasks under challenging conditions.
Deep learning has significantly improved the performance of fingerprint liveness detection, while susceptibility to adversarial attacks remains a critical security challenge. Existing input transformation-based defense methods, including JPEG compression, total variance minimization (TVM), high-level representation guided denoiser (HGD), and Defed, are typically designed for specific attacks, resulting in limited generalization across diverse adversarial scenarios. Experimental analysis indicates that among the four defense methods based on input transformation, Defed achieves the best overall performance when evaluated against both momentum iterative fast gradient sign method (MI-FGSM) and DeepFool attacks. However, Defed exhibits strong robustness against MI-FGSM attacks but demonstrates insufficient defense effectiveness against DeepFool attacks. To address this issue, an improved method of Defed has been proposed by integrating a learnable Gaussian noise module into the core structure to enable adaptive suppression of adversarial perturbations, and by employing 1 x 1 convolutions to allow cross-channel information interaction, thereby enhancing feature consistency and overall robustness. Experimental results on the LivDet 2015 dataset demonstrate that the defense success rate against DeepFool attacks has increased by 3%-5%, while strong robustness against MI-FGSM attacks has been maintained, substantially improving the security and reliability of fingerprint liveness detection systems.
This study introduces a reinforcement training framework for face recognition systems (FRSs) that leverages facial morphing techniques to generate counterfactual visual instances for model enhancement. Two complementary morphing strategies were employed: a geometric approach based on Delaunay–Voronoi triangulation (DVT-Morph) and a generative approach using latent diffusion and autoencoder-based models (diffusion-based morphing [MorDIFF]). The generated morphs act as controlled counterfactuals, representing minimally modified facial images that induce changes in FRS verification decisions. The proposed method integrates these counterfactuals into the training process of two state-of-the-art recognition systems, ArcFace and MagFace, to strengthen their decision boundaries and improve their robustness, calibration, and explainability. By combining morphing-based counterfactual generation with eXplainable Artificial Intelligence (XAI) techniques, the framework enables a more interpretable embedding space and increased resilience against morphing and adversarial perturbations. The experimental results demonstrate that the inclusion of morph-based counterfactuals significantly enhances the verification accuracy and decision transparency of modern FRSs. Moreover, the methodology is model- and morphing-agnostic and can be applied to any FRS architecture, regardless of the morphing generation technique.
Large fingerprint datasets, while important for training and evaluation, are time-consuming and expensive to collect and require strict privacy measures. Researchers are exploring the use of synthetic fingerprint data to address these issues. This article presents a novel approach for generating synthetic fingerprint images (both spoof and live), addressing concerns related to privacy, cost, and accessibility in biometric data collection. Our approach utilizes conditional StyleGAN2-ADA and StyleGAN3 architectures to produce high-resolution synthetic live fingerprints, conditioned on specific finger identities (thumb through little finger). Additionally, we employ CycleGANs to translate these into realistic spoof fingerprints, simulating a variety of presentation attack materials (e.g., EcoFlex, Play-Doh). These synthetic spoof fingerprints are crucial for developing robust spoof detection systems. Through these generative models, we created two synthetic datasets (DB2 and DB3), each containing 1500 fingerprint images of all 10 fingers with multiple impressions per finger, and including corresponding spoofs in eight material types. The results indicate robust performance: our StyleGAN3 model achieves a Fréchet inception distance (FID) as low as 5, and the generated fingerprints achieve a true acceptance rate (TAR) of 99.47% at a 0.01% false acceptance rate (FAR). The StyleGAN2-ADA model achieved a TAR of 98.67% at the same 0.01% FAR. We assess fingerprint quality using standard metrics (NFIQ2, MINDTCT), and notably, matching experiments confirm strong privacy preservation, with no significant evidence of identity leakage, confirming the strong privacy-preserving properties of our synthetic datasets.
In the face of the mounting challenges posed by cybersecurity threats, there is an imperative for the development of robust identity authentication systems to safeguard sensitive user data. Conventional biometric authentication methods, such as fingerprinting and facial recognition, are vulnerable to spoofing attacks. In contrast, electrocardiogram (ECG) signals offer distinct advantages as dynamic, “liveness”-assured biomarkers, exhibiting individual specificity. This study proposes a novel fusion network model, the convolutional neural network (CNN)-transformer fusion network (CTFN), to achieve high-precision ECG-based identity authentication by synergizing local feature extraction and global signal correlation analysis. The proposed framework integrates a multistage enhanced CNN to capture fine-grained local patterns in ECG morphology and a transformer encoder to model long-range dependencies in heartbeat sequences. An adaptive weighting mechanism dynamically optimizes the contributions of both modules during feature fusion. The efficacy of CTFN was evaluated in three critical real-world scenarios: single/multi-heartbeat authentication, cross-temporal consistency, and emotional variability resistance. The evaluation was conducted on 283 subjects from four public ECG databases: CYBHi, PTB, ECG-ID, and MIT-BIH. The CYBHi dataset revealed that CTFN exhibited a state-of-the-art recognition accuracy of 98.46%, 80.95%, and 90.76%, respectively, signifying its remarkable performance. Notably, the model attained a 100% authentication accuracy rate using only six heartbeats. This represents a 25% decrease in input requirements when compared to prior works, while concurrently maintaining its robust performance against physiological variations induced by emotional states or temporal gaps. These results demonstrate that CTFN significantly advances the practicality of ECG biometrics by balancing high accuracy with minimal data acquisition demands, offering a scalable and spoof-resistant solution for secure authentication systems.
Unified face attack detection (UAD), which simultaneously addresses physical presentation attacks (PAs) and digital forgery attacks (DAs) with a vision-language model, remains challenging due to the difficulty of effectively separating live and fake cues. The challenges mainly arise from two aspects: (1) text prompts are insufficiently aligned with visual features across layers, and (2) patch tokens containing live and fake cues often overlap, leading to ambiguous attribution and small decision margins. To address these problems, we propose a novel Layer-wise Cue Alignment framework (LCA) that leverages textual features to extract both layer-wise and global cues from patch tokens, and we further introduce a new training strategy to improve the separation of live and fake cues. Specifically, the layer-wise prompts are obtained by the cue matching block (CMB), which matches textual features with patch embeddings at each transformer layer, and the layer-level cues are injected into the visual features of each layer and further aggregated by the cue fusion block (CFB) to form comprehensive prompts that enhance the overall visual representation. Moreover, we design a complementary supervision mechanism (CSM) that suppresses forgery cues in live faces while enforcing mutual exclusivity between live and fake cues in attack samples to improve the reliability of cue separation. Extensive experiments on multiple benchmarks demonstrate that our framework achieves state-of-the-art performance on most protocols of the datasets.
