A Survey of Adversarial Attacks on Deep Neural Network in Computer Vision
Qi Wang, Jie Jiang, Jinyuan Mo, Qiuyu Kong, Yishan Li, Suru Feng · 2022
As an important part of artificial intelligence, deep learning has been widely used in computer vision. However, since deep neural networks lack of theoretical support, their security has been questioned by lots of researchers. Recent studies have shown that deep neural networks are susceptible to subtle perturbation. Researchers call the method of using perturbation to mislead the model adversarial attack. In this paper, we introduce some basic concepts of adversarial attacks in the computer vision field and analyze the causes of adversarial examples. We describe and compare some attack methods. Finally, we review the development of adversarial attack technology in computer vision and prospect potential development directions in this field.