This paper tends to serve as a concise review of the recent literature on multimodal biometrics recognition systems, in particular, on the fusion strategies that are utilized at each stage of biometric data processing. These are feature-level, score-level, decision-level, serial, and hybrid fusions. The survey underscores that the fusion of multiple biometric traits, including fingerprints, face, iris, palmprints, voice, signature, etc., is more effective in increasing system performance, reliability, and spoof resistance than unimodal techniques. Classifying the studies based on the fusion strategies that they followed, this paper reviews the techniques, performance indicators, datasets, and application of the reviewed studies. The score-level fusion method gave the best reported accuracy of 100%, and the serial fusion obtained lower accuracy owing to the limitation of adaptability and dataset dependency. The review discusses and describes common problems and future directions for research as well. The insights drawn are targeted towards possibly the more secure, efficient and versatile multimodal biometric systems as far as the real-world applications are concerned.

Abstract Image compression techniques are essential for efficient storage and transmission; however, they often lead to significant degradation in image quality, including the loss of details and the introduction of visual artifacts. The research proposes a novel approach for optimizing compressed image quality and enhancing features using Singular Value Decomposition (SVD) combined with Deep Learning (DL). The objective is to address the limitations of traditional compression methods, which frequently compromise crucial image elements such as edges, textures, and contrast. In the proposed method, the input image is first processed using SVD to extract its fundamental structural components. A threshold-based modification is then applied to the singular values to enhance the representation of low contrast and low resolution features, commonly observed in compressed surveillance images. The images are resized and denoised using a Bilateral Filter to standardize dimensions for uniform processing. Subsequently, the Adaptive Bird Swarm-driven Compressed Singular Value-Enriched VariationalAutoencoder (ABS-CSV-VAE) technique is used to increase the compression ratio and decrease redundancy without significant loss of perceptual quality. The performance of the proposed ABS-CSV-VAE method is evaluated using a dataset of low quality images. Quantitative metrics such as loss (0.8223), PSNR (34.474 dB), SSIM (0.895), BPP (0.4134), parameter count (700 K), and MACs of 51.32 G. The results indicate that the proposed method achieves a PSNR improvement compared to traditional algorithms while restoring key facial features more effectively. Experimental findings confirm that the technique enhances the visual quality of compressed images and preserves essential features, making it highly suitable for applications in video surveillance, biometric recognition, and low bandwidth image transmission systems.

The Integrated Smart Campus Bus Management System (ISC-BMS) has been developed to provide a comprehensive and effective approach to managing transportation operations through the use of intelligent automation and real time monitoring. The various functional modules are combined into one system including driver log-in based attendance; Student Bus Pass Application Workflows, and Centralized Oversight Administrator Dashboard. It is proposed that using a face recognition subsystem will automate the process of taking attendance as students board the buses thereby improving accuracy in seating arrangements, reducing the opportunity for proxies, and increasing overall safety and security on campus. Real-time GPS location tracking will provide constant visibility of bus routes, as well as notification of the movement of the bus, delays, and whether the student is on board or not to parents. Several key limitations are addressed by the proposed framework including the limitations of current systems that only track or schedule services separately without incorporating any behavioral verification or automating identity authentication process(es). By integrating the above with biometric recognition, digital pass processing, and multi-role communication capabilities all into a single platform; school districts will be able to improve the safety, operational transparency, and user experience of their transportation operations thereby creating a more reliable and sustainable campus transportation management system.

This study presents a secure machine learning model for drug authentication to combat the proliferation of counterfeit pharmaceuticals. A web-based system was developed by integrating a FastTree binary classification model in ML.NET with biometric facial recognition for user authentication. The system enables real-time verification of pharmaceutical products using structured metadata, including batch numbers, expiration dates, manufacturer identifiers, and barcodes. A dataset obtained from the U.S. Food and Drug Administration’s OpenFDA repository was used for model training and evaluation, with preprocessing implemented through a C# schema class. To enhance system security and prevent unauthorized access, a facial recognition module was incorporated as an additional authentication layer. Performance evaluation using 10-fold cross-validation yielded strong results, achieving 97.8% accuracy, an F1-score of 96.8%, and an AUC of 0.981. The proposed system provides a lightweight, scalable, and secure framework that integrates machine-learning-based drug authentication with biometric access control. This approach enhances the reliability, integrity, and security of pharmaceutical verification systems. Future work may explore larger datasets and blockchain integration for improved traceability

Biometric recognition of human gait is a promising, non-invasive method for the identification of people that does not require their engagement. Existing solutions mainly focus on the identification effectiveness under laboratory conditions, frequently overlooking factors that disrupt the gait of test subjects. The present work considers the issue of identifying a person on the basis of ground reaction forces in cases where their gait is disrupted through asymmetric loading. This paper proposes a solution based on ensemble classifiers utilizing various types of deep neural networks as base classifiers. To further increase the ability to generalize base models, data augmentation was used. The proposed solution was tested on a sample of 215 people (7351 gait cycles) and two strategies for combining classifier decisions. The accuracy results obtained, ranging between 99.8, 98.55, and 98.85% correct recognitions depending on the scenario analyzed, are very good and significantly exceed other methods presented in the literature to date.

Retraction Note: Multimodal of Ear and Face Biometric Recognition Using Adaptive Approach Runge–Kutta Threshold segmentation and Classifier with Score Level Fusion

Finger vein recognition has emerged as a highly robust and intrinsically stable biometric technology, demonstrating great potential in identity authentication and intelligent security applications. However, conventional methods still suffer from constraints in feature representation and computational efficiency, particularly under challenging conditions such as illumination variation, pose deviation, and noise interference. To address these challenges, this study presents an efficient finger vein recognition approach based on a lightweight convolutional neural network (LCNN) architecture. The proposed framework integrates a multi-stage image preprocessing pipeline for automatic vein region detection, advanced denoising, and refined texture enhancement, which is subsequently followed by compact feature modeling within a lightweight deep network. Extensive experiments on the public Shandong University Machine Learning and Applications-Homologous Multi-Modal Traits (SDUMLA-HMT) dataset and a self-acquired Laboratory Finger-Vein (Lab-Vein) dataset validate the superiority of the proposed method, achieving recognition accuracies of 97.1% and 98.3%, respectively, surpassing existing benchmark models. Moreover, the model demonstrates notable reductions in parameter complexity and computational cost, achieving an average inference time of only 12.6 ms, which confirms its strong real-time capability and suitability for embedded deployment. Overall, the proposed approach attains a desirable trade-off between accuracy and efficiency, offering meaningful implications for the advancement of lightweight biometric recognition systems.

Existing biometric systems are predominantly built upon 2D biometrics which are vulnerable to presentation attacks, and have a limited coverage of the biometric surface. 3D biometrics is emerging due to the rapid development of 3D sensing technology. 3D biometrics is effective in defending against spoofing attacks and potentially offer more robust performance under different conditions. However, there exist many challenges in developing effective 3D biometric systems in terms of 3D biometrics reconstruction and recognition. In this paper, we present a systematic survey on the latest developments of 3D biometric systems guided by our proposed taxonomy of methods for 3D biometrics.

Advances in optoelectronic synapses (OES) have relied on complex device configurations and fabrication processes, which limit their practical implementation. Here, we exploit the untapped potential of ultrathin Bi2Te3 to construct a multifunctional OES device for a range of applications in neuromorphic computing, biometric recognition, and artificial visual perception. The Te vacancies in the film trap and de-trap charges, leading to persistent photoconductivity as the operating mechanism. Specifically, we demonstrate successful defect engineering by controlling the annealing temperature of the Bi2Te3 films and directly correlate the OES performance with the defect density. The role of the Te vacancies in OES is further confirmed by first-principles calculations. The OES devices show excellent metrics such as 191.7% paired-pulse facilitation and 37.2 fJ per spike of energy consumption. The device successfully simulates Pavlov's classic associative learning experiment. A 6 × 6 device array, serving as an artificial retina for image processing, displays excellent retention of the learned optical information and memory performance by 57.4%. The OES devices demonstrate high accuracy in facial recognition (93.3%) and urban traffic scene segmentation (86.7%) tasks after 100 epochs. Finally, successful optical logic gate operations and Morse code for optical signal recognition and wireless communication are demonstrated using the OES devices.

Forensic odontology research has shown that teeth are the hardest structures in the human body and can serve as reliable indicators for individual identification. In this study, we propose a Point Transformer based algorithm called MultiDentFormer, which leverages local-to-global feature fusion to achieve 3D dental biometric recognition. The global features refer to the high-level semantic representations of the full dental arch, capturing the overall structure and arrangement of the teeth, while the local features focus on the detailed geometry of individual teeth. To address the limited capability of the original Point Transformer in capturing local information at the early stages of the network, this study introduces a single-point local feature aggregation module. This module aggregates local geometric information and low-level feature representations into each local center point, thereby enhancing the expressiveness of the extracted features. Additionally, to address the issue of detail information loss during the network downsampling process, which weakens feature representation capability, this paper introduces a multi-scale feature fusion strategy. This strategy effectively integrates the rich detail features from shallow layers with the high-level semantic features extracted from deeper layers, compensating for the information lost during downsampling. Experimental evaluations on a IOS dataset comprising 104 individuals and a CBCT dataset with 51 individuals demonstrate that MultiDentFormer achieves superior recognition performance compared to common algorithms in 3D point cloud classification. The local feature aggregation strategy and multi-scale feature fusion strategy enhance the discriminative ability and classification performance of the 3D point cloud feature representations.

Abstract Due to their ease of use and computational effectiveness, traditional face recognition methods like Local Binary Patterns (LBP) and Principal Component Analysis (PCA) have been thoroughly researched conventional face recognition techniques such as Local Binary Patterns (LBP) and Principal Component Analysis (PCA) are widely used because they are simple to implement and computationally efficient. However, their performance in real-world situations is still limited. This paper presents an experimental study to examine how elements like changes in circumstances of lighting, variations in face pose, complex backgrounds influence the performance of conventional face recognition methods. Recognition accuracy is assessed using benchmark datasets representing controlled, semi-controlled, and unconstrained environments. The experimental results show that classical methods perform well in controlled conditions but their accuracy decreases significantly when illumination varies or when backgrounds become complex. These findings highlight the limitations of manually designed feature-based approaches and indicate the need for more advanced representation learning methods for reliable biometric recognition. Keywords: Face Recognition, Illumination Variation, Pose Variation, Background Complexity, PCA, LBP, Classical Biometrics

This study delves into the vulnerabilities of deep learning-based gait recognition systems against adversarial attacks, a critical issue considering the increasing reliance on these technologies in high-security environments. We highlight a major issue concerning the susceptibility of these systems to adversarial interventions that compromise their reliability. The importance of this issue stems from the critical role of gait recognition in applications where security and accuracy are paramount. Our approach introduces an advanced methodology that integrates Proximal Policy Optimization (PPO) with Generative Adversarial Networks (GANs) to create and deploy adversarial attacks in the form of targeted adversarial patches. These patches are designed to deceive gait recognition algorithms without detection by human oversight, exploiting the models' weaknesses to induce misclassification. This methodology not only leverages the strengths of GANs to produce deceptive examples but also innovatively utilizes PPO to ascertain their optimal placements, thereby maximizing the disruption on gait recognition systems. We assess the impact of these attacks using the CASIA Gait Database: Dataset B and the OU-ISIR Treadmill Dataset B - Clothes variation-, covering both real-world and controlled environments. Our results demonstrate a significant decline in recognition accuracy post-attack, underscoring the effectiveness of our adversarial approach. These findings underscore critical security flaws and actively inform the broader discussion aimed at boosting the robustness of gait recognition systems. The impact of our research extends significantly, providing crucial insights that aid in the creation of more secure, attack-resistant biometric recognition systems, thereby enhancing the resilience of gait recognition technologies against the backdrop of advancing cyber threats.

Deep-learning models, including those used in biometric recognition, have achieved remarkable performance on benchmark datasets as well as real-world recognition tasks. However, a major drawback of these models is their lack of transparency in decision-making. Mechanistic interpretability has emerged as a promising research field intended to help us gain insights into such models, but its application to biometric data remains limited. In this work, we bridge this gap by introducing the FaceMINT library, a publicly available Python library (build on top of Pytorch) that enables biometric researchers to inspect their models through mechanistic interpretability. It provides a plug-and-play solution that allows researchers to seamlessly switch between the analyzed biometric models, evaluate state-of-the-art sparse autoencoders, select from various image parametrizations, and fine-tune hyperparameters. Using a large scale Glint360K dataset, we demonstrate the usability of FaceMINT by applying its functionality to two state-of-the-art (deep-learning) face recognition models: AdaFace, based on Convolutional Neural Networks (CNN), and SwinFace, based on transformers. The proposed library implements various sparse auto-encoders (SAEs), including vanilla SAE, Gated SAE, JumpReLU SAE, and TopK SAE, which have achieved state-of-the-art results in the mechanistic interpretability of large language models. Our study highlights the promise of mechanistic interpretability in the biometric field, providing new avenues for researchers to explore model transparency and refine biometric recognition systems. The library is publicly available at www.gitlab.com/peterrot/facemint . • We have developed a library for the mechanistic interpretability of face rec. models. • It supports CNN and transformer architectures for face recognition analysis. • Integrates sparse autoencoders to reveal disentangled latent features. • Offers activation maximization with multiple image parametrizations. • We present initial findings in feature discovery, validated by a human survey.

A novel physics-informed multimodal biometric recognition framework that unifies iris and fingerprint modalities through a physically interpretable computational architecture. Conventional unimodal biometric systems are often constrained by intra-class variability, environmental distortions, and spoofing vulnerabilities. To address these, we model fingerprint textures as quasi-static surface deformations influenced by erosion and pressure dynamics, while the iris is treated as a dynamic optical structure modulated by pupil dilation and angular displacements. These physical analogs inform the feature extraction process, which employs the Wavelet Scattering Transform to capture frequency and motion-invariant features across spatial scales. A Siamese Neural Network is trained to perform metric-based classification, discerning identity through an abstract similarity embedding. Quantitative results demonstrate substantial gains: at 150 training epochs, the model achieves 98.4% training, 98.1% validation, and 98.0% test accuracy, with minimal loss. Furthermore, a SHAP based interpretability module yields a Decision Robustness of 95.73% and a Feature Attribution Strength of 91.23%, confirming the model's transparency and consistency. This interdisciplinary approach, rooted in dynamic systems theory, wave-based feature modeling, and interpretable deep learning, offers a promising pathway for resilient and explainable biometric authentication under real-world variabilities.

The uniqueness of the electroencephalogram (EEG), a distinct biometric marker inherent to each individual, yields significant advantages for user authentication and identification in brain-computer interface (BCI) systems. However, EEG features can easily change according to the user's state, which may affect the performance of biometric recognition systems based on EEG. Notably, in EEG data collection for such systems, fatigue levels can fluctuate over time-a factor that has yet to be thoroughly investigated concerning its impact on individual recognition performance. This study explores the implications of fatigue on EEG-based personal recognition systems. We derived six sub-datasets from two simulated driving datasets, each labeled with varying levels of fatigue. From each sub-dataset, we extracted six features for identity recognition within and across different fatigue levels. Single-session and cross-session studies revealed that the disparity of EEG fatigue levels between the training and testing sets increased, and system recognition accuracy experienced a decline. Specifically, recognition accuracy typically fell by over 30 after 90min of simulated driving. Furthermore, identity recognition results are better when the training set includes EEG in more fatigued states compared to the test set, rather than the other way around. Crucially, the method based on functional connectivity features shows the best recognition accuracy under different fatigue levels. This research emphasizes the potential benefits of considering fatigue variations in EEG-based personal recognition systems.

This paper introduces an innovative three-phase cascade framework designed for biometric recognition systems, particularly suited for small-scale applications. By integrating multiple biometric modalities - dorsal vein, wrist vein, and palm print - the framework aims to improve recognition accuracy and robustness. The first phase focuses on extracting unique features from each modality using a moment-based approach that is transformation-invariant and computationally efficient. In the second phase, an asymmetric aggregator operator is employed to merge these features into a unified representation. The final phase utilizes spectral clustering to classify and match the fused feature vectors, effectively addressing unseen samples. Evaluated on 350 samples from the COEP and FYO benchmark databases, the framework achieved an impressive accuracy of around 98% for unseen samples, outperforming existing methods like Zernike moment and hierarchical clustering. This work not only enhances biometric authentication but also broadens its applicability across various domains, marking a significant advancement in the field.

Fourier Analysis Network (FAN) was recently proposed as a simple way to improve neural network performance by replacing part of ReLU activations with sine and cosine functions. Although several studies have reported small but consistent gains across tasks, the underlying mechanism behind these improvements has remained unclear. In this work, we show that only the sine activation contributes positively to performance, whereas the cosine activation tends to be detrimental. Our analysis reveals that the improvement is not a consequence of the sine function’s periodic nature; instead, it stems from the function’s local behavior near <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x</i> = 0, where its non-zero derivative mitigates the vanishing-gradient problem. We further show that FAN primarily alleviates the dying-ReLU problem, in which a neuron consistently receives negative inputs, produces zero gradients, and stops learning. Although modern ReLU-like activations, such as Leaky ReLU, GELU, and Swish, reduce ReLU’s zero-gradient region, they still contain input domains where gradients remain significantly diminished, contributing to slower optimization and hindering rapid convergence. FAN addresses this limitation by introducing a more stable gradient pathway. This analysis shifts the understanding of FAN’s benefits from a spectral interpretation to a concrete analysis of training dynamics, leading to the development of the Dual-Activation Layer (DAL), a more efficient convergence accelerator. We evaluate DAL on three tasks: classification of noisy sinusoidal signals versus pure noise, MNIST digit classification, and ECG-based biometric recognition. In all cases, DAL models converge faster and achieve equal or higher validation accuracy compared to models with conventional activations.

Inderscience is a global company, a dynamic leading independent journal publisher disseminates the latest research across the broad fields of science, engineering and technology; management, public and business administration; environment, ecological economics and sustainable development; computing, ICT and internet/web services, and related areas.

Gait recognition has emerged as a robust biometric approach for human identification in surveillance, healthcare, and forensic applications. However, the efficiency of deep-learning-based gait recognition largely depends on the optimization algorithm used for model training and hyperparameter tuning. While traditional gradient-based methods such as Stochastic Gradient Descent (SGD) and Adam are widely adopted, their convergence behavior often deteriorates in high-dimensional non-convex spaces. Recent studies employing metaheuristic algorithms such as Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) have demonstrated performance gains but remain constrained by local optima and unstable convergence dynamics. This study benchmarks a newly introduced metaheuristic; the Hippopotamus Optimization Algorithm (HOA) against four well-established optimizers: Adam, SGD, PSO, and GA, as reported in previous deep learning optimization studies. The developed HOA-CNN-LSTM hybrid model integrates the HOA for global hyperparameter optimization and Adam for fine-tuned gradient updates. Experiments conducted on the TUM-GAID dataset show that HOA achieves 97.4% accuracy and 98.5% Genuine Acceptance Rate (GAR) with a reduced convergence time of 39s per epoch. These results surpass comparative benchmarks reported for Adam [10], SGD [11], PSO [12], and GA [13], confirming HOA’s superior balance between exploration and exploitation. By situating HOA’s performance within a metaheuristic benchmarking framework, this work provides empirical evidence that HOA represents a promising optimization paradigm for next-generation spatiotemporal deep learning and biometric recognition applications.

Reliable and robust personal authentication technologies have become indispensable in modern digital and physical security infrastructures. Traditional unimodal biometric systems—using a single biometric trait—often suffer from noise, spoofing vulnerabilities, and intra-class variability. To overcome these limitations, multimodal biometric systems fuse evidence from multiple biometric sources. However, feature-level fusion, despite yielding richer discriminatory information, produces high-dimensional feature spaces that demand efficient dimensionality reduction or feature selection. This review presents a consolidated analysis of three optimization-driven multimodal biometric recognition systems: Particle Swarm Optimization (PSO) for fingerprint–palmprint fusion, Genetic Algorithm (GA) for iris–fingerprint fusion, and Artificial Bee Colony (ABC) optimization for iris–palmprint fusion. We critically examine preprocessing techniques, feature extraction schemes, fusion strategies, dimensionality-reduction approaches, classifier performance, and comparative advantages. The review highlights trends, challenges, and future research directions in optimization-enhanced multimodal biometrics.