Low-rank matrix recovery with total generalized variation for defending adversarial examples

作     者：Wen LI Hengyou WANG Lianzhi HUO Qiang HE Linlin CHEN Zhiquan HE Wing W.Y.Ng Wen LI;Hengyou WANG;Lianzhi HUO;Qiang HE;Linlin CHEN;Zhiquan HE;Wing W.Y.Ng

作者机构：School of ScienceBeijing University of Civil Engineering and ArchitectureBeijing 100044China School of Computer Science and EngineeringSouth China University of TechnologyGuangzhou 510006China Aerospace Information Research InstituteChinese Academy of SciencesBeijing 100094China Guangdong Key Laboratory of Intelligent Information ProcessingShenzhen 518060China Institute of Big Data Modeling and TechnologyBeijing University of Civil Engineering and ArchitectureBeijing 100044China 

出 版 物：《Frontiers of Information Technology & Electronic Engineering》 (信息与电子工程前沿（英文版）)

年 卷 期：2024年第25卷第3期

页      面：432-445页

核心收录：

学科分类：08[工学] 081201[工学-计算机系统结构] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：Project supported by the National Natural Science Foundation of China(No.62072024) the Outstanding Youth Program of Beijing University of Civil Engineering and Architecture,China(No.JDJQ20220805) the Shenzhen Stability Support General Project(Type A),China(No.20200826104014001) 

主　　题：Total generalized variation Low-rank matrix Alternating direction method of multipliers Adversarial example 

摘      要：Low-rank matrix decomposition with first-order total variation(TV)regularization exhibits excellent performance in exploration of image *** advantage of its excellent performance in image denoising,we apply it to improve the robustness of deep neural ***,although TV regularization can improve the robustness of the model,it reduces the accuracy of normal samples due to its *** our work,we develop a new low-rank matrix recovery model,called LRTGV,which incorporates total generalized variation(TGV)regularization into the reweighted low-rank matrix recovery *** the proposed model,TGV is used to better reconstruct texture information without *** reweighted nuclear norm and Li-norm can enhance the global structure ***,the proposed LRTGV can destroy the structure of adversarial noise while re-enhancing the global structure and local texture of the *** solve the challenging optimal model issue,we propose an algorithm based on the alternating direction method of *** results show that the proposed algorithm has a certain defense capability against black-box attacks,and outperforms state-of-the-art low-rank matrix recovery methods in image restoration.