A5037878863- Face recognition and analysis
- Biometric Identification and Security
- Domain Adaptation and Few-Shot Learning
- Adversarial Robustness in Machine Learning
- Anomaly Detection Techniques and Applications
- Multimodal Machine Learning Applications
- Topic Modeling
- Face and Expression Recognition
- Generative Adversarial Networks and Image Synthesis
- Digital Media Forensic Detection
- Spine and Intervertebral Disc Pathology
- Advanced Measurement and Detection Methods
- Infrared Target Detection Methodologies
- Optical measurement and interference techniques
- Video Surveillance and Tracking Methods
- Cervical and Thoracic Myelopathy
- Traumatic Brain Injury and Neurovascular Disturbances
- Spinal Fractures and Fixation Techniques
- Misinformation and Its Impacts
- Reconstructive Facial Surgery Techniques
- Cardiac Arrest and Resuscitation
- Artificial Immune Systems Applications
- Machine Learning and ELM
- Forensic and Genetic Research
- Advanced Steganography and Watermarking Techniques
Beijing Chao-Yang Hospital
2025
Capital Medical University
2024-2025
Harbin Institute of Technology
2023-2025
China Mobile (China)
2025
Shenzhen Institute of Information Technology
2025
Tianjin People's Hospital
2024
Nanyang Technological University
2022-2023
Hong Kong Baptist University
2017-2022
Tianjin Chengjian University
2020
Baidu (China)
2018
Face presentation attacks have become an increasingly critical issue in the face recognition community. Many anti-spoofing methods been proposed, but they cannot generalize well on "unseen" attacks. This work focuses improving generalization ability of from perspective domain generalization. We propose to learn a generalized feature space via novel multi-adversarial discriminative deep framework. In this framework, is performed under dual-force triplet-mining constraint. ensures that learned...
Face presentation attacks have become an increasingly critical concern when face recognition is widely applied. Many anti-spoofing methods been proposed, but most of them ignore the generalization ability to unseen attacks. To overcome limitation, this work casts as a domain (DG) problem, and attempts address problem by developing new meta-learning framework called Regularized Fine-grained Meta-learning. let our model generalize well attacks, proposed trains perform in simulated shift...
In recent years, software-based face presentation attack detection (PAD) methods have seen a great progress. However, most existing schemes are not able to generalize well in more realistic conditions. The objective of this competition is evaluate and compare the generalization performances mobile PAD techniques under some real-world variations, including unseen input sensors, instruments (PAI) illumination conditions, on larger scale OULU-NPU dataset using its standard evaluation protocols...
With a large number of video surveillance systems installed for the requirement from industrial security, task object tracking, which aims to locate objects interest in videos, is very important. Although numerous tracking algorithms RGB videos have been developed decade, performance and robustness these may be degraded dramatically when information unreliable (e.g., poor illumination conditions or low resolution). To address this issue, paper presents new system, combine infrared modalities...
Three-dimensional mask spoofing attacks have been one of the main challenges in face recognition. Compared with a 3D mask, real displays different facial motion patterns that are reflected by dynamic textures. However, large portion these differences is subtle. We find subtle can be fully captured multiple deep textures from convolutional layer neural network, but not all spatial regions and channels useful for differentiation motions between faces masks. In this paper, we propose novel...
Misinformation has become a pressing issue. Fake media, in both visual and textual forms, is widespread on the web. While various deepfake detection text fake news methods have been proposed, they are only designed for single-modality forgery based binary classification, let alone analyzing reasoning subtle traces across different modalities. In this paper, we high-light new research problem multi-modal namely Detecting Grounding Multi-Modal Media Manipulation (DGM <sup...
3D mask spoofing attack has been one of the main challenges in face recognition. A real displays a different motion behaviour compared to spoof attempt, which is reflected by facial dynamic textures. However, information usually exists subtle texture level, cannot be fully differentiated traditional hand-crafted texture-based methods. In this paper, we propose novel method for anti-spoofing, namely deep convolutional learning, learns robust from fine-grained features. Moreover,...
Misinformation has become a pressing issue. Fake media, in both visual and textual forms, is widespread on the web. Whilevarious deepfake detection text fake news methods have been proposed, they are only designed for single-modality forgery based binary classification, let alone analyzing reasoning subtle traces across different modalities. In this paper, we highlight new research problem multi-modal namely <bold xmlns:mml="http://www.w3.org/1998/Math/MathML"...
This paper proposes a framework for the calibration of multi-camera and multi-LiDAR system. It utilizes Apriltags to build environment solve poses extrinsics cameras then deploy ICP-like algorithm LiDARs. Compared previous extrinsic methods, this proposed naturally applies systems with different numbers configurations The procedure produces not only accurately robustly, but efficiently without inconvenient human manipulation. provides general, efficient standardizable solution autonomous...
Face presentation attack detection (fPAD) plays a critical role in the modern face recognition pipeline. An fPAD model with good generalization can be obtained when it is trained images from different input distributions and types of spoof attacks. In reality, training data (both real images) are not directly shared between owners due to legal privacy issues. this article, motivation circumventing challenge, we propose federated (FedPAD) framework that simultaneously takes advantage rich...
Predicting neurological prognosis after cardiac arrest remains challenging. Somatosensory evoked potential N20 absence is highly specific but lacks sensitivity. Glial fibrillary acidic protein and gene product 9.5 are biomarkers for brain injury, yet their roles in patients with preserved somatosensory remain underexplored. From January 2023 to December 2024, 69 were enrolled, of whom 46 had responses. Serum glial protein, neuron-specific enolase levels measured at 72 h post-resuscitation....
As biological control agents, bacteriophages can both inhibit the pathogenic bacteria and remove bacterial biofilms from seafood. Vibrio cholerae is pathogen of cholera severe infection could lead watery diarrhea even death. The double-layer agar plate method was used to isolate screen V. samples aquaculture water sewage. Purified were examined through genome sequencing, as well morphological characterizations. Among isolated bacteriophages, bacteriophage VC3 found be a long-tailed...
Infrared sensors have been deployed in many video surveillance systems because of the insensibility their imaging procedure to some extreme conditions (e.g. low illumination condition, dim environment). To reduce human labor monitoring and perform intelligent infrared understanding, an important issue we need consider is how locate object interest consecutive frames accurately. Therefore, developing a robust tracking algorithm for videos necessary. However, information may not be reliable...
In multimedia analysis, one objective of unsupervised visual domain adaptation is to train a classifier that works well on target given labeled source samples and unlabeled samples. Feature alignment two domains the key issue which should be addressed achieve this objective. Inspired by recent study Generative Adversarial Networks (GAN) in adaptation, paper proposes new model based Network, named Hierarchical Deep Network (HADN), jointly optimizes feature-level pixel-level adversarial within...
This study aimed to explore the characteristics of abnormal regional resting-state functional magnetic resonance imaging (rs-fMRI) activity in comatose patients early period after cardiac arrest (CA), and investigate their relationships with neurological outcomes. We also explored correlations between jugular venous oxygen saturation (SjvO2) rs-fMRI resuscitated patients. examined relationship amplitude N20-baseline within intracranial conduction pathway somatosensory evoked potentials...