da/sec scientific talk on Biometrics

Topic: AI-Powered Dataset Collection: Leveraging Agents and Knowledge Graphs for Constructing Face Morphing Datasets

by Robert Nichols
D19/2.03a, February 20, 2025 (Thursday), 12.00 noon

Keywords — Biometric datasets, Machine learning, Face morphing, Big data

Abstract

„Morphing attacks in the border control scenario remain a threat to public security, which makes them an area of continued scientific interest. However, one of the persistent limiting factors in biometric machine- and deep-learning research, specifically morphing attack detection, is the lack of available, high-quality, and accurately labeled data spanning up to 10 years of sample age intervals. To this end, a novel data collection approach is proposed that integrates multiple state-of-the-art computer vision and natural language processing models with technologies that constitute the World Wide Web. The primary data source is live and archived news programs from German public television, and accurate labels are supplied by large-scale public knowledge graphs and the recent concept of AI Agents. State-of-the-art image morphing methods are applied to create a sophisticated database for future development of novel morphing attack detection systems. The results indicate a feasible approach to addressing the ubiquitous challenge of limited data availability in regulated jurisdictions.“