|
Amina Bassit I am a postdoctoral researcher advised by Vishnu Boddeti in the Human Analysis Lab at Michigan State University (US). I received my Ph.D. degree from the University of Twente (NL) in ML-based biometric recognition and applied cryptography as part of the Marie Skłodowska-Curie funded PriMa (Privacy Matters) project, under the supervision of Raymond Veldhuis, Andreas Peter, and Florian Hahn. I won the EAB Industry Award 2024 for my Ph.D. dissertation, Fast and Accurate Biometric Search under Encryption and its patent (NL2035809B1). I was a Program Chair of IWBF 2025, a co-organizer of the Trustworthy Biometrics special session at Eusipco 2025, and a presenter of the Computer Vision over Homomorphically Encrypted Data tutorial at CVPR 2025. |
|
ResearchMy research focuses on making AI technologies trustworthy and secure by preserving user privacy and protecting the intellectual property (IP) of models. I integrate advanced machine learning and applied cryptography to develop scalable, privacy-preserving solutions that build trust for both users and model owners. |
Publications |
|
Practical Biometric Search under Encryption: Meeting the NIST Runtime Requirement without Loss of Accuracy
Amina Bassit, Florian Hahn, Raymond Veldhuis, Andreas Peter IEEE T-BIOM, 2025 Encrypted biometric search at scale becomes practical with FHE supporting SIMD and vertical chunking on the database, enabling accurate one-in-a-million search in under 10 seconds. |
|
|
Template Recovery Attack on Encrypted Face Recognition Systems with Unprotected Decision using Synthetic Faces
Amina Bassit, Florian Hahn, Zohra Rezgui, Hatef Otroshi Shahreza, Raymond Veldhuis, Andreas Peter Frontiers in Imaging, 2025 Even HE-based BTPs cannot hide from synthetic spoofs—this study shows how attackers can recover and reconstruct facial templates using only synthetic faces and a handful of comparison scores. |
|
|
A Study on the Next Generation of Digital Travel Credentials
Ana-Teodora Radutoiu, Amina Bassit, Raymond Veldhuis, Christoph Busch IEEE BIOSIG, 2024 Future digital travel credentials—tomorrow's passports—can be secured using iris biometrics, enabling clone-resistant, device-portable authentication for seamless and secure travel. |
|
|
Improved Multiplication-Free Biometric Recognition under Encryption
Amina Bassit, Florian Hahn, Raymond Veldhuis, Andreas Peter IEEE TBIOM, 2024 By turning similarity computations into lookup tables, we unlock fast and secure biometric recognition under encryption without multiplications, achieving decisions in under 50ms. |
|
|
Fast and Accurate Biometric Search Under Encryption
Amina Bassit PhD Thesis - Dec 15th, 2023 - University of Twente Lookup table-based comparisons turn biometric verification and search under encryption into milliseconds and seconds instead of minutes and hours, enabling real-world deployment of privacy-preserving biometrics without accuracy loss. Supervised by Raymond Veldhuis, Andreas Peter, and Florian Hahn Funded by the PriMa (Privacy Matters) project under the Marie Skłodowska-Curie grant No. 860315 EAB Industry Award 2024 Patent published as NL2035809B1; WO2025058514A1 |
|
|
Template Recovery Attack on Homomorphically Encrypted Biometric Recognition Systems with Unprotected Threshold Comparison
Amina Bassit, Florian Hahn, Zohra Rezgui, Una Kelly, Raymond Veldhuis, Andreas Peter IEEE IJCB, 2023 Revealing comparison scores compromises BTPs' privacy. Our attack shows that a handful of fake templates and cleartext scores are enough to fully recover protected biometric templates and reconstruct raw biometric samples. |
|
|
Remote Cancelable Biometric System for Verification and Identification Applications
Hatef Otroshi Shahreza, Amina Bassit, Sébastien Marcel, Raymond Veldhuis IEEE BIOSIG, 2023 Dynamic protection meets cancelable biometrics. Our protocol enables secure, user-specific template replacement in both identification and verification without sacrificing flexibility or privacy. |
|
|
Multiplication-Free Biometric Recognition for Faster Processing under Encryption
Amina Bassit, Florian Hahn, Raymond Veldhuis, Andreas Peter IEEE IJCB, 2022 Multiplications out, lookup tables in! Our approach accelerates biometric verification under encryption without compromising recognition accuracy. Patent published as WO2025058509A1 |
|
|
Hybrid Biometric Template Protection: Resolving the Agony of Choice Between Bloom Filters and Homomorphic Encryption
Amina Bassit, Florian Hahn, Raymond Veldhuis, Andreas Peter IET Biometrics, 2022 Unlinkability meets efficiency! Our hybrid scheme blends Bloom filters with homomorphic encryption to deliver fast, accurate, and privacy-preserving iris recognition. |
|
|
Transferability analysis of adversarial attacks on gender classification to face recognition: Fixed and variable attack perturbation
Zohra Rezgui, Amina Bassit, Raymond Veldhuis IET Biometrics, 2022 Adversarial attacks reimagined! We show that fast, fixed-magnitude perturbations on gender classifiers can stealthily transfer to face recognition—unlocking a new path to soft biometric privacy without sacrificing identity utility. |
|
|
Fast and Accurate Likelihood Ratio-Based Biometric Verification Secure Against Malicious Adversaries
Amina Bassit, Florian Hahn, Joep Peeters, Tom Kevenaar, Raymond Veldhuis, Andreas Peter IEEE TIFS, 2021 Honest-but-curious isn't enough for biometric recognition. Our solution resists malicious adversaries by combining homomorphic encryption with zero-knowledge proofs, delivering fast, accurate biometric verification without leakage. |
|
|
Bloom Filter vs Homomorphic Encryption: Which approach protects the biometric data and satisfies ISO/IEC 24745?
Amina Bassit, Florian Hahn, Chris Zeinstra, Raymond Veldhuis, Andreas Peter IEEE BIOSIG, 2021 Not all protections are standard-proof. Our analysis reveals that only homomorphic encryption with verifiable computation satisfies ISO/IEC 24745 in trustless settings—while Bloom filters fall short on key privacy guarantees. Best Paper Award |
|
|
Transferability analysis of an adversarial attack on gender classification to face recognition
Zohra Rezgui, Amina Bassit IEEE BIOSIG, 2021 Task boundaries don't prevent adversarial transfer. Our analysis shows that attacks on gender classification can fool face recognition models, exposing cross-task vulnerabilities in shared biometric pipelines. |
Activities |
