Improving the JPEG-resistance of Adversarial Attacks on Face Recognition by Interpolation Smoothing

• DOI: 10.48550/arxiv.2402.16586
• Publication Date: 2024-02-26
• Authors: Kefu Guo; Fengfan Zhou; Hefei Ling; Ping Li; Hui Liu
• Keywords: Smoothing; Interpolation; JPEG

JPEG compression can significantly impair the performance of adversarial face examples, which previous attacks on recognition (FR) have not adequately addressed. Considering this challenge, we propose a novel attack FR that aims to improve resistance examples against compression. Specifically, during iterative process generating interpolate into smaller size. Then utilize these interpolated create in next iteration. Subsequently, restore their original size by interpolating. Throughout entire process, our proposed method smooth perturbations, effectively mitigating presence high-frequency signals crafted are typically eliminated Our experimental results demonstrate effectiveness improving JPEG-resistance examples.