da/sec scientific talk on Explainability in Biometrics

Topic: Explaining OFIQ: Understanding Face Image Quality Assessments

by Ana Real
D19/2.03a (also online via the corresponding BBB room), January 15, 2026 (Thursday), 12.00 noon

Keywords — Face image, Biometrics, FIQA, Large Language Models, Data analysis

Abstract

„Face image quality assessment is used to evaluate the suitability of face images for biometric purposes, including individual verification and recognition. Given the current deployment of unsupervised face-recognition systems and self-enrolment applications, guidance on capturing high-quality face images is required. For this reason, the integration of descriptive explanations concerning face image quality assessment during enrolment is considered important. In this scientific talk, the accuracy of large language models in identifying defects will be discussed, together with the relationship between quality scores obtained by analysing images with OFIQ. The correct classification and description of such defects is expected to improve the usability of FIQA scores in unsupervised face-image enrolment scenarios.“