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

作　　者：王逸飞 林建明 WANG Yifei;LIN Jianming(College of Computer and Infomation Technology of China Three Gorges University,Yichang 443002,China;Zhejiang Open University,Hangzhou 310012,China)

机构地区：[1]三峡大学计算机与信息学院,湖北宜昌443002 [2]浙江开放大学,浙江杭州310012

出　　处：《长江信息通信》2024年第4期47-50,共4页Changjiang Information & Communications

摘　　要：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.

关 键 词：对抗攻击 生物识别系统 深度学习 未来发展 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]