Deep image prior based defense against adversarial examples

作者：

Highlights：

• A defense method against adversarial examples.

• Our method captures the image internal priors of the input itself.

• Our method follows a two-stage feature learning: robust and non-robust feature learning.

• We develop an adaptive stopping strategy.

• Our method is attack- and model-agnostic, and is robust to various attacks.

摘要

•A defense method against adversarial examples.•Our method captures the image internal priors of the input itself.•Our method follows a two-stage feature learning: robust and non-robust feature learning.•We develop an adaptive stopping strategy.•Our method is attack- and model-agnostic, and is robust to various attacks.

论文关键词：Deep neural network,Adversarial example,Image prior,Defense

论文评审过程：Received 14 August 2020, Revised 2 August 2021, Accepted 11 August 2021, Available online 7 September 2021, Version of Record 1 October 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108249