Face-Off: Adversarial Face Obfuscation

Abstract Advances in deep learning have made face recognition technologies pervasive. While useful to social media platforms and users, this technology carries significant privacy threats. Coupled with the abundant information they about service providers can associate users interactions, visited places, activities, preferences–some of which user may not want share. Additionally, facial models used by various agencies are trained data scraped from platforms. Existing approaches mitigate associated risks result an imbalanced trade-off between utility. In paper, we address proposing Face-Off, a privacy-preserving framework that introduces strategic perturbations images user’s prevent it being correctly recognized. To realize overcome set challenges related black-box nature commercial services, scarcity literature for adversarial attacks on metric networks. We implement evaluate Face-Off find deceives three services Microsoft, Amazon, Face++. Our study 423 participants further shows come at acceptable cost users.