Pub Date : 2026-02-12DOI: 10.1109/TIFS.2026.3664007
Xiaolan Zhu;Junfeng Wang;Wenhan Ge;Xinbo Han
Although encryption offers strong anonymity, it also facilitates the concealment of malicious activities, allowing adversaries to evade detection, and posing a great challenge to cybersecurity surveillance. Many existing encrypted traffic classification methods struggle to integrate flow- and packet-level tasks effectively, as they are trained independently, which is redundancy. Additionally, packet header and payload are treated equally, leading to the rich information in raw bytes remains fully unexplored, particularly in the abundant payload data. Moreover, they neglect the semantic invariance and common features between data samples, which ultimately results in suboptimal performance. To address these challenges, we propose an effective Multi-Task model using Dual Embedding and Graph Contrastive Learning (MT-DEGCL). Based on the byte-packet-flow structure of network traffic, a parallel dual embedding embeds the header and payload separately, followed by a cross-gated feature fusion strategy to capture the strong local packet-level representation. Then, we construct the traffic interaction graph and further utilize graph contrastive learning to extract the robust global flow-level representation. Finally, a multi-task model is trained for joint flow- and packet-level classification, leveraging the complementary learning between tasks to enhance overall performance. The experimental results on four real datasets highlight the effectiveness of MT-DEGCL, demonstrating superior performance in both tasks. Specifically, on the ISCX-Tor dataset, MT-DEGCL achieves F1 scores of 98.63% for flow-level classification and 98.10% at the packet level, surpassing the state-of-the-art (i.e., DE-GNN) by 2.03% and 83.21%, respectively. Furthermore, MT-DEGCL maximizes the rich information in raw payload bytes, significantly reducing or even nearly eliminating classification loss when using only payload data.
{"title":"MT-DEGCL: Multi-Task Encrypted Traffic Classification With Dual Embedding and Graph Contrastive Learning","authors":"Xiaolan Zhu;Junfeng Wang;Wenhan Ge;Xinbo Han","doi":"10.1109/TIFS.2026.3664007","DOIUrl":"10.1109/TIFS.2026.3664007","url":null,"abstract":"Although encryption offers strong anonymity, it also facilitates the concealment of malicious activities, allowing adversaries to evade detection, and posing a great challenge to cybersecurity surveillance. Many existing encrypted traffic classification methods struggle to integrate flow- and packet-level tasks effectively, as they are trained independently, which is redundancy. Additionally, packet header and payload are treated equally, leading to the rich information in raw bytes remains fully unexplored, particularly in the abundant payload data. Moreover, they neglect the semantic invariance and common features between data samples, which ultimately results in suboptimal performance. To address these challenges, we propose an effective Multi-Task model using Dual Embedding and Graph Contrastive Learning (MT-DEGCL). Based on the byte-packet-flow structure of network traffic, a parallel dual embedding embeds the header and payload separately, followed by a cross-gated feature fusion strategy to capture the strong local packet-level representation. Then, we construct the traffic interaction graph and further utilize graph contrastive learning to extract the robust global flow-level representation. Finally, a multi-task model is trained for joint flow- and packet-level classification, leveraging the complementary learning between tasks to enhance overall performance. The experimental results on four real datasets highlight the effectiveness of MT-DEGCL, demonstrating superior performance in both tasks. Specifically, on the ISCX-Tor dataset, MT-DEGCL achieves F1 scores of 98.63% for flow-level classification and 98.10% at the packet level, surpassing the state-of-the-art (i.e., DE-GNN) by 2.03% and 83.21%, respectively. Furthermore, MT-DEGCL maximizes the rich information in raw payload bytes, significantly reducing or even nearly eliminating classification loss when using only payload data.","PeriodicalId":13492,"journal":{"name":"IEEE Transactions on Information Forensics and Security","volume":"21 ","pages":"2220-2235"},"PeriodicalIF":8.0,"publicationDate":"2026-02-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146169663","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
As cloud computing advances, data owners increasingly upload large volumes of data to the cloud. Attribute-based searchable encryption (ABSE) empowers data owners to manage fine-grained access over encrypted cloud files, and supports keyword-based search for authorized users. However, current multi-owner searchable encryption schemes often suffer from efficiency limitations and vulnerabilities to keyword guessing attacks. Furthermore, access policies are typically stored in plain form, exposing confidential details about data owners and authorized users. To tackle the aforementioned issues, we put forward an expressive attribute-based searchable encryption scheme with full policy concealment. Our design leverages the reduced ordered binary decision diagram (ROBDD) for access control targeting multi-user and multi-owner environments. In our scheme, users can flexibly select data owners and utilize a single trapdoor to search across shared datasets. The integration of a warrant server that signs obfuscated keywords prevents the cloud server from launching effective keyword guessing attacks. The adoption of ROBDD enables complex access policies via boolean operations, thereby significantly enhancing the efficiency and flexibility of access control. Full policy hiding is achieved by mapping ROBDD paths to an improved bloom filter, preventing access policy leakage. We present formal definitions and security models of the proposed approach, along with rigorous security proofs. Performance evaluation is conducted through theoretical analysis and simulations. Experimental indicate that our scheme achieves superior efficiency over state-of-the-art alternatives, offering a robust solution for secure and flexible cloud data management.
{"title":"Expressive and Fully Policy-Hidden Attribute-Based Searchable Encryption Scheme for Multi-Owner","authors":"Qing Miao;Jiguo Li;Yang Lu;Hang Cheng;Yichen Zhang;Jian Shen","doi":"10.1109/TIFS.2026.3663992","DOIUrl":"10.1109/TIFS.2026.3663992","url":null,"abstract":"As cloud computing advances, data owners increasingly upload large volumes of data to the cloud. Attribute-based searchable encryption (ABSE) empowers data owners to manage fine-grained access over encrypted cloud files, and supports keyword-based search for authorized users. However, current multi-owner searchable encryption schemes often suffer from efficiency limitations and vulnerabilities to keyword guessing attacks. Furthermore, access policies are typically stored in plain form, exposing confidential details about data owners and authorized users. To tackle the aforementioned issues, we put forward an expressive attribute-based searchable encryption scheme with full policy concealment. Our design leverages the reduced ordered binary decision diagram (ROBDD) for access control targeting multi-user and multi-owner environments. In our scheme, users can flexibly select data owners and utilize a single trapdoor to search across shared datasets. The integration of a warrant server that signs obfuscated keywords prevents the cloud server from launching effective keyword guessing attacks. The adoption of ROBDD enables complex access policies via boolean operations, thereby significantly enhancing the efficiency and flexibility of access control. Full policy hiding is achieved by mapping ROBDD paths to an improved bloom filter, preventing access policy leakage. We present formal definitions and security models of the proposed approach, along with rigorous security proofs. Performance evaluation is conducted through theoretical analysis and simulations. Experimental indicate that our scheme achieves superior efficiency over state-of-the-art alternatives, offering a robust solution for secure and flexible cloud data management.","PeriodicalId":13492,"journal":{"name":"IEEE Transactions on Information Forensics and Security","volume":"21 ","pages":"2416-2429"},"PeriodicalIF":8.0,"publicationDate":"2026-02-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146169661","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Website fingerprinting (WF) attacks, which covertly monitor user communications to identify the web pages they visit, pose a serious threat to user privacy. Existing WF defenses attempt to reduce attack accuracy by disrupting traffic patterns, but attackers can retrain their models to adapt, making these defenses ineffective. Meanwhile, their high overhead limits deployability. To overcome these limitations, we introduce a novel controllable website fingerprinting defense called TrapFlow based on backdoor learning. TrapFlow exploits the tendency of neural networks to memorize subtle patterns by injecting crafted trigger sequences into targeted website traffic, causing the attacker’s model to build incorrect associations during training. If the attacker attempts to adapt by training on such noisy data, TrapFlow ensures that the model internalizes the trigger as a dominant feature, leading to widespread misclassification across unrelated websites. Conversely, if the attacker ignores these patterns and trains only on clean data, the trigger behaves as an adversarial patch at inference time, causing model misclassification. To achieve this dual effect, we optimize the trigger using the Fast Levenshtein-like distance to maximize both its learnability and distinctiveness from normal traffic. Experiments show that TrapFlow significantly reduces the accuracy of the RF attack from 99% to 6% with 74% data overhead. This compares favorably against two SOTA defenses: FRONT reduces accuracy by only 2% at a similar overhead, while Palette achieves 32% accuracy, but with 48% more overhead. We further validate the practicality of our method in a real Tor network environment.
{"title":"TrapFlow: Controllable Website Fingerprinting Defense via Dynamic Backdoor Learning","authors":"Siyuan Liang;Jiajun Gong;Tianmeng Fang;Aishan Liu;Tao Wang;Xiaochun Cao;Dacheng Tao;Ee-Chien Chang","doi":"10.1109/TIFS.2026.3663989","DOIUrl":"10.1109/TIFS.2026.3663989","url":null,"abstract":"Website fingerprinting (WF) attacks, which covertly monitor user communications to identify the web pages they visit, pose a serious threat to user privacy. Existing WF defenses attempt to reduce attack accuracy by disrupting traffic patterns, but attackers can retrain their models to adapt, making these defenses ineffective. Meanwhile, their high overhead limits deployability. To overcome these limitations, we introduce a novel controllable website fingerprinting defense called TrapFlow based on backdoor learning. TrapFlow exploits the tendency of neural networks to memorize subtle patterns by injecting crafted trigger sequences into targeted website traffic, causing the attacker’s model to build incorrect associations during training. If the attacker attempts to adapt by training on such noisy data, TrapFlow ensures that the model internalizes the trigger as a dominant feature, leading to widespread misclassification across unrelated websites. Conversely, if the attacker ignores these patterns and trains only on clean data, the trigger behaves as an adversarial patch at inference time, causing model misclassification. To achieve this dual effect, we optimize the trigger using the Fast Levenshtein-like distance to maximize both its learnability and distinctiveness from normal traffic. Experiments show that TrapFlow significantly reduces the accuracy of the RF attack from 99% to 6% with 74% data overhead. This compares favorably against two SOTA defenses: FRONT reduces accuracy by only 2% at a similar overhead, while Palette achieves 32% accuracy, but with 48% more overhead. We further validate the practicality of our method in a real Tor network environment.","PeriodicalId":13492,"journal":{"name":"IEEE Transactions on Information Forensics and Security","volume":"21 ","pages":"2610-2625"},"PeriodicalIF":8.0,"publicationDate":"2026-02-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146169659","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-09DOI: 10.1109/tifs.2026.3663056
Mengsha Kou, Xiaoyu Xia, Ibrahim Khalil, Ziqi Wang, Xiuzhen Zhang, Lin Yao, Minhui Xue
{"title":"Data Flipping Attack and Defense in Web Edge Caching Systems","authors":"Mengsha Kou, Xiaoyu Xia, Ibrahim Khalil, Ziqi Wang, Xiuzhen Zhang, Lin Yao, Minhui Xue","doi":"10.1109/tifs.2026.3663056","DOIUrl":"https://doi.org/10.1109/tifs.2026.3663056","url":null,"abstract":"","PeriodicalId":13492,"journal":{"name":"IEEE Transactions on Information Forensics and Security","volume":"60 25 1","pages":""},"PeriodicalIF":6.8,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146160269","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Recently, backdoor attack, which aims to implant malicious logic into deep learning models (DLMs), has attracted so extensive research attention. Among them, the non-poisoning-based backdoor attack appears considerable development prospects owing to the posed threats against the DLMs-based artificial intelligence applications in cyberspace. However, previous non-poisoning-based backdoor attacks for DLMs are limited to the impractical attacking forms, resulting in certain weaknesses in both attacking complexity and attacking adaptability. To tackle the mentioned issues, this paper proposes a novel backdoor attack framework, namely the shell code injection (SCI), to perform backdoor attacks against DLMs with lower complexity and higher adaptability. Specifically, for alleviating the attacking complexity, we elaborate the logic-driven stealthy backdoor shell motivated by the biological behavior in nature, e.g., the camouflage and attack strategy of crabs. By introducing the trigger consistency verification and short-circuit code packaging strategies, the SCI misleads the victim models to output wrong predictions without training requirements according to the preset poisonous decision logic. For enhancing the attacking adaptability, we design the LLM-assisted adaptive attacking target code generation that consists of the model concept detection module and the attack target adjusting module. Since the attacking goals could be generated dynamically according to the aware victim model information and appointed attacker preset instructions, the SCI could achieve more flexible attacking performance. Extensive experiments are conducted to demonstrate that the proposed backdoor attack framework appears awesome attacking ability (almost 100% ASR) under various settings. Additionally, we provide a case study on combining the cyber attack with SCI, which also exhibits certain space for imagination of new-type backdoor attacks. The code is released at https://github.com/WDQhello/Shell_attack/
{"title":"Practical and Flexible Backdoor Attack Against Deep Learning Models via Shell Code Injection","authors":"Jiakai Wang;Hao Liu;Renshuai Tao;Jian Sun;Xianglong Liu;Yao Zhao","doi":"10.1109/TIFS.2026.3662587","DOIUrl":"10.1109/TIFS.2026.3662587","url":null,"abstract":"Recently, backdoor attack, which aims to implant malicious logic into deep learning models (DLMs), has attracted so extensive research attention. Among them, the non-poisoning-based backdoor attack appears considerable development prospects owing to the posed threats against the DLMs-based artificial intelligence applications in cyberspace. However, previous non-poisoning-based backdoor attacks for DLMs are limited to the impractical attacking forms, resulting in certain weaknesses in both attacking complexity and attacking adaptability. To tackle the mentioned issues, this paper proposes a novel backdoor attack framework, namely the shell code injection (SCI), to perform backdoor attacks against DLMs with lower complexity and higher adaptability. Specifically, for alleviating the attacking complexity, we elaborate the logic-driven stealthy backdoor shell motivated by the biological behavior in nature, e.g., the camouflage and attack strategy of crabs. By introducing the trigger consistency verification and short-circuit code packaging strategies, the SCI misleads the victim models to output wrong predictions without training requirements according to the preset poisonous decision logic. For enhancing the attacking adaptability, we design the LLM-assisted adaptive attacking target code generation that consists of the model concept detection module and the attack target adjusting module. Since the attacking goals could be generated dynamically according to the aware victim model information and appointed attacker preset instructions, the SCI could achieve more flexible attacking performance. Extensive experiments are conducted to demonstrate that the proposed backdoor attack framework appears awesome attacking ability (almost 100% ASR) under various settings. Additionally, we provide a case study on combining the cyber attack with SCI, which also exhibits certain space for imagination of new-type backdoor attacks. The code is released at <uri>https://github.com/WDQhello/Shell_attack/</uri>","PeriodicalId":13492,"journal":{"name":"IEEE Transactions on Information Forensics and Security","volume":"21 ","pages":"2268-2283"},"PeriodicalIF":8.0,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146160270","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-09DOI: 10.1109/tifs.2026.3662200
Haonan Yuan, Wenyuan Wu, Jingwei Chen
{"title":"Privacy-Preserving, Efficient and Accurate Dimensionality Reduction","authors":"Haonan Yuan, Wenyuan Wu, Jingwei Chen","doi":"10.1109/tifs.2026.3662200","DOIUrl":"https://doi.org/10.1109/tifs.2026.3662200","url":null,"abstract":"","PeriodicalId":13492,"journal":{"name":"IEEE Transactions on Information Forensics and Security","volume":"16 1","pages":""},"PeriodicalIF":6.8,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146160271","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-03DOI: 10.1109/TIFS.2026.3660595
Dongyu Cao;Bixin Li;Huijie Zhang;Yong Wang;Lulu Wang
Blockchain technology improves supply chain management by ensuring the immutability of transaction records and facilitating process tracking. However, the transparency of blockchain raises significant privacy concerns, as sensitive information such as buyer and supplier qualifications, product specifications, and transaction amounts is often exposed. Compliance verification, which needs access to specific sensitive data for compliance checks, becomes challenging in blockchain-based privacy-preserving supply chains. This paper introduces ZKVeil, an innovative scheme utilizing zero-knowledge proof technology to maintain the confidentiality of sensitive information while ensuring compliance verification. Additionally, ZKVeil uses decentralized identifiers and verifiable credentials to ensure the authenticity of transaction data. A theoretical security analysis demonstrates the effectiveness of ZKVeil in safeguarding real sensitive data and ensuring compliance with regulations. To evaluate the performance of our scheme, we implement ZKVeil on a private blockchain of 100 nodes. Taking the shipbuilding supply chain transaction as an example, the experimental results demonstrate that ZKVeil incurs low gas consumption, execution time, and memory overhead.
{"title":"ZKVeil: A Privacy-Preserving Compliance Verification Scheme for Blockchain-Enabled Supply Chain Transactions","authors":"Dongyu Cao;Bixin Li;Huijie Zhang;Yong Wang;Lulu Wang","doi":"10.1109/TIFS.2026.3660595","DOIUrl":"10.1109/TIFS.2026.3660595","url":null,"abstract":"Blockchain technology improves supply chain management by ensuring the immutability of transaction records and facilitating process tracking. However, the transparency of blockchain raises significant privacy concerns, as sensitive information such as buyer and supplier qualifications, product specifications, and transaction amounts is often exposed. Compliance verification, which needs access to specific sensitive data for compliance checks, becomes challenging in blockchain-based privacy-preserving supply chains. This paper introduces ZKVeil, an innovative scheme utilizing zero-knowledge proof technology to maintain the confidentiality of sensitive information while ensuring compliance verification. Additionally, ZKVeil uses decentralized identifiers and verifiable credentials to ensure the authenticity of transaction data. A theoretical security analysis demonstrates the effectiveness of ZKVeil in safeguarding real sensitive data and ensuring compliance with regulations. To evaluate the performance of our scheme, we implement ZKVeil on a private blockchain of 100 nodes. Taking the shipbuilding supply chain transaction as an example, the experimental results demonstrate that ZKVeil incurs low gas consumption, execution time, and memory overhead.","PeriodicalId":13492,"journal":{"name":"IEEE Transactions on Information Forensics and Security","volume":"21 ","pages":"1858-1873"},"PeriodicalIF":8.0,"publicationDate":"2026-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146110445","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-03DOI: 10.1109/TIFS.2026.3660599
Haoyang Huang;Fengwei Zhang
Confidential Compute Architecture (CCA) is the latest Trusted Execution Environment (TEE) system on Arm. It offers a VM-level execution environment designed to host applications that manage security-sensitive tasks and safeguard them from malicious system software. Although this VM-level design simplifies TEE adoption, it introduces a large attack surface. Attackers can break isolation by exploiting vulnerabilities in any component of the VM. In this paper, we present HiveTEE, a scalable intra-TEE isolation architecture that leverages Realm Management Extension (RME) and Memory Tagging Extension (MTE). HiveTEE allows developers to partition applications into multiple isolated domains (SDoms), preventing a compromise in one part of the application from propagating across the entire TEE. To evaluate the performance overhead introduced by HiveTEE, we apply it to three real-world applications: OpenSSL, SQLite, and Memcached. The evaluation results show that HiveTEE incurs a small performance overhead (<3%).
{"title":"HiveTEE: Scalable and Fine-Grained Isolated Domains With RME and MTE Co-Assisted","authors":"Haoyang Huang;Fengwei Zhang","doi":"10.1109/TIFS.2026.3660599","DOIUrl":"10.1109/TIFS.2026.3660599","url":null,"abstract":"Confidential Compute Architecture (CCA) is the latest Trusted Execution Environment (TEE) system on Arm. It offers a VM-level execution environment designed to host applications that manage security-sensitive tasks and safeguard them from malicious system software. Although this VM-level design simplifies TEE adoption, it introduces a large attack surface. Attackers can break isolation by exploiting vulnerabilities in any component of the VM. In this paper, we present HiveTEE, a scalable intra-TEE isolation architecture that leverages Realm Management Extension (RME) and Memory Tagging Extension (MTE). HiveTEE allows developers to partition applications into multiple isolated domains (SDoms), preventing a compromise in one part of the application from propagating across the entire TEE. To evaluate the performance overhead introduced by HiveTEE, we apply it to three real-world applications: <monospace>OpenSSL</monospace>, <monospace>SQLite</monospace>, and <monospace>Memcached</monospace>. The evaluation results show that HiveTEE incurs a small performance overhead (<3%).","PeriodicalId":13492,"journal":{"name":"IEEE Transactions on Information Forensics and Security","volume":"21 ","pages":"2300-2312"},"PeriodicalIF":8.0,"publicationDate":"2026-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146110444","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Unsupervised Visible-Infrared Person Re-Identification (USL-VI-ReID) aims to match person images across visible and infrared modalities without identity annotations, addressing challenges such as cross-modal discrepancy and unlabeled data. Existing methods, however, often suffer from excessive sub-clusters, identity mixing, and unreliable cross-modal associations, which degrade matching performance. To overcome these issues, we propose MACHANet, a novel framework. The Memory Learning via Progressive Hybrid Clustering (MLPHC) module reduces excessive sub-clustering and enhances memory representations by first applying Harmonic Discrepancy Clustering with harmonic constraints and a core-edge mechanism, then gradually transitioning to DBSCAN as features become more discriminative. The Global Cross-Modal Positive Sample Alignment (GCPSA) module constructs a global set of cross-modal positive pairs, selecting the most similar visible–infrared samples of the same identity and computing alignment losses across intra- and inter-modalities. By maximizing mutual information and minimizing cross-modal distribution gaps, GCPSA effectively reduces modality discrepancies and suppresses noisy identity associations. Finally, the Multi-Modal Support Sample Expansion Alignment (MSSEA) module dynamically expands multi-modal support samples and incorporates residual-based representations to refine clusters, separate mixed identities, and progressively merge sub-identities. Extensive experiments on SYSU-MM01 and RegDB show that MACHANet outperforms existing state-of-the-art methods, including some supervised approaches. The source code will be publicly released.
{"title":"MACHANet: Memory-Augmented Cross Modal Hybrid Alignment Network for Unsupervised Visible-Infrared Person Re-Identification","authors":"Tingyu Yang;Weiqing Yan;Guanghui Yue;Wujie Zhou;Chang Tang","doi":"10.1109/TIFS.2026.3660597","DOIUrl":"10.1109/TIFS.2026.3660597","url":null,"abstract":"Unsupervised Visible-Infrared Person Re-Identification (USL-VI-ReID) aims to match person images across visible and infrared modalities without identity annotations, addressing challenges such as cross-modal discrepancy and unlabeled data. Existing methods, however, often suffer from excessive sub-clusters, identity mixing, and unreliable cross-modal associations, which degrade matching performance. To overcome these issues, we propose MACHANet, a novel framework. The Memory Learning via Progressive Hybrid Clustering (MLPHC) module reduces excessive sub-clustering and enhances memory representations by first applying Harmonic Discrepancy Clustering with harmonic constraints and a core-edge mechanism, then gradually transitioning to DBSCAN as features become more discriminative. The Global Cross-Modal Positive Sample Alignment (GCPSA) module constructs a global set of cross-modal positive pairs, selecting the most similar visible–infrared samples of the same identity and computing alignment losses across intra- and inter-modalities. By maximizing mutual information and minimizing cross-modal distribution gaps, GCPSA effectively reduces modality discrepancies and suppresses noisy identity associations. Finally, the Multi-Modal Support Sample Expansion Alignment (MSSEA) module dynamically expands multi-modal support samples and incorporates residual-based representations to refine clusters, separate mixed identities, and progressively merge sub-identities. Extensive experiments on SYSU-MM01 and RegDB show that MACHANet outperforms existing state-of-the-art methods, including some supervised approaches. The source code will be publicly released.","PeriodicalId":13492,"journal":{"name":"IEEE Transactions on Information Forensics and Security","volume":"21 ","pages":"1914-1925"},"PeriodicalIF":8.0,"publicationDate":"2026-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146110443","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：