Tools
Tools
Lopez Lopez, Roque Ivan (2025) TR-SONIC: A Multimodal Transformer-Based Sonar 3D Face Recognition System for Biometric Verification. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (936kB) | Preview |
Preview |
PDF (Configuration Manual)
Download (2MB) | Preview |
Abstract
This research presents TR-SONIC, a multimodal biometric authentication system that combines 3D facial geometry with simulated ultrasonic echoes processed using efficient transformer architectures. This work addresses the limitations of traditional 2D/3D vision-based facial recognition systems, which are vulnerable to poor lighting conditions, facial occlusions, impersonation attacks, and demographic biases. TR-SONIC proposes an innovative approach inspired by sonar technology, integrating two complementary models: SonarSimilarityTransformer, which compares acoustic signatures generated from 3D facial meshes, and FacialLandmarkTransformer, which employs K-Means clustering to detect anatomical landmarks. This dual-model architecture enables accurate and robust verification even in challenging scenarios.
The experiments, conducted on the FaceScape dataset, yielded an overall system accuracy of 95.7%, with the FacialLandmarkTransformer achieving 97.3% accuracy and the SonarSimilarityTransformer achieving 94.1% accuracy, and a 70% reduction in demographic bias metrics, while maintaining processing times of 0.42 seconds per authentication suitable for real-time environments. Furthermore, the system incorporates an intuitive graphical interface and an ethics monitoring framework that ensures compliance with emerging regulations on algorithmic fairness. TR-SONIC represents a significant advance toward more secure, fair, and suitable biometric systems for critical applications such as mobile banking and digital verification.
| Item Type: | Thesis (Masters) |
|---|---|
| Supervisors: | Name Email Haque, Rejwanul UNSPECIFIED |
| Uncontrolled Keywords: | Biometric Authentication; Transformer Architectures; Multimodal Biometrics; Acoustic Sonar; 3D Face Recognition/Modelling; Bias Mitigation |
| Subjects: | Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence T Technology > TK Electrical engineering. Electronics. Nuclear engineering > Biometric Identification |
| Divisions: | School of Computing > Master of Science in Artificial Intelligence for Business |
| Depositing User: | Ciara O'Brien |
| Date Deposited: | 24 Aug 2026 10:30 |
| Last Modified: | 24 Aug 2026 10:30 |
| URI: | https://norma.ncirl.ie/id/eprint/9590 |
Actions (login required)
 |
View Item |