FIVA: Facial Image and Video Anonymization and Anonymization Defense

In this paper, we present a new approach for facial anonymization in imagesand videos, abbreviated as FIVA. Our proposed method is able to maintain thesame face anonymization consistently over frames with our suggestedidentity-tracking and guarantees a strong difference from the original face.FIVA allows for 0 true positives for a false acceptance rate of 0.001. Our workconsiders the important security issue of reconstruction attacks andinvestigates adversarial noise, uniform noise, and parameter noise to disruptreconstruction attacks. In this regard, we apply different defense andprotection methods against these privacy threats to demonstrate the scalabilityof FIVA. On top of this, we also show that reconstruction attack models can beused for detection of deep fakes. Last but not least, we provide experimentalresults showing how FIVA can even enable face swapping, which is purely trainedon a single target image.