Ensemble-Based Biometric Verification: Defending Against Multi-Strategy Deepfake Image Generation
Name
computers-14-00225.pdf
Size
1.15 MB
Format
Adobe PDF
Checksum (MD5)
1b88a8bd00cf3bb5f02c8ec2e65470e6
Author(s)
Zen, Hilary
Wagh, Rohan
Wanderley, Miguel
Bicalho, Gustavo
Park, Rachel
Sun, Megan
Palacios, Rafael
Carvalho, Lucas
Rinaldo, Guilherme
Gupta, Amar
Date Issued
June 9, 2025
Journal
Computers
Publisher
Multidisciplinary Digital Publishing Institute
Citation
Zen, H.; Wagh, R.; Wanderley, M.; Bicalho, G.; Park, R.; Sun, M.; Palacios, R.; Carvalho, L.; Rinaldo, G.; Gupta, A. Ensemble-Based Biometric Verification: Defending Against Multi-Strategy Deepfake Image Generation. Computers 2025, 14, 225.
Version
Final published version
Abstract
Deepfake images, synthetic images created using digital software, continue to present a serious threat to online platforms. This is especially relevant for biometric verification systems, as deepfakes that attempt to bypass such measures increase the risk of impersonation, identity theft and scams. Although research on deepfake image detection has provided many high-performing classifiers, many of these commonly used detection models lack generalizability across different methods of deepfake generation. For companies and governments fighting identify fraud, a lack of generalization is challenging, as malicious actors may use a variety of deepfake image-generation methods available through online wrappers. This work explores if combining multiple classifiers into an ensemble model can improve generalization without losing performance across different generation methods. It also considers current methods of deepfake image generation, with a focus on publicly available and easily accessible methods. We compare our framework against its underlying models to show how companies can better respond to emerging deepfake generation methods.
MIT Department
Sloan School of Management
Terms of Use
Creative Commons Attribution
Persistent DSpace Link
DOI of Published Version
https://doi.org/10.3390/computers14060225
