On Adversarial Robustness of Deep Image Deblurring

TL;DR

This paper investigates the vulnerability of deep learning-based image deblurring methods to adversarial attacks, showing that small distortions can severely impair their performance and highlighting the need for robust training.

Contribution

It introduces adversarial attacks specific to image deblurring networks and evaluates their impact, emphasizing the importance of robustness in image recovery tasks.

Findings

Imperceptible distortions significantly degrade deblurring performance

Adversarial attacks can produce drastically different output images

Robust training is necessary for reliable image deblurring

Abstract

Recent approaches employ deep learning-based solutions for the recovery of a sharp image from its blurry observation. This paper introduces adversarial attacks against deep learning-based image deblurring methods and evaluates the robustness of these neural networks to untargeted and targeted attacks. We demonstrate that imperceptible distortion can significantly degrade the performance of state-of-the-art deblurring networks, even producing drastically different content in the output, indicating the strong need to include adversarially robust training not only in classification but also for image recovery.