生物识别系统中的对抗攻击研究综述

A Review of Research on Adversarial Attacks in Biometric Systems

生物识别系统是当下一个备受关注的研究方向,其安全性也得到了公众的广泛关注。基于深度学习的生物识别系统容易受到对抗攻击,添加对抗扰动的图像会让深度学习模型做出错误的输出预测,从而破坏生物识别系统的安全性,达到非法攻击的目的。文章对针对生物识别系统的对抗攻击进行了调查,并研究了对抗攻击防御方法及对抗攻击在隐私保护方面的应用。最后,本文讨论了未来生物识别系统中的对抗攻击的发展方向,为未来的对抗攻击方式提供了可实现的解决方案。

Biometric recognition systems are currently a highly anticipated research direction,and their safety has also received widespread public attention.Deep learning based biometric systems are susceptible to adversarial attacks.Adding adversarial perturbations to images can cause the deep learning model to make incorrect output predictions,thereby undermining the security of the biometric system and achieving the goal of illegal attacks.This paper investigates adversarial attacks against biometric systems and studies the defense methods and applications of adversarial attacks in privacy protection.Finally,this paper discusses the development direction of adversarial attacks in future biometric systems,providing feasible solutions for future adversarial attack methods.