LD-GAN: Learning perturbations for adversarial defense based on GAN structure

作者：

Highlights：

• We propose a novel noise-injection module to improve the flexibility of the network.

• We reproduce the image and avoid the risk of radically changing the feature.

• The experimental results demonstrate the superiority of our method.

摘要

•We propose a novel noise-injection module to improve the flexibility of the network.•We reproduce the image and avoid the risk of radically changing the feature.•The experimental results demonstrate the superiority of our method.

论文关键词：Adversarial attacks,Adversarial defense,Adversarial robustness,Image classification

论文评审过程：Received 20 July 2021, Revised 8 December 2021, Accepted 31 January 2022, Available online 8 February 2022, Version of Record 18 February 2022.

论文官网地址：https://doi.org/10.1016/j.image.2022.116659