Towards adversarial robustness verification of no-reference image- and video-quality metrics

Ekaterina Shumitskaya, Anastasia Antsiferova, Dmitriy S. Vatolin

Abstract: Highlights•Modern image- and video-quality metrics are vulnerable to adversarial attacks.•It is possible to attack metrics fast even for high-resolution images/video frames.•Metrics that currently lead in existing benchmarks are most vulnerable to attacks.•Input transformations improve the robustness of vulnerable quality metrics.