Face Recognition Task Research Articles

Class‐wised domain decoupling‐guided adversarial feature learning for cross‐age face recognition

AbstractHow to extract age‐independent identity features from face images has long been the major challenge in cross‐age face recognition task. In this letter, a class‐wised domain decoupling‐guided adversarial feature learning network to extract age‐independent face features is proposed. In this model, an adversarial feature learning module is designed to decompose identity features and age features. Considering that artificially designed age interference suppression strategies are difficult to fit the complex relationship between identity features and age features, an adversarial training strategy incorporated with an attention mechanism in the adversarial feature learning module is introduced to suppress age interference adaptively. Besides, under the idea of maximizing the distribution difference of identity feature domain and age feature domain, a class‐wised domain decoupling module is further designed to guide the model to extract age‐independent identity features. The proposed model has been validated on the well‐known public datasets CALFW and CACD‐VS, where it achieved remarkable recognition accuracy rates of 94.5% and 99.5%, respectively. This represents improvements of 4.2% and 0.1% over the state‐of‐the‐art single‐task‐based models, clearly showcasing the effectiveness of the proposed class‐wised domain decoupling‐guided adversarial feature learning model.

Comprehensive comparison between vision transformers and convolutional neural networks for face recognition tasks

This paper presents a comprehensive comparison between Vision Transformers and Convolutional Neural Networks for face recognition related tasks, including extensive experiments on the tasks of face identification and verification. Our study focuses on six state-of-the-art models: EfficientNet, Inception, MobileNet, ResNet, VGG, and Vision Transformers. Our evaluation of these models is based on five diverse datasets: Labeled Faces in the Wild, Real World Occluded Faces, Surveillance Cameras Face, UPM-GTI-Face, and VGG Face 2. These datasets present unique challenges regarding people diversity, distance from the camera, and face occlusions such as those produced by masks and glasses. Our contribution to the field includes a deep analysis of the experimental results, including a thorough examination of the training and evaluation process, as well as the software and hardware configurations used. Our results show that Vision Transformers outperform Convolutional Neural Networks in terms of accuracy and robustness against distance and occlusions for face recognition related tasks, while also presenting a smaller memory footprint and an impressive inference speed, rivaling even the fastest Convolutional Neural Networks. In conclusion, our study provides valuable insights into the performance of Vision Transformers for face recognition related tasks and highlights the potential of these models as a more efficient solution than Convolutional Neural Networks.

Scene effects on disgusted facial expression detection in individuals with social anxiety: The role of emotional intensity

Individuals exhibiting high social anxiety (HSA) typically encounter challenges in identifying threatening stimuli with varying levels of intensity in different social scenes, ultimately affecting their social interactions. However, it is not well understood how social scenes, emotional intensity, and interaction influence the recognition of threat stimuli among HSA individuals (HSAs). To address this issue, a face recognition task was administered to 20 HSA participants and 22 individuals exhibiting low social anxiety (LSA) in this study. Results indicated that during the social scene presentation stage, HSAs produced larger P2 amplitude than LSA individuals (LSAs) no matter the valence of the scenes. During the face recognition stage, HSAs had smaller N170 amplitude than LSAs and exhibited lower recognition time for 2 % disgusted faces compared to LSAs. Furthermore, the consistency between scenes and faces led to faster recognition of disgusted faces in HSAs, but not in LSAs. Consequently, our findings suggested that HSAs exhibited unique cognitive processing patterns in social scenes, manifested by increased attention to scenes and decreased attention to faces. In addition, the emotional congruence between the scene and the faces could facilitate the recognition of faces by HSAs.

Face recognition using deep learning on Raspberry Pi

Abstract Facial recognition on resource-limited devices such as the Raspberry Pi poses a challenge due to inherent processing limitations. For real-time applications, finding efficient and reliable solutions is critical. This study investigated the feasibility of using transfer learning for facial recognition tasks on the Raspberry Pi and evaluated transfer learning that leverages knowledge from previously trained models. We compared two well-known deep learning (DL) architectures, InceptionV3 and MobileNetV2, adapted to face recognition datasets. MobileNetV2 outperformed InceptionV3, achieving a training accuracy of 98.20% and an F1 score of 98%, compared to InceptionV3’s training accuracy of 86.80% and an F1 score of 91%. As a result, MobileNetV2 emerges as a more powerful architecture for facial recognition tasks on the Raspberry Pi when integrated with transfer learning. These results point to a promising direction for deploying efficient DL applications on edge devices, reducing latency, and enabling real-time processing.

Enhancing the Transferability of Adversarial Patch via Alternating Minimization

Adversarial patches, a type of adversarial example, pose serious security threats to deep neural networks (DNNs) by inducing erroneous outputs. Existing gradient stabilization methods aim to stabilize the optimization direction of adversarial examples through accumulating gradient momentum to enhance attack transferability on black-box models. However, they are not fully effective for adversarial patches. The accumulated momentum during optimization often misaligns with the optimization direction, reducing their efficacy in black-box scenarios. We introduce an optimization method called Alternating Minimization for Adversarial Patch (AMAP). This method decomposes the original AP into multiple sub-patches and utilizes their update direction to stabilize the optimization of the original AP. Additionally, we propose an adaptive step size optimization method that accelerates convergence and boosts the attack performance. In face recognition tasks, AMAP outperforms baseline methods by a remarkable 5.21% and exceeds the second-best method by 1.4%. Furthermore, AMAP demonstrates practical feasibility in the physical domain, highlighting its potential for robust computer security testing applications.

Multi-model fusion of DSFD and Zero-DCE for low-light face detection

Due to the wide application of face detection, face recognition task under extreme lighting conditions inevitably becomes one of the challenges. Under low-light conditions, the face detection task becomes more complex and difficult due to the presence of more noise and lower visibility in the image. For this extreme environment, this paper proposes a multi-model-assisted solution that combines image enhancement and face detection models to handle this task. The DarkFace dataset was used in this study's validation of a number of distinct face identification models and picture enhancement techniques. The experimental findings demonstrate that, on this dataset, the DSFD (Dual Shot Face Detector) and Zero-DCE (Zero-Reference Deep Curve Estimation) methods perform noticeably better than the other techniques, achieving a recall of 24.61%. Therefore, these two algorithms are selected as the core processing flow in this paper to improve the facial detection system's low-light performance. This combination offers a workable answer to the problem of face detection in low light by effectively increasing the visibility of the picture and raising the accuracy of face detection.

FORMICARY SWARM OPTIMIZED DEEP CNN FOR FACIAL EMOTION RECOGNITION FROM HUMAN FACIAL EXPRESSIONS

Facial emotion recognition (FER) is a dominant research area that captures the biological facial features and matches the facial data with existing databases to analyze the individual’s emotional state. Numerous techniques have been formulated for attaining effective FER. However, the occlusions, different head positions, deformed faces, motion blur under unrestricted settings, and complicated backgrounds make it complex to analyze the facial images. In this paper, the formicary swarm optimization-based deep convolutional neural network (FSO-opt DCNN) model is utilized for facial emotion detection for which JAFFE and the RAVDESS facial expression datasets are used. DCNNs are proficient with built-in feature extraction strategies from images to map the various facial expressions to the corresponding emotional states adopted for effective FER. In addition, the intensity, directional, and edge patterns as well as the correlation features extracted utilizing the hybrid textual pattern, RESNET 101 and VGG 16-based correlation modules assist the DCNN to attain the informative feature from the high-resolution images. Further, the formicary swarm optimization (FSO) is incorporated that effectively tunes the DCNN to capture the complex relationships between the learned features that excel the FER capability. Evaluating the metrics, the face recognition model using the RAVDESS dataset achieves notable efficiencies during a training percentage (TP) of 90%, with values of 97.51%, 95.48%, 99.55%, 95.48%, 97.48%, 96.47%, and a minimum loss of 2.49%. Simultaneously, the emotion detection model demonstrates robust efficiencies with metric values of 96.75%, 98.49%, 95.01%, 98.49%, 96.72%, 97.59%, and a minimum loss of 3.25%. Finally, the obtained results reveal the efficacy of the FSO-opt DCNN, particularly in face recognition tasks, as it outperforms the existing models across various datasets, showcasing its versatility and potential in facial analysis applications.

Perbandingan Dataset Labelled Faces in the Wild (LFW) dan faces94 Menggunakan Algoritma Convolutional Neural Networks (CNN) untuk Pengenalan Wajah

This research compares the performance of two popular datasets, Labelled Faces in the Wild (LFW) and faces94, in the task of face recognition using Convolutional Neural Networks (CNN) algorithms. The LFW dataset is known for its high variation in pose, lighting, and expression, while faces94 is more structured with more uniform lighting and pose conditions. CNNs were chosen for their ability to extract important features from face images for classification. In this study, a CNN model was trained on both datasets and its performance was evaluated using accuracy, precision, and recall metrics. The experimental results showed that the model trained on the faces94 dataset achieved higher accuracy compared to the model trained on the LFW dataset. However, the model on the LFW dataset demonstrated better resilience to variations in lighting and pose conditions. These findings indicate that while a more structured dataset like faces94 can produce a model with high accuracy under testing conditions similar to the training data, a dataset with greater variation like LFW is more suitable for real-world applications involving diverse conditions. This study provides important insights into the selection of datasets for developing robust and accurate face recognition systems.

Kernel Angle Dependence Measures in Metric Spaces

Measuring and testing dependence between data in separable metric spaces is of great importance in modern statistics. Most existing work relied on the distance between random variables, which inevitably required the moment conditions to guarantee the distance is well-defined. Based on the geometry element “angle”, we develop a novel class of nonlinear dependence measures for data in metric space that can avoid such conditions. Specifically, by making use of the reproducing kernel Hilbert space equipped with Gaussian measure, we introduce kernel angle covariances that can be applied to various types of data, including low dimensional vector, high dimensional vectors, non-Euclidean data like symmetric positive definite matrices, and compositional data. We estimate kernel angle covariances based on U-statistic and establish the corresponding independence tests via gamma approximation. Our kernel angle independence tests, imposing no-moment conditions on kernels, are robust with heavy-tailed random variables. We conduct comprehensive simulation studies and apply our proposed methods to a facial recognition task. Our kernel angle covariances-based tests show remarkable performances in dealing with image data. All the codes and proofs are included in the supplementary materials.

Face recognition's practical relevance: Social bonds, not social butterflies

Research on individual differences in face recognition has provided important foundational insights: their broad range, cognitive specificity, strong heritability, and resilience to change. Elusive, however, has been the key issue of practical relevance: do these individual differences correlate with aspects of life that go beyond the recognition of faces, per se? Though often assumed, especially in social realms, such correlates remain largely theoretical, without empirical support. Here, we investigate an array of potential social correlates of face recognition. We establish social relationship quality as a reproducible correlate. This link generalises across face recognition tasks and across independent samples. In contrast, we detect no robust association with the sheer quantity of social connections, whether measured directly via number of social contacts or indirectly via extraversion-related personality indices. These findings document the existence of a key social correlate of face recognition and provide some of the first evidence to support its practical relevance. At the same time, they challenge the naive assumption that face recognition relates equally to all social outcomes. In contrast, they suggest a focused link of face recognition to the quality, not quantity, of one's social connections.

Integration of Manifold Learning and Density Estimation for Fine-Tuned Face Recognition

With the rapid advancements in data analysis and the increasing complexity of high-dimensional datasets, traditional dimensionality reduction techniques like Local Linear Embedding (LLE) often face challenges in maintaining accuracy and efficiency. This research aims to overcome the limitations of LLE, specifically its reliance on the nearest neighbor concept, its inability to distinguish differences among manifold points, and its underutilization of data discrimination information. To address these issues, we propose an advanced LLE algorithm that integrates decision tree-based neighbor recognition with Gaussian kernel density estimation. Decision trees accurately determine neighboring relationships, which are then optimized using Gaussian kernel density estimation to better reflect the distribution of sample points on the manifold. The algorithm also incorporates data discrimination information to enhance classification accuracy and efficiency. Evaluations in facial recognition tasks using SVM classifiers demonstrate significant improvements. Integrating decision trees (LLE-DT) yielded accuracy gains, with LFW at 98.75%, CFP 96.10%, and Olivetti 92.18%. Gaussian density estimation (LLE-GDE) achieved further enhancements, especially in LFW (99.13%), with CFP at 96.85%, and Olivetti at 91.82%. Combining both methods (LLE-DT-GDE) led to substantial improvements: LFW 99.61%, CFP 97.23%, and Olivetti 93.56%, highlighting the synergy between decision trees and Gaussian estimation. This advanced LLE algorithm effectively addresses the limitations of traditional approaches, showing promising results in complex data processing tasks such as facial recognition. These findings suggest its potential for broader applications in fields requiring robust data analysis and classification.

Convolutional neural networks to classify human stress that occurs during in‐field sugarcane harvesting: A case study

AbstractAssessing human stress in agriculture proves to be a complex and time‐intensive endeavor within the field of ergonomics, particularly for the development of agricultural systems. This methodology involves the utilization of instrumentation and the establishment of a dedicated laboratory setup. The complexity arises from the need to capture and analyze various physiological and psychological indicators, such as heart rate (HR), muscle activity, and subjective feedback to comprehensively assess the impact of farm operations on subjects. The instrumentation typically includes wearable devices, sensors, and monitoring equipment to gather real‐time data of subject during the performance of farm operations. Deep learning (DL) models currently achieve human performance levels on real‐world face recognition tasks. In this study, we went beyond face recognition and experimented with the recognition of human stress based on facial features during the drudgery‐prone agricultural operation of sugarcane harvesting. This is the first research study for deploying artificial intelligence‐driven DL techniques to identify human stress in agriculture instead of monitoring several ergonomic characteristics. A total of 20 (10 each for male and female) subjects comprising 4300 augmented RGB images (215 per subject) were acquired during sugarcane harvesting seasons and then these images were deployed for training (80%) and validation (20%). Human stress and nonstress states were determined based on four ergonomic physiological parameters: heart rate (ΔHR), oxygen consumption rate (OCR), energy expenditure rate (EER), and acceptable workload (AWL). Stress was defined when ΔHR, OCR, EER, and AWL reached or exceeded certain standard threshold values. Four convolutional neural network‐based DL models (1) DarkNet53, (2) InceptionV3, (3) MobileNetV2 and (4) ResNet50 were selected due to their remarkable feature extraction abilities, simple and effective implementation to edge computation devices. In all four DL models, training performance results delivered training accuracy ranging from 73.8% to 99.1% at combinations of two mini‐batch sizes and four levels of epochs. The maximum training accuracies were 99.1%, 99.0%, 97.7%, and 95.4% at the combination of mini‐batch size 16 and 25 epochs for DarkNet53, InceptionV3, ResNet50, and MobileNetV2, respectively. Due to the best performance, DarkNet53 was tested further on an independent data set of 100 images and found 89.8%–93.3% confident to classify stressed images for female subjects while 92.2%–94.5% for male subjects, though it was trained on the integrated data set. The comparative classification of the developed model and ergonomic measurements for stress classification was carried out with a net accuracy of 88% where there were few instances of wrong classifications.

Residual serialized cross grouping transformer for small scale sketch face recognition

Sketch face recognition has recently gained significant attention in the field of computer vision due to its ability to quickly identify matched pairs of optical and sketch images. This technology has the potential to greatly improve the efficiency of law enforcement agencies in criminal investigations. However, there are still challenges that need to be addressed in sketch face recognition algorithms, such as modal differences and limited sample sizes. To overcome these issues, this study proposes a Residual Serialized Cross Grouping Transformer (RSCGT), which contains a residual serialized module to reduce the computation complexity, a two-layer Cross Grouping Transformer module that is capable of extracting modality-invariant context features, a domain adaptive module to mitigate the impact of modal differences. Additionally, we introduce a meta-learning training strategy to augment the generalization ability of this model. Experimental results demonstrate that the RSCGT achieves high accuracy in sketch face recognition tasks, even with small-scale datasets.

“Sparse + Low-Rank” tensor completion approach for recovering images and videos

Recovering color images and videos from highly undersampled data is a fundamental and challenging task in face recognition and computer vision. By the multi-dimensional nature of color images and videos, in this paper, we propose a novel tensor completion approach, which is able to efficiently explore the sparsity of tensor data under the discrete cosine transform (DCT). Specifically, we introduce two “sparse + low-rank” tensor completion models as well as two implementable algorithms for finding their solutions. The first one is a DCT-based sparse plus weighted nuclear norm induced low-rank minimization model. The second one is a DCT-based sparse plus p-shrinking mapping induced low-rank optimization model. Moreover, we accordingly propose two implementable augmented Lagrangian-based algorithms for solving the underlying optimization models. A series of numerical experiments including color image inpainting and video data recovery demonstrate that our proposed approach performs better than many existing state-of-the-art tensor completion methods, especially for the case when the ratio of missing data is high.

Robust sparse representation based on fitting error decomposition

Sparse representation (SR) of a signal aims at finding the minimum number of atoms for its representation. In several practical scenarios, the signal is vulnerable to outliers and thus robustness is required for SR based algorithms. However, most existing robust schemes are designed on the assumption that the anomalies are independently distributed, which may not perform well encountering complex interferences, especially the correlated ones. To deal with this drawback, a robust SR model is proposed, where both independent and correlated outliers are considered. Specifically, the fitting error is decomposed as the combination of a low-rank component and a sparse part, corresponding to the correlated gross error and independent outlier, respectively. Then, the group sparsity of the representation coefficient is utilized. Moreover, ℓ0-norm and ℓ2,0-norm are adopted as the sparsity regularization for the sparse outlier and representation coefficients, respectively. The solutions to ℓ0/ℓ2,0-norm minimization are generated by the hard-thresholding strategy, where the decision threshold is adaptively determined using median absolute deviation operator. As for the low-rank regularization, due to the NP-hardness of rank minimization, we employ matrix logarithmic norm as the rank surrogate to lessen the approximation gap. Finally, we apply the proposed model to face recognition task, and the excellent performance demonstrates its effectiveness.

Social identity threat attenuates own-race bias in face recognition under the “Asian-Caucasian” context

People typically remember faces of their own race better than the faces of other races, which is known as the own-race bias. Social-cognitive theories on the own-race bias debate if motivation plays a role in racial face recognition under the “Asian-Caucasian” context. We investigated the effect of social-identity-threat based motivation on own-race bias. In Experiment 1, Chinese participants were primed with a negative image of own-race by associating Chinese (i.e., own-race) faces with negative personality traits. Results showed that such social identity threat marginally affect the own-race bias in face recognition sensitivity (Experiment 1a), and it did not influence participants' implicit positive attitudes toward their ethnicity (Experiment 1b). In Experiments 2, 3 and 4, the social identity threat was manipulated by reading an article before the face recognition task. Experiment 2 found that the social identity threat significantly affected the own-race bias in face recognition false alarm. Although we did not find the modulation effect of group on ORB with ANOVA in Experiments 3 and 4, we observed the significant indirect effect of group on ORB through the mediator of generalized threat. Together, the current study provides evidence of social motivation in the own-race bias in the “Asian-Caucasian” setting.

Closing a conceptual gap in race perception research: A functional integration of the other-race face recognition and "who said what?" paradigms.

White people confuse Black faces more than their own-race faces. This is an example of the other-race effect, commonly measured by the other-race face recognition task. Like this task, the "Who said what?" paradigm uses within-race confusions in memory, but to measure social categorization strength. The former finds a strongly asymmetrical pattern of interrace perception, the other-race effect, yet the latter usually finds symmetrical patterns (equally strong categorization of own-race and other-race faces). In a "Who said what?" meta-analysis, racial categorization and individuation across races were only weakly asymmetrical (Study 1, n = 2,669). We aimed to resolve this empirical misalignment. As tested in other-race face recognition tasks, the weak asymmetry was not due to the limited number of portrait stimuli (Study 2, N = 99) nor to the longer duration of stimulus presentation in the "Who said what?" task (Study 4, n = 358). Pairing portraits with statements reduced the other-race effect (Study 3, n = 126). Showing each portrait repeatedly also reduced the other-race effect (Study 4, n = 358; Study 5, n = 470) but did not decrease infrahumanization of Black portraits (Study 6, n = 487). Consequently, presenting portraits only once in the "Who said what?" paradigm (Study 7, N = 112) resulted in strong interrace categorization and individuation asymmetries. This finding bridges a central conceptual gap between the other-race effect and social categorization strength. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Device-edge collaborative occluded face recognition method based on cross-domain feature fusion

The lack of facial features caused by wearing masks degrades the performance of facial recognition systems. Traditional occluded face recognition methods cannot integrate the computational resources of the edge layer and the device layer. Besides, previous research fails to consider the facial characteristics including occluded and unoccluded parts. To solve the above problems, we put forward a device-edge collaborative occluded face recognition method based on cross-domain feature fusion. Specifically, the device-edge collaborative face recognition architecture gets the utmost out of maximizes device and edge resources for real-time occluded face recognition. Then, a cross-domain facial feature fusion method is presented which combines both the explicit domain and the implicit domain facial. Furthermore, a delay-optimized edge recognition task scheduling method is developed that comprehensively considers the task load, computational power, bandwidth, and delay tolerance constraints of the edge. This method can dynamically schedule face recognition tasks and minimize recognition delay while ensuring recognition accuracy. The experimental results show that the proposed method achieves an average gain of about 21% in recognition latency, while the accuracy of the face recognition task is basically the same compared to the baseline method.

Brain morphological changes and functional neuroanatomy related to cognitive and emotional distractors during working memory maintenance in post-traumatic stress disorder

Post-traumatic stress disorder (PTSD) is associated with abnormalities in the processing and regulation of emotion as well as cognitive deficits. This study evaluated the differential brain activation patterns associated with cognitive and emotional distractors during working memory (WM) maintenance for human faces between patients with PTSD and healthy controls (HCs) and assessed the relationship between changes in the activation patterns by the opposing effects of distraction types and gray matter volume (GMV). Twenty-two patients with PTSD and twenty-two HCs underwent T1-weighted magnetic resonance imaging (MRI) and event-related functional MRI (fMRI), respectively. Event-related fMRI data were recorded while subjects performed a delayed-response WM task with human face and trauma-related distractors. Compared to the HCs, the patients with PTSD showed significantly reduced GMV of the inferior frontal gyrus (IFG) (p < 0.05, FWE-corrected). For the human face distractor trial, the patients showed significantly decreased activities in the superior frontal gyrus and IFG compared with HCs (p < 0.05, FWE-corrected). The patients showed lower accuracy scores and slower reaction times for the face recognition task with trauma-related distractors compared with HCs as well as significantly increased brain activity in the STG during the trauma-related distractor trial was observed (p < 0.05, FWE-corrected). Such differential brain activation patterns associated with the effects of distraction in PTSD patients may be linked to neural mechanisms associated with impairments in both cognitive control for confusable distractors and the ability to control emotional distraction.

Understanding of facial features in face perception: insights from deep convolutional neural networks.

Face recognition has been a longstanding subject of interest in the fields of cognitive neuroscience and computer vision research. One key focus has been to understand the relative importance of different facial features in identifying individuals. Previous studies in humans have demonstrated the crucial role of eyebrows in face recognition, potentially even surpassing the importance of the eyes. However, eyebrows are not only vital for face recognition but also play a significant role in recognizing facial expressions and intentions, which might occur simultaneously and influence the face recognition process. To address these challenges, our current study aimed to leverage the power of deep convolutional neural networks (DCNNs), an artificial face recognition system, which can be specifically tailored for face recognition tasks. In this study, we investigated the relative importance of various facial features in face recognition by selectively blocking feature information from the input to the DCNN. Additionally, we conducted experiments in which we systematically blurred the information related to eyebrows to varying degrees. Our findings aligned with previous human research, revealing that eyebrows are the most critical feature for face recognition, followed by eyes, mouth, and nose, in that order. The results demonstrated that the presence of eyebrows was more crucial than their specific high-frequency details, such as edges and textures, compared to other facial features, where the details also played a significant role. Furthermore, our results revealed that, unlike other facial features, the activation map indicated that the significance of eyebrows areas could not be readily adjusted to compensate for the absence of eyebrow information. This finding explains why masking eyebrows led to more significant deficits in face recognition performance. Additionally, we observed a synergistic relationship among facial features, providing evidence for holistic processing of faces within the DCNN. Overall, our study sheds light on the underlying mechanisms of face recognition and underscores the potential of using DCNNs as valuable tools for further exploration in this field.