da/sec scientific talk on Biometrics

Topic: Towards Reliable Image Quality Assessment for Identity Documents

by Gregor Grote
D19/2.03a (also online via the corresponding BBB room), July 16, 2026 (Thursday), 12.00 noon

Keywords — Identity Documents, Computer Vision, Image Quality, OFIQ

Abstract

„The automatic processing of identity document images is a key component in remote enrollment and digital identity verification systems. However, the performance of downstream tasks such as presentation attack detection (PAD) is highly dependent on image quality, while no standardized approach for identity document image quality assessment currently exists. In this presentation, we adapt and extend established face image quality measures to the domain of identity cards and introduce additional document-specific metrics. We evaluate the proposed measures by analyzing their impact on PAD performance across multiple models and datasets using Error versus Discard Characteristic (EDC) curves. Furthermore, we propose a normalized aggregation scheme to enable consistent comparison across heterogeneous scenarios and discuss the relationship between perceived image quality and task-specific utility.“