Tools
Tools
Nagaraj, Sahana (2025) Biometrics Security: Securing Facial Recognition Systems Against Deepfakes Spoofing Attacks. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (709kB) | Preview |
Preview |
PDF (Configuration Manual)
Download (650kB) | Preview |
Abstract
Biometric systems, with facial recognition technologies in particular, are increasingly susceptible to new forms of spoofing as embodied in deepfakes in the age of artificial intelligence (AI)-based generated media. Deepfakes enable generation of extremely realistic looking, yet entirely synthetic face representations. They are generated with advanced algorithms of deep learning networks like variational autoencoders (VAEs), diffusion model, and Generative Adversarial Networks (GANs). The fact these attacks can bypass traditional liveness detection and sensor-based countermeasures makes them dangerous to facial biometric identification systems. Of utmost importance then would be to ensure that facial recognition technology will never be manipulated as it is finding greater use in identity verification, control of access, and safe transactions.
In the present thesis, the deepfake detection system is proposed based on the DenseNet121 convolutional neural network (CNN) architecture and validated. The model is trained based on the Deepfake Detection Challenge (DFDC) dataset as one of the common benchmarks in deepfake research implemented through a transfer learning technique. The metrics of performance such as accuracy, precision, recall, F1-score, and ROC-AUC are deployed within the methodology to assess the skill in detecting the spoofing as well as generalizability in many different spoofing scenarios. By experimental results, the DenseNet121-based system will be more effective than the ordinary systems such as VGG-16 and ResNet-50, particularly regarding detecting high-fidelity changes to the face.
Actions (login required)
 |
View Item |