Face Recognition Research Articles

Faculty presence detection and alert system

The exponential growth of inspection systems has increased the demand for efficient and cost-effective solutions. This project introduces a faculty verification system that uses the webcam of the system and OpenCV for real-time facial recognition. The system integrates a Flask-based web interface to provide an intuitive and dynamic user experience. The key features are live detection and system-level video capture by using the Haar Cascade classifier. It gives more importance to accessibility and user-friendliness. Unlike some solutions that work on an external camera or even a Raspberry Pi module, this system works on solely built-in resources. Therefore, the value is guaranteed. The study evaluates performance in face recognition under varying conditions. Focus has been laid on accuracy, responsiveness, and scalability.

Countermeasures against bias and spoofing in modern facial recognition systems

Countermeasures against bias and spoofing in modern facial recognition systems

A novel integral and SE(3)-invariant description for 3D face recognition

A novel integral and SE(3)-invariant description for 3D face recognition

Human-like face pareidolia emerges in deep neural networks optimized for face and object recognition.

The human visual system possesses a remarkable ability to detect and process faces across diverse contexts, including the phenomenon of face pareidolia--seeing faces in inanimate objects. Despite extensive research, it remains unclear why the visual system employs such broadly tuned face detection capabilities. We hypothesized that face pareidolia results from the visual system's optimization for recognizing both faces and objects. To test this hypothesis, we used task-optimized deep convolutional neural networks (CNNs) and evaluated their alignment with human behavioral signatures and neural responses, measured via magnetoencephalography (MEG), related to pareidolia processing. Specifically, we trained CNNs on tasks involving combinations of face identification, face detection, object categorization, and object detection. Using representational similarity analysis, we found that CNNs that included object categorization in their training tasks represented pareidolia faces, real faces, and matched objects more similarly to neural responses than those that did not. Although these CNNs showed similar overall alignment with neural data, a closer examination of their internal representations revealed that specific training tasks had distinct effects on how pareidolia faces were represented across layers. Finally, interpretability methods revealed that only a CNN trained for both face identification and object categorization relied on face-like features-such as 'eyes'-to classify pareidolia stimuli as faces, mirroring findings in human perception. Our results suggest that human-like face pareidolia may emerge from the visual system's optimization for face identification within the context of generalized object categorization.

Factors influencing behavioral intention to use facial recognition payment in the restaurant industry: a comparative analysis of China and Malaysia

ABSTRACT This study examines factors influencing behavioral intention (BI) to use facial recognition payment (FRP) in restaurants across China (403 responses) and Malaysia (395 responses). Using PLS-SEM, the study explores the valence framework and trust transfer theory. Relative advantage, personal innovativeness, and perceived herd positively influence BI, while perceived sensitivity of facial information, perceived risks, and technophobia negatively affect trust in FRP. Trust in FRP systems and service providers significantly impact BI. Multi-group analysis highlights cultural differences: personal innovativeness is more influential in China, while technophobia is more detrimental in Malaysia. Findings underscore the need for culturally tailored trust-building strategies.

Transfer of Share Ownership and Change of Position of Director of a Limited Liability Company Due to Misuse of GLA Online Accounts

The misuse of General Legal Administration (GLA) Online accounts, a system developed to streamline public services for notaries, has caused significant issues, including unauthorized transfers of share ownership and changes in director positions within Limited Liability Companies. This research aims to analyze the legal mechanisms surrounding these corporate actions, investigate the challenges posed by account misuse, and propose effective legal remedies for affected parties. Employing a doctrinal research method, the research relies on literature analysis, interviews with experts, and a case research of the Cibinong District Court Decision Number 9/Pdt.G/2024/PN.Cbi. Primary legal materials include laws and court decisions, while secondary materials encompass books, journals, and credible online sources. The findings reveal that both the transfer of shares and the replacement of directors, as a result of account misuse, violate the Company Law and corporate governance principles. To mitigate these issues, the Directorate General of GLA suggests procedural remedies, including police reports, Regional Supervisory Council investigations, and State Administrative Court filings. This research highlights the need for enhanced digital security measures, such as face recognition, and emphasizes stricter notary accountability. Future research can explore the broader implications of digital platform governance in corporate law, aiming to improve legal frameworks and prevent similar misuse.

Can artificial intelligence understand our emotions? Deep learning applications with face recognition

Abstract Objective The aim of this study is to evaluate the ability to detect emotions from human facial expressions via facial recognition technologies and analyze the effectiveness of deep learning models in this process. Method This research was conducted between 01.04 and 01.07.2024. The data of the study were taken from the open access site https://www.kaggle.com/datasets/msambare/fer2013 (Kaggle, 2024). Python 3.8 is used in this study. The FER-2013 (Facial Expression Recognition 2013) dataset is a comprehensive collection of facial images labeled with various emotions. The dataset contains 35,887 grayscale facial images. Each image has a size of 48 × 48 pixels. The dataset consists of images belonging to 7 emotion categories: anger, disgust, fear, happy, sad, confused, and neutral. Results In our experiments on the FER2013 dataset, we evaluated the performance of three different models: MobileNetV3-L, EfficientNetV2-L, and our proposed EfficientMobileNet. The evaluation criteria were based on sensitivity, specificity, accuracy, and F1 scores to assess the effectiveness of each model comprehensively. The EfficientMobileNet model outperformed MobileNetV3-L and EfficientNetV2-L in all measured performance metrics. EfficientMobileNet was the most successful model for predicting emotions, with an accuracy of 77.6%. Conclusion The impressive results obtained by EfficientMobileNet on the Fer2013 dataset show potential for wider application, especially in image classification tasks involving low-quality or small-scale images. This performance supports the idea of the potential for further improvements in neural network architecture and model efficiency and accuracy. Future work should focus on optimizing the model for more challenging datasets, studying the impact of different architectural adjustments, and investigating the scalability of EfficientMobileNet across various domains and applications.

Learnable Anchor Embedding for Asymmetric Face Recognition

Face verification and identification traditionally follow a symmetric matching approach, where the same model (e.g., ResNet-50 vs. ResNet-50) generates embeddings for both gallery and query images, ensuring compatibility. However, real-world scenarios often demand asymmetric matching, especially when query devices have limited computational resources or employ heterogeneous models (e.g., ResNet-50 vs. SwinTransformer). This asymmetry can degrade face recognition performance due to incompatibility between embeddings from different models. To tackle this asymmetric face recognition problem, we introduce the Learnable Anchor Embedding (LAE) model, which features two key innovations: the Shared Learnable Anchor and a Light Cross-Attention Mechanism. The Shared Learnable Anchor is a dynamic attractor, aligning heterogeneous gallery and query embeddings within a unified embedding space. The Light Cross-Attention Mechanism complements this alignment process by reweighting embeddings relative to the anchor, efficiently refining their alignment within the unified space. Extensive evaluations of several facial benchmark datasets demonstrate LAE’s superior performance, particularly in asymmetric settings. Its robustness and scalability make it an effective solution for real-world applications such as edge-device authentication, cross-platform verification, and environments with resource constraints.

Generation of artificial facial drug abuse images using deep de-identified anonymous dataset augmentation through genetics algorithm (3DG-GA)

Abstract In biomedical research and artificial intelligence, access to large, well-balanced, and representative datasets is crucial for developing trustworthy applications that can be used in real-world scenarios. However, obtaining such datasets can be challenging, as they are often restricted to hospitals and specialized facilities. To address this issue, the study proposes to generate highly realistic synthetic faces exhibiting drug abuse traits through augmentation. The proposed method, called ”3DG-GA”, Deep De-identified anonymous Dataset Generation, uses a Genetic Algorithm as a strategy for synthetic face generation. The algorithm includes GAN-based artificial face generation, forgery detection, and face recognition. Initially, a dataset of 120 images of actual facial drug abuse is used. By preserving the drug traits, the 3DG-GA provides a dataset containing 3000 synthetic facial drug abuse images. The dataset will be open to the scientific community, allowing others to reproduce our results and benefit from the generated datasets while avoiding legal or ethical restrictions. Additionally, we validated the dataset by training a CNN model on the synthetic images and validating it on previously unseen real images. The model achieved an accuracy of 97.2% on the unseen real images, demonstrating the high quality and applicability of the synthetic data.

Real-time detection and monitoring of public littering behavior using deep learning for a sustainable environment

With the global population surpassing 8 billion, waste production has skyrocketed, leading to increased pollution that adversely affects both terrestrial and marine ecosystems. Public littering, a significant contributor to this pollution, poses severe threats to marine life due to plastic debris, which can inflict substantial ecological harm. Additionally, this pollution jeopardizes human health through contaminated food and water sources. Given the annual global plastic consumption of approximately 475 million tons and the pervasive issue of public littering, addressing this challenge has become critically urgent. The Surveillance and Waste Notification (SAWN) system presents an innovative solution to combat public littering. Leveraging surveillance cameras and advanced computer vision technology, SAWN aims to identify and reduce instances of littering. Our study explores the use of the MoViNet video classification model to detect littering activities by vehicles and pedestrians, alongside the YOLOv8 object detection model to identify individuals responsible through facial recognition and license plate detection. Collecting appropriate data for littering detection presented significant challenges due to its unavailability. Consequently, project members simulated real-life littering scenarios to gather the required data. This dataset was then used to train different models, including LRCN, CNN-RNN, and MoViNets. After extensive testing, MoViNets demonstrated the most promising results. Through a series of experiments, we progressively improved the model’s performance, achieving accuracy rates of 93.42% in the first experiment, 95.53% in the second, and ultimately reaching 99.5% in the third experiment. To detect violators’ identities, we employed YOLOv8, trained on the KSA vehicle plate dataset, achieving 99.5% accuracy. For face detection, we utilized the Haar Cascade from the OpenCV library, known for its real-time performance. Our findings will be used to further enhance littering behavior detection in future developments.

Quantitative Study of Swin Transformer and Loss Function Combinations for Face Anti-Spoofing

Face anti-spoofing (FAS) has always been a hidden danger in network security, especially with the widespread application of facial recognition systems. However, some current FAS methods are not effective at detecting different forgery types and are prone to overfitting, which means they cannot effectively process unseen spoof types. Different loss functions significantly impact the classification effect based on the same feature extraction without considering the quality of the feature extraction. Therefore, it is necessary to find a loss function or a combination of different loss functions for spoofing detection tasks. This paper mainly aims to compare the effects of different loss functions or loss function combinations. We selected the Swin Transformer as the backbone of our training model to extract facial features to ensure the accuracy of the ablation experiment. For the application of loss functions, we adopted four classical loss functions: cross-entropy loss (CE loss), semi-hard triplet loss, L1 loss and focal loss. Finally, this work proposed combinations of Swin Transformers and different loss functions (pairs) to test through in-dataset experiments with some common FAS datasets (CelebA-Spoofing, CASIA-MFSD, Replay attack and OULU-NPU). We conclude that using a single loss function cannot produce the best results for the FAS task, and the best accuracy is obtained when applying triplet loss, cross-entropy loss and Smooth L1 loss as a loss combination.

BabyFM: Towards accurate 3D baby facial models using spectral decomposition and asymmetry swapping.

In this paper, we present the first publicly available 3D statistical facial shape model of babies, the Baby Face Model (BabyFM). Constructing a model of the facial geometry of babies entails specific challenges, such as occlusions, extreme and uncontrollable expressions, and data shortage. We address these challenges by proposing (1) a non-template dependent method that jointly estimates a 3D facial baby-specific template and the point-to-point correspondences; (2) a novel method to establish correspondences based on the spectral decomposition of the Laplace Beltrami Operator, which provides a more robust theoretical foundation than state-of-the-art methods; and (3) an asymmetry-swapping strategy to alleviate the shortage of large scale datasets by decoupling the identity-related and the asymmetry-related shape deformation fields. The latter leads to a data augmentation technique that we integrate within the Gaussian Process Morphable Model framework, providing a simple way of combining synthetic or sample covariance functions. We exhaustively evaluate each stage of our method and demonstrate that (1) when aiming at the 3D facial geometry of a baby, a specific model of babies is needed, since the pre-built publicly available models constructed with adults or older children are not able to accurately represent the facial shape of babies; (2) our spectral approach improves correspondences accuracy with respect to state-of-the-art-methods; and (3) the proposed data augmentation technique enhances the robustness of the BabyFM.

Exploring the Challenges of Aadhaar based Face Recognition in Unrestricted Environments

- Technology improvements have resulted in a high criminal rate, which has raised serious concerns about security-related issues. Face recognition can be used to identify people because every person tends to possess a particular attribute. One of the main applications of facial recognition is in the field of video surveillance. To reduce the increasing criminal rates, this technology is responsible for extracting features from the human face and further identifying them. The CCTV footage might also be used to identify suspects at a crime scene. However, the criminals would be found by recognizing and comparing the suspect's face to the database repository that would include data on people based on their Aadhaar card. Aadhaar photos can be used to identify and recognize faces in a facial recognition system since Aadhaar is a unique identification system in India and mandatory for every Indian resident. However, there are many challenges faced by Aadhaar based face recognition. Face recognition in uncontrolled environments like public surveillance faces significant challenges such as low-resolution images, pose variations, and masked faces. This paper highlights the challenges faced by Aadhaar enabled face recognition in surveillance scenes. This article aims to provide a brief survey of the Aadhaar-enabled face recognition system for criminal identification and recognizing individuals. There is limited work with the Aadhaar dataset due to privacy and security reasons. This article also emphasizes the difficulties that will be encountered in implementing facial recognition as well as strategies for enhancing it while taking various tradeoffs into account. It also recommends opportunities for further study in the application of facial identification in other fields. Key Words: Aadhaar, Surveillance, Face recognition, Criminals

Enhancing social functioning using multi-user, immersive virtual reality

Rates of loneliness and other forms of social disconnection have been increasing worldwide. Prior studies have suggested that brief behavioral interventions can teach skills that may improve social functioning and connection but, currently, access to such interventions is limited. One previously untested approach for addressing this gap is to teach these skills using immersive, multi-user virtual reality (VR). To measure the feasibility of this approach, 33 young adults experiencing social discomfort were enrolled in a study of a VR-based application called Reconnecting with Ourselves and Others in virtual Meetings (ROOM), which delivered a previously validated, mindfulness-based intervention. Satisfaction ratings, qualitative feedback, and quantitative measures of aspects of social functioning, including social comfort, interpersonal distance, and facial affect recognition, were collected from participants. All of the participants attended more than half of the sessions and 90.3% found ROOM to be beneficial and useful. In addition, comfort with others in the real world (p = 0.02) and facial affect recognition accuracy (p = 0.02) significantly increased, while interpersonal distances remained unchanged, following ROOM. An immersive, multi-user VR intervention may represent one feasible, scalable approach for disseminating skills that can improve social functioning and connection. Trial registration Clinicaltrial.org # NCT06360562.

Deficits in facial emotion recognition and cognitive function among baby boomers.

Facial emotion recognition (FER), a key component of social cognition, plays a critical role in social interactions. In the aging process, FER among older adults holds significant potential as a tool for diagnosing cognitive function or enhancing interpersonal relationships. However, research in this area remains limited. This study aims to address this gap by examining the impact of cognitive function on FER among Korean baby boomers aged 60 to 69. Eighty-one participants completed the Korean version of the Montreal Cognitive Assessment (K-MOCA) and FER tasks. Of the participants, 69% had normal cognition, while 31% had mild cognitive impairment. Participants with normal cognition were 1.59 times more likely to recognize facial expressions correctly than those with impaired cognition (AOR=1.59, p<0.0001). They showed significantly higher odds of recognizing happy (AOR=9.68), anger (AOR=2.25), disgust (AOR=1.95), neutral (AOR=3.02), and surprise (AOR=2.27) expressions (p<0.001). Participants with normal cognition participants also scored higher on overall intensity for every emotion except sadness (p<0.001). Among the seven domains of the MoCA, three sub-domains-visuospatial/executive function, language, and orientation-showed significant associations with overall FER. These results highlight the diagnostic potential of FER tasks in identifying cognitive disorders and enhancing social skills in clinical and practical settings.

Application of Facial Emotion Recognition in Teaching-Learning Process for Quality Assessment and Enhancement

Application of Facial Emotion Recognition in Teaching-Learning Process for Quality Assessment and Enhancement

Interracial contact shapes racial bias in the learning of person-knowledge

During impression formation, perceptual cues facilitate social categorization while person-knowledge can promote individuation and enhance person memory. Although there is extensive literature on the cross-race recognition deficit, observed when racial ingroup faces are recognized more than outgroup faces, it is unclear whether a similar deficit exists when recalling individuating information about outgroup members. To better understand how perceived race can bias person memory, the present study examined how self-identified White perceivers’ interracial contact impacts learning of perceptual cues and person-knowledge about perceived Black and White others over five sessions of training. While person-knowledge facilitated face recognition accuracy for low-contact perceivers, face recognition accuracy did not differ for high-contact perceivers based on person-knowledge availability. The results indicate a bias towards better recall of ingroup person knowledge, which decreased for high-contact perceivers across the five-day training but simultaneously increased for low-contact perceivers. Overall, the elimination of racial bias in recall of person-knowledge among high-contact perceivers amid a persistent cross-race deficit in face recognition suggests that contact may have a greater impact on the recall of person-knowledge than on face recognition.

EGEFACE: A new face memory test with static and dynamic images

Face memory is a crucial cognitive ability necessary for maintaining a healthy social life. Recent studies reveal large individual differences in face recognition ability. Face memory tests are used to evaluate this ability. The main purpose of this study was to develop a new face memory test (EGEFACE) addressing the limitations of existing tests using both static and dynamic stimuli to increase ecological validity; employing face recognition algorithms to adjust test difficulty; measuring face memory accuracy independently of response bias by including both target-absent and target-present trials and using ROC analysis; and developing a test to measure both ends of the face recognition ability spectrum. After building a new database of static and dynamic faces, we created three difficulty levels using a face recognition algorithm. We collected data from 703 participants in two steps and examined the internal consistency, split-half reliability, and item–total score correlations. The reliability analysis confirmed that both target-absent and target-present trials of EGEFACE were reliable. High EGEFACE performers scored near super recognizer levels on CFMT+, while low performers showed limited overlap with prosopagnosic-level performance on CFMT+, suggesting EGEFACE’s sensitivity across different levels of face recognition ability. Overall, results indicated a moderate positive correlation between EGEFACE and CFMT+, showing that both tests assess similar cognitive skills, while a low to moderate correlation with KFMT suggests that EGEFACE measures cognitive ability that is related to yet distinct from face perception. The results suggest that EGEFACE shows promise as an ecologically valid and effective alternative tool for assessing individual differences in face memory.

Exploring predementia: Understanding the characteristics of subjective cognitive decline plus from India.

Subjective cognitive decline (SCD) is the early predementia syndrome. that occurs even before the development of objective cognitive decline. SCD plus refers to an additional set of criteria that increases the likelihood of developing mild cognitive impairment and further progressing to Alzheimer's disease (AD). Studying the progression of SCD-plus participants will help in understanding the importance of diagnosing this condition at an early stage and delaying its onset. The present tries to examine neurocognitive changes in individuals who met the criteria of SCD-plus patients. The study also investigated the imaging correlates of these individuals in both cohorts. This study included 94 participants from Srinivaspura Aging, Neuro Senescence, and COGnition (SANSCOG) and Tata Longitudinal Study of Aging (TLSA) cohorts who satisfy the criteria of SCD plus. Mann-Whitney U test was used to compare the SCD plus participants and healthy controls. Regression analysis was performed to find the association between SCD plus and cognition. The SCD-plus group performed poorer than the healthy group in episodic memory delayed recall (p = 0.049), name face recognition (p = 0.023), and letter fluency (p = 0.004) tasks. The generalized linear model revealed that the SCD-plus group had lower left cerebellar cortex (p = 0.010) and right inferior occipital cortex (p = 0.016) volumes than the healthy control group. The participants in the SCD-plus group performed poorly on memory and language-related tasks, and the volumes of the associated brain regions decreased. This study suggested that the SCD-plus group had characteristics similar to AD group and can help in identifying AD at the earliest.

Assessing bias and computational efficiency in vision transformers using early exits

Face recognition with deep learning is generally approached as a problem of capacity. The field has seen progressively deeper, more complex models or larger, more highly variant data sets. The data sets can be problematic, as they are often scraped indiscriminately from the internet. This results in an uncertain, and often heavily unbalanced distribution of race, gender, age and other aspects of the subjects, which is then manifested in the decisions of the models trained on them. The carbon footprint of machine learning is a concern. A real push is developing to reduce the energy consumption of machine learning as we strive for a more eco-friendly society. In addition, due to many instances of misuse by law enforcement and other agencies, unbiased models for face recognition are now fundamental to the practical application of the field. We present an approach using the state of the art Vision Transformer and Early Exits for reducing compute budget without significantly affecting performance. We develop a system for face recognition and identification with a closed-set gallery and show that with a small reduction in performance, a reasonable reduction in compute cost can be obtained using our method. Second, we investigate how these early Exits interact with the bias model through a robust evaluation of matching scores on a racially balanced data set. We show that matching scores vary heavily between cohorts, and these variations are magnified at the earlier exits.