2026-10-14T00:15:16Z https://api.figshare.com/v2/oai

oai:figshare.com:article/32633799 2026-10-01T17:00:19Z portal_1172 item_type_8 month_year_10_2026

Face-based biometric systems with deep learning in non-homogenous settings Glen Brown (24165915) PUREID: 411053350 Face recognition deep learning machine learning In the last decade, deep learning has led to a paradigm shift in the field of face-based biometrics. The modern performance of such facial verification / re-identification systems has made them ubiquitous: from being embedded in smart devices to operating at the nation-state / population level. State-of-the-art models require millions of images for training, and the acquisition of that data can be time-consuming / logistically difficult, and in the specific case of facial images for use in surveillance settings, fraught with potential ethical and legal violations. This thesis therefore focuses on analysing and reducing the performance gap between face-based biometric experiments, and real-world systems which often perform worse than the literature would otherwise suggest. This is attempted holistically, with an eye to lessening the aforementioned logistical and ethical issues throughout. Firstly, the performance of some typical deep learning models trained for facial verification / re-identification is investigated with regards to real-world cross-domain / multi-modal scenarios. This is achieved by contributing a novel dataset designed for the benchmarking of such systems. Data augmentation techniques, including the incorporation of synthetically generated data, are then explored for the purpose of mitigating the performance penalty when operating in cross-domain / multi-modal settings. Secondly, domain adaptation techniques are employed to overcome the domain gap in non-homogeneous settings. Initially, this is achieved by utilising domain transfer techniques during the post-training verification / re-identification stage, leading to a 12% performance improvement over baseline. Using such domain translation models to overcome the synthetic-to-real domain gap is also examined. Finally, the often overlooked aspect of neural network security is scrutinized: specifically, the issue of feature misuse in face-based biometric systems. This is where personally identifiable information / protected attributes are extracted from said feature vectors by a malicious actor. A new training paradigm is proposed to obfuscate such protected information rendering it irretrievable, and a comprehensive evaluation framework comprising three attack scenarios with varying levels of attacker prior-knowledge is contributed. 2026-10-01T17:00:19Z Text Thesis 10.17034/32633799.v1 https://figshare.com/articles/thesis/Face-based_biometric_systems_with_deep_learning_in_non-homogenous_settings/32633799 All Rights Reserved