Video Anonymisation

Data on the one hand has undoubtedly brought technological advances, on the other hand has also fostered the growing concern regarding privacy issues, in particular for applications where data transmission is performed for fog or cloud computing. In this project, we address the problem of visual privacy preservation for video content by obfuscating the personally identifiable information (PII) of a data subject, whose face is often the most identity-informative part. Classic face anonymisation techniques, e.g., blurring or pixelation, can effectively remove PII. However, this comes at a high cost of degrading other vision-related tasks, in particular for the action/emotion recognition where the facial poses play an essential role. Thanks to recent advances in generative adversarial networks (GANs), several anonymisation solutions have been proposed to generate natural-looking faces that correspond to different identities, while preserving the original facial poses either with or without the facial landmarks as the guidance. We further advance GAN-based face anonymisation techniques to address challenges in real-world video recordings. This work is funded by the EU project MARVEL.

Related publications

Graph-based Generative Face Anonymisation with Pose Preservation

N. Dall’Asen, Y. Wang, H. Tang, L. Zanella, E. Ricci, ICIAP, Lecce, IT, May 2022 [Paper] [Code]

Video Presentation is coming soon!