Impact and Mitigation of Quality Degradation for Differential Morphing Attack Detection

Differential Morphing Attack Detection (D-MAD) aims to detect morphed face images by comparing the suspected potential morph against a trusted live capture. While the quality of a suspected image used in a passport or similar will typically be controlled to avoid degradation by environmental factors such as lighting, the quality of the trusted live capture could more easily vary in real conditions. This paper examines how a DMAD system is affected by varying trusted live capture quality, how the same quality variations impact Face Image Quality Assessment (FIQA), and demonstrates how said FIQA can be utilized to substantially mitigate the D-MAD performance impact via another model. The experiments in particular consider four synthetic and thus clearly controlled image defect types, two corresponding to environmental lighting variation and two to blur, all based on approaches from the NIST FATE Quality SIDD report. The tested D-MAD system is based on deep face representations and the tested FIQA algorithms are parts of the recently established OFIQ project.