معرفی
Sahar Husseini is a Research Fellow at EURECOM's Digital Security department. Her research focuses on deepfake detection, face verification, and cybersecurity. She has contributed to evaluating deepfake generators using frameworks like MetaHumans and 3D-assisted methods, addressing challenges such as beautification filters and spoofing attacks. Her work spans computer vision, biometrics, and machine learning, with publications at top conferences like ICCV, ICASSP, and ICIP. She collaborates with experts like Jean-Luc Dugelay on topics ranging from authentication systems to signal processing. Her thesis emphasizes strengthening face authentication in the deepfake era.
Research Interests:
- Developing robust methods against deepfake spoofing attacks.
- Enhancing face verification models via adaptive alignment techniques.
- Investigating the impact of image filters on deepfake detection.
- Leveraging 3D data and MetaHumans for generator evaluation.
- Optimizing CNN models for small datasets in color constancy tasks.
- Exploring optical flow for object tracking.
Key Contributions:
- Introduced RAW data's critical role in effective deepfake detection.
- Proposed AlignFace for pose/expression/illumination adaptation.
- Demonstrated vulnerabilities of detectors to beautification filters.
- Created frameworks for quantitative head motion replication analysis.
- Authored a comprehensive deepfake generator evaluation benchmark.
Labs/Teams: Active member of EURECOM's Digital Security research group, working on cutting-edge AI security solutions.
