Advances in Facial Expression Recognition: A Survey of Methods, Benchmarks, Models, and Datasets

Information, Journal Year: 2024, Volume and Issue: 15(3), P. 135 - 135

Published: Feb. 28, 2024

Recent technological developments have enabled computers to identify and categorize facial expressions determine a person’s emotional state in an image or video. This process, called “Facial Expression Recognition (FER)”, has become one of the most popular research areas computer vision. In recent times, deep FER systems primarily concentrated on addressing two significant challenges: problem overfitting due limited training data availability, presence expression-unrelated variations, including illumination, head pose, resolution, identity bias. this paper, comprehensive survey is provided FER, encompassing algorithms datasets that offer insights into these intrinsic problems. Initially, paper presents detailed timeline showcasing evolution methods expression recognition (FER). illustrates progression development techniques resources used FER. Then, review introduced, basic principles (components such as preprocessing, feature extraction classification, methods, etc.) from pro-deep learning era (traditional using handcrafted features, i.e., SVM HOG, era. Moreover, brief introduction related benchmark (there are categories: controlled environments (lab) uncontrolled (in wild)) evaluate different comparison models. Existing neural networks strategies designed for based static images dynamic sequences, discussed. The remaining challenges corresponding opportunities future directions designing robust also pinpointed.

Language: Английский

---

Hybrid approach for suspicious object surveillance using video clips and UAV images in cloud-IoT-based computing environment

Cluster Computing, Journal Year: 2023, Volume and Issue: 27(1), P. 761 - 785

Published: Feb. 17, 2023

Language: Английский

---

Advancing Facial Expression Recognition in Online Learning Education Using a Homogeneous Ensemble Convolutional Neural Network Approach

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(3), P. 1156 - 1156

Published: Jan. 30, 2024

Facial expression recognition (FER) plays a crucial role in understanding human emotions and is becoming increasingly relevant educational contexts, where personalized empathetic interactions are essential. The problems with existing approaches typically solved using single deep learning method, which not robust complex datasets, such as FER data, have characteristic imbalance multi-class labels. In this research paper, an innovative approach to homogeneous ensemble convolutional neural network, called HoE-CNN, presented for future online education. This paper aims transfer the knowledge of models classification ensembled conventional network architectures. challenging because there many real-world applications consider, adaptive user interfaces, games, education, robot integration. HoE-CNN used improve performance on dataset, encompassing seven main multi-classes (Angry, Disgust, Fear, Happy, Sad, Surprise, Neutral). experiment shows that proposed framework, uses models, performs better than model. summary, model will increase efficiency results solve FER2013 at accuracy 75.51%, addressing both imbalanced datasets application applications.

Language: Английский

---

The Use of Facial Recognition Technology by Law Enforcement in Europe: a Non-Orwellian Draft Proposal

European Journal on Criminal Policy and Research, Journal Year: 2022, Volume and Issue: 29(4), P. 515 - 533

Published: June 1, 2022

Language: Английский

---

Quantum face recognition protocol with ghost imaging

Scientific Reports, Journal Year: 2023, Volume and Issue: 13(1)

Published: Feb. 10, 2023

Face recognition is one of the most ubiquitous examples pattern in machine learning, with numerous applications security, access control, and law enforcement, among many others. Pattern classical algorithms requires significant computational resources, especially when dealing high-resolution images an extensive database. Quantum have been shown to improve efficiency speed tasks, as such, they could also potentially complexity face process. Here, we propose a quantum learning algorithm for based on principal component analysis, independent analysis. A novel finding dissimilarity faces computation trace determinant matrix (image) proposed. The overall our [Formula: see text]-N image dimension. As input these algorithms, consider experimental obtained from imaging techniques correlated photons, e.g. "interaction-free" or "ghost" imaging. Interfacing processor provides that possess better signal-to-noise ratio, lower exposures, higher resolution, thus speeding up process further. Our fully system inputs promises much-improved acquisition identification potential extending beyond recognition, e.g., medical diagnosing sensitive tissues biology protein identification.

Language: Английский

---

A Comprehensive Review of Face Recognition Techniques, Trends, and Challenges

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 107903 - 107926

Published: Jan. 1, 2024

Face Recognition (FR) is the technology used to identify and verify individuals based on their facial features. In recent decades, FR plays a crucial role in various sectors including security, healthcare, banking, criminal identification. For effective FR, numerous techniques are currently under development which range from appearance hybrid approaches. Most of existing methods offer diverse solutions describe face image either by focusing specific features or considering entire face. This study explores such challenges related FR. The were analysed with respect perspectives inputs, viz., illumination, pose variation, expressions, occlusions, aging led prominent implementation systems. primary contribution this survey lies comprehensive review state-of-the-art deriving taxonomy categorizing these into classes Moreover, proposed detailed highlights significant most research developed also, provide classification video-based methods, highlighting major advancements core processing steps for handling huge volume datasets. outlines current trends available datasets emphasizing enhancements. also aims valuable resource researchers practitioners offering insights latest developments identifying open problems that require further investigation.

Language: Английский

---

Study on an Improved YOLOv7-Based Algorithm for Human Head Detection

Electronics, Journal Year: 2025, Volume and Issue: 14(9), P. 1889 - 1889

Published: May 7, 2025

In response to the decreased accuracy in person detection caused by densely populated areas and mutual occlusions public spaces, a human head-detection approach is employed assist detecting individuals. To address key issues dense scenes—such as poor feature extraction, rough label assignment, inefficient pooling—we improved YOLOv7 network three aspects: adding attention mechanisms, enhancing receptive field, applying multi-scale fusion. First, large amount of surveillance video data from crowded spaces was collected compile dataset. Then, based on YOLOv7, optimized follows: (1) CBAM module added neck section; (2) Gaussian field-based label-assignment strategy implemented at junction between original feature-fusion head; (3) SPPFCSPC used replace multi-space pyramid pooling. By seamlessly uniting CBAM, RFLAGauss, SPPFCSPC, we establish novel collaborative optimization framework. Finally, experimental comparisons revealed that model’s increased 92.4% 94.4%; recall 90.5% 93.9%; inference speed 87.2 frames per second 94.2 second. Compared with single-stage object-detection models such YOLOv8, model demonstrated superior speed. Its also significantly outperforms Faster R-CNN, Mask DINOv2, RT-DETRv2, markedly both small-object (head) performance efficiency.

Language: Английский

---

Hardware Accelerators for Real-Time Face Recognition: A Survey

IEEE Access, Journal Year: 2022, Volume and Issue: 10, P. 83723 - 83739

Published: Jan. 1, 2022

Real-time face recognition has been of great interest in the last decade due to its wide and variant critical applications which include biometrics, security public places, identification login systems. This encouraged researchers design fast accurate embedded portable systems that are capable detect recognize a large number faces at almost video frame rate. Due increasing volume reference faces, traditional general purpose computing engines such as ones based on Intel's Pentium processors have shown not be adequate various dedicated hardware accelerators either Graphical Processing Unit (GPU), Field Programmable Gate Arrays (FPGA), Application Specific Integrated Circuit (ASIC), or even multi-core Central Units (CPU) emerged. Earlier published review papers detection/recognition discussed detection algorithms enhancement improve accuracy. Nevertheless, none them reviewed used for this application. Accordingly, paper aims provide comprehensive most recent associated targeting real-time performance. A detailed comparison between neural network non-neural network-based terms accuracy processing time is provided. Discussions their suitability implemented into parallel architectures Single Instruction Multiple Thread (SIMT) Data (SIMD) also discussed.

Language: Английский

---

Automatic Face Recognition System Using Deep Convolutional Mixer Architecture and AdaBoost Classifier

Applied Sciences, Journal Year: 2023, Volume and Issue: 13(17), P. 9880 - 9880

Published: Aug. 31, 2023

In recent years, advances in deep learning (DL) techniques for video analysis have developed to solve the problem of real-time processing. Automated face recognition runtime environment has become necessary surveillance systems urban security. This is a difficult task due occlusion, which makes it hard capture effective features. Existing work focuses on improving performance while ignoring issues like small dataset, high computational complexity, and lack lightweight efficient feature descriptors. this paper, (FR) using Convolutional mixer (AFR-Conv) algorithm handle occlusion problems. A novel AFR-Conv architecture designed by assigning priority-based weight different patches along with residual connections an AdaBoost classifier automatically recognizing human faces. The also leverages strengths pre-trained CNNs extracting features ResNet-50, Inception-v3, DenseNet-161. combines these features’ weighted votes predict labels testing images. To develop system, we use data augmentation method enhance number datasets then used extract robust from Finally, recognize identity, utilized. For training evaluation model, set images collected online sources. experimental results approach are presented terms precision (PR), recall (RE), detection accuracy (DA), F1-score metrics. Particularly, proposed attains 95.5% PR, 97.6% RE, 97.5% DA, 98.5% 8500 show that our scheme outperforms advanced methods classification.

Language: Английский

---

Generative Data Augmentation applied to Face Recognition

2022 International Conference on Information Networking (ICOIN), Journal Year: 2023, Volume and Issue: unknown, P. 242 - 247

Published: Jan. 11, 2023

In this paper, we present a data augmentation method whose goal is to generate face images and maximize faces variation in the training set. The main objective break free from traditional techniques used deep neural networks such as geometric photometric transformations. Our consists generating using Deep Convolutional Generative Adversarial Networks (DC-GAN) feed with light pose variations of 2D plane. Its selective feature space augmentation. Then, apply resolution enhancement based on Enhanced Super Resolution GAN (ESRGAN), since generated are inferior noisy. As final step, perform verification Neural (CNNs) confirm robustness pipeline. found results achieves comparable performance comparison state-of-the-art methods.

Language: Английский

---