Existing steganalysis methods based on deep neural networks often rely on first-order statistical representations or generic attention mechanisms to amplify embedding disturbances, which may show limited sensitivity to weak, spatially distributed steganographic perturbations. Prior studies also focus on local feature extraction and underutilize broader spatial information. To address these challenges, we propose VADNet, a statistical representation framework for spatial-domain steganalysis that enhances weak embedding disturbances through variance-guided feature recalibration and multi-scale spatial dependency extraction with low overhead. We analyze the statistical characteristics of steganographic embedding, emphasizing the limitations of first-order-statistics-based attention mechanisms in characterizing weak embedding-induced feature fluctuations. A Variance Attention Module (VAM) based on second-order statistics is introduced in the preprocessing stage to recalibrate channel weights by calculating spatial channel-feature variance, thereby enhancing embedding-disturbance-related features and suppressing irrelevant channel features. A multi-scale feature extraction structure with dilated convolutions enlarges the receptive field and captures embedding-disturbance distribution characteristics at different spatial scales, improving discriminative ability. Experiments on BOSSBase-1.01 and BOWS2 demonstrate effectiveness across various payloads. The proposed approach improves detection accuracy by 0.4%-18.22% for WOW, 0.37%-15.89% for S-UNIWARD, and 0.5%-13.18% for HILL. Compared with representative attention mechanisms, VADNet achieves the highest detection accuracy while maintaining competitive computational efficiency.

Worldwide image geo-localization aims to determine the capture location of an image on a global scale. Existing methods often mislocalize images by matching them to visually similar scenes from different geographic regions, which limits reliability in practical applications. To address this issue, we propose TransGeoCLIP, a novel retrieval-based framework that integrates a location attention mechanism and large multimodal models (LMMs). Using the Transformer encoder with location attention to encode GPS coordinates, TransGeoCLIP can effectively distinguish geographic features among visually similar images. The framework consists of two stages: 1) Retrieval database construction, which employs Transformers equipped with location attention mechanisms to encode labeled GPS coordinates and enhance location semantics, subsequently enables joint image-text-GPS embedding through CLIP; 2) Retrieval-augmented inference, which leverages LMMs to infer the final image location prediction from retrieved database results. Extensive experimental results on diverse datasets, including IM2GPS, IM2GPS3k, YFCC4k, and YFCC26k, demonstrate that TransGeoCLIP significantly enhances localization performance for visually similar images. Particularly, street-level localization accuracy (within 1 km error) is substantially improved, surpassing state-of-the-art methods by 1.5

The misuse of deep learning-based facial manipulation poses a serious threat to civil rights. To prevent such fraud at its source, proactive defense methods have been proposed that embed invisible adversarial perturbations into images, disrupting the manipulation process and rendering the forged output unconvincing to observers. However, non-targeted disruption of the output may leave identifiable facial features intact, potentially leading to the stigmatization of individuals. In this work, we propose a universal framework for combating facial manipulation, termed ID-Guard. The framework employs a single forward pass of an encoder-decoder network to generate cross-model transferable adversarial perturbations. We introduce a novel Identity Destruction Module (IDM) to suppress identifiable features in manipulated faces. The perturbation generation is optimized by formulating the disruption of various manipulation types as a multi-task learning problem, with a dynamic weighting strategy designed to enhance cross-model performance. Experimental results show that ID-Guard effectively defends against diverse facial manipulation models while degrading identifiable regions in manipulated images. It also enables disrupted images to evade facial inpainting and facial recognition systems. Moreover, ID-Guard can be seamlessly integrated as a plug-and-play component into other tasks, such as adversarial training.

Digital images have been widely applied in various fields such as government affairs, healthcare, and finance, making the protection of image content crucial. Particularly with the rapid development of current deepfake technology, the establishment of effective image anti-counterfeiting and copyright authentication mechanisms has become increasingly urgent. However, existing forgery defense methods based on image camouflage and active interference often face issues such as the need for improved reconstructed image quality and unclear ownership due to insufficient utilization of the latent space. To address the above problems, we propose an image content protection framework based on semantic transfer. First, this framework encodes the secret image into a high-dimensional latent semantic vector by training a private diffusion model. Second, a public diffusion model is used to decode this vector, ensuring that the camouflage image possesses the specified semantics. Finally, on the premise of not damaging the reconstructed semantics of the secret image, reversible pixel embedding technology is employed to embed ownership information into the secret image. Experimental results show that the proposed framework can effectively protect image content: the peak signal-to-noise ratio (PSNR) of the image reconstructed by authorized users is 17.27% higher than that of the comparison methods, while unauthorized users cannot reconstruct the secret image at all. In addition, while achieving image content protection, the proposed framework can add authentication information and realize 100.00% accurate recovery of ownership identifiers.

Website fingerprinting attacks leverage encrypted traffic features to identify specific services accessed by users within anonymity networks such as Tor. Although existing WF methods achieve high accuracy on static datasets using deep learning techniques, they struggle in dynamic environments where anonymous Websites continually evolve. These methods typically require full retraining on composite datasets, resulting in substantial computational and storage burdens, and are particularly vulnerable to classification bias caused by data imbalance and concept drift. To address these challenges, we propose EIL-WF, a dynamic WF framework based on incremental learning that enables efficient adaptation to newly emerging websites without the need for full retraining. EIL-WF incrementally trains lightweight, independent classifiers for new website classes and integrates them through classifier normalization and energy alignment strategies grounded in energy-based model theory, thereby constructing a unified and robust classification model. Comprehensive experiments on two public Tor traffic datasets demonstrate that EIL-WF outperforms existing incremental learning methods by 6.2%-20.2% in identifying new websites and reduces catastrophic forgetting by 5.4%-20%. Notably, EIL-WF exhibits strong resilience against data imbalance and concept drift, maintaining stable classification performance across evolving distributions. Furthermore, EIL-WF decreases training time during model updates by 2-3 orders of magnitude, demonstrating substantial advantages over conventional full retraining paradigms.

Most existing image steganography without modification (SWM) methods leverage stable features for covert transmission that resist common channel attacks. However, their limited feature space restricts hiding capacity. To this end, we propose a robust SWM method based on co-occurrence labels (CooL-SWM). It exploits both the semantic stability and co-occurrence patterns of object category labels. Specifically, CooL-SWM extracts co-occurrence features from natural images and constructs co-occurrence matrices through label permutations, expanding the feature space. By using unique spatial positions, CooL-SWM establishes a bijective mapping between secret messages and co-occurrence features. This improves feature space utilization and hiding capacity. We also introduce a structured text prompt template and a label-to-image semantic consistency strategy. These ensure the generated stego images strictly match the co-occurrence label sequences, improving hiding success rate and extraction accuracy. Different from existing object-based SWM methods, CooL-SWM requires no auxiliary information such as the number of stego objects per transmission, obviating additional transmission risks. Experimental results based on VOC2012 and COCO2014 demonstrate that, while ensuring complete hiding of secret messages, the proposed CooL-SWM achieves 1.75 times and 2.33 times higher hiding capacity compared with existing object-based SWM methods, with average extraction accuracy improvements of 23.48% and 30.59%, respectively, showcasing its application prospects in real-world lossy network channels.

Binary Code Similarity Detection (BCSD) plays a critical role in software security. However, accurately measuring the similarity between binaries compiled from the same source code under different compilation configurations remains a major challenge. To address this issue, we propose BinEncoder, a novel BCSD framework that extracts high-level semantics of binary functions through a microcode-based representation. Specifically, BinEncoder lifts binary code into Hex-Rays microcode, an intermediate representation, and decomposes composite instructions into nested instructions to expose fine-grained semantics. It then employs a Transformer model pre-trained on three tailored tasks to capture structural and contextual information. Extensive experiments demonstrate that BinEncoder achieves superior performance over representative BCSD baselines across multiple BCSD tasks and shows promising potential for retrieval-based vulnerability detection.

IP geolocation technique determines the geographic location of a target network device based on its IP address, which gives it significant application value in the field of cybersecurity for services such as cyber attack traceback and cybercrime tracking. Client-independent IP geolocation technology can obtain a user’s geographic location without requiring active user cooperation. The current state-of-the-art client-independent IP geolocation methods are based on graph learning. In particular, subgraph-based IP geolocation methods have demonstrated significant advantages. However, these methods face two major challenges: (1) if there is no common last-hop router between the target and landmarks, an IP subgraph cannot be constructed, leading to the failure of geolocating some target IPs; (2) the constructed IP subgraphs often contain noisy landmarks, resulting in degraded geolocation accuracy. To address these issues, we propose an IP geolocation method based on hop-constrained subgraphs and binary gates, named HB-Geo. Specifically, we design a novel IP subgraph construction method. The hop-constrained IP subgraphs do not rely on the common last-hop routers, allowing all reachable target IPs to be included in the subgraphs, thereby ensuring successful geolocation for every reachable target IP. Furthermore, we propose a binary gate-based IP geolocation framework. This framework can filter out noisy landmarks through binary gates, enabling target IPs to learn more valuable geographical supervision signals. Experiments on five real-world IPv4/IPv6 networks demonstrate that HB-Geo successfully geolocates all target IPs, while reducing the mean error by 0.36–40.17

The vastness and structural heterogeneity of the IPv6 address space fundamentally complicates the discovery of network periphery devices. Existing approaches consequently suffer from low probing precision and limited coverage, yielding an incomplete and biased view of the periphery. To address these challenges, we present 6Get, an Internet-scale active measurement method for IPv6 network periphery discovery. 6Get proposes a multi-source data mapping mechanism and an active prefix expansion mechanism to improve the precision and coverage of active-prefix generation. Using this active-prefix set, 6Get performs adaptive budget allocation and hierarchical subnet detection, substantially improving probe utilization during periphery device discovery. In real network deployments, 6Get identifies 13.76 million active /48 prefixes and 242.6 million periphery devices, achieving a probing precision of 10.41%. Compared to existing state-of-the-art methods, it achieves a 27.2× increase in coverage, a 7× improvement in probing precision, and a 2.37× increase in the number of discovered periphery devices. These results show that 6Get improves Internet-scale IPv6 network periphery discovery in both precision and coverage.

The exceptional performance of diffusion models establishes them as high-value intellectual property but exposes them to unauthorized replication. Existing protection methods either modify the model to embed watermarks, which impairs performance, or extract model fingerprints by manipulating the denoising process, rendering them incompatible with black-box APIs. In this paper, we propose TrajPrint, a completely lossless and training-free framework that verifies model copyright by extracting unique manifold fingerprints formed during deterministic generation. Specifically, we first utilize a watermarked image as an anchor and exactly trace the path back to its trajectory origin, effectively locking the model fingerprint mapped by this path. Subsequently, we implement a joint optimization strategy that employs dual-end anchoring to synthesize a specific fingerprint noise, which strictly adheres to the target manifold for robust watermark recovery. As input, it enables the protected target model to recover the watermarked image, while failing on non-target models. Finally, we achieved verification via atomic inference and statistical hypothesis testing. Extensive experiments demonstrate that TrajPrint achieves lossless verification in black-box API scenarios with superior robustness against model modifications.

Generative retrieval presents a promising approach to information retrieval, streamlining both indexing and retrieval processes through end-to-end optimization. This method typically involves assigning a unique identifier to each document, with the retrieval goal being the generation of the correct document identifier in response to a query. Although generative retrieval has demonstrated empirical success in various tasks, designing an effective document identifier remains a challenge. Previous studies have either depended excessively on one-to-one discrete identifiers, leading to increased retrieval latency and loss of semantics in documents or have used retrieval-agnostic dense document identifiers, which can hinder performance. To this end, we propose to integrate the benefits of generative retrieval and dense retrieval using an encoder-decoder-based pre-trained language model. Particularly, the decoder, i.e., the discrete identifier, functions as a coarse retriever, effectively reducing the retrieval space in an end-to-end manner. As a complement, the encoder, i.e., the dense vector, serves as a fine-grained retriever, efficiently and precisely ranking documents in a condensed space. Accordingly, we introduce a three-stage end-to-end learning framework that optimizes identifiers and vectors. Extensive experiments reveal that the proposed method exceeds the current models in terms of effectiveness and time efficiency, across both small and larger corpus sets.

Covert communication technology embeds important information into carrier files and transmits them through open channels to realize safe information sharing without causing external perception. Blockchain is a new technology that has gradually developed with the rise of digital cryptocurrencies. Blockchain has the characteristics of anonymity, anti-tampering, and decentralization, which can solve the problems of traditional covert communication, such as the anonymity of the communicating parties being challenging to guarantee, the integrity of secret information being vulnerable to the channel, and the communication process is overly dependent on the third parties. Based on this, this paper analyzes and researches the existing covert communication technology using digital cryptocurrency as the transactions. First, the consensus mechanisms and cryptographic algorithms used by the current mainstream digital cryptocurrencies are expounded. Next, the covert communication schemes are classified based on digital encryption currency, the advantages and disadvantages of these schemes are sorted out, and the performance analysis and comparison of these schemes are carried out. Finally, this paper implements three typical covert communication schemes, conducts multi-dimensional comparative experiments using different digital cryptocurrencies as the transportation medium, and further discusses the covert communication scheme that improves the cost transmission performance of transmission according to the practical requirements.

IP geolocation can establish mapping relationships from cyberspace to geographical space by estimating the geographical locations of IP addresses. It is widely needed in many location-based internet services, spanning from personalized recommendations to cybercrime tracking. Currently, the state-of-the-art (SOTA) IP geolocation methods are mainly based on Graph Neural Networks (GNNs). Their accuracy significantly outperforms traditional rule-based geolocation methods if there are sufficient landmarks near the target IPs. However, landmarks are usually hard to find except in major city centers. In networks with sparse landmarks, the accuracy of GNN-based methods is heavily affected, sometimes even worse than traditional methods. GNN-based methods also require much more hyperparameter tuning time than traditional methods. To better adapt to scenarios with limited landmarks and time constraints, we propose an efficient IP geolocation framework based on Graph Transformer, named GT-Geo. Different from GNN-based methods, in GT-Geo, Graph Transformer layers are designed to capture the critical geographical signals of long-range landmarks for target IPs in few-landmark networks. Meanwhile, to reduce the hyperparameter-tuning time for Graph learning-based IP geolocation methods, a graph entropy-based Node and Edge Embedding Selection (NEEDS) algorithm is designed to pre-calculate the ideal graph node and edge embedding dimensions according to the minimum entropy principle. Extensive experiments in five real-world IPv4/IPv6 networks demonstrate that: in few-landmark networks, GT-Geo significantly outperforms the SOTA methods in geolocation accuracy by an average of 4.05% in median error distance and 6.02% in mean error distance. Meanwhile, GT-Geo is also 1.59 to 4.53 times faster than the SOTA GNN-based methods.

The geographic location of social media users is crucial for understanding user behavior, optimizing advertising, and supporting location-based services such as emergency awareness and event monitoring services. However, existing Twitter user geolocation methods primarily focus on explicit social relationships between users while overlooking implicit location correlations, which affects the accuracy of user geolocation. To address this, this article proposes a Twitter user geolocation method (GeoICMF) based on implicit location correlations and multi-scale feature fusion. GeoICMF introduces a novel location association graph construction method to effectively capture implicit location correlations among users, an innovative multi-scale feature fusion model to dynamically fuse multi-scale features and generate richer user representations, and a pioneering geographic partitioning method to better adapt to user location distributions and enhance geolocation accuracy. Extensive experiments on three real-world datasets demonstrate that GeoICMF outperforms state-of-the-art baseline methods in Twitter user geolocation tasks, validating the effectiveness and superiority of the proposed method.

JPEG image forensics has become a critical area of research due to the increasing prevalence of compressed images and their vulnerability to tampering. This paper provides a comprehensive survey of three major tasks within JPEG image forensics: JPEG Compression Detection (JCD), JPEG Quantization Step Estimation (JQE), and the Application of JPEG Features (AJF). We explore the key features and techniques for detecting compression traces and estimating quantization parameters, emphasizing their applicability in both forensic analysis and anti-forensic strategies. JCD focuses on identifying JPEG compression artifacts, while JQE estimates the specific quantization steps used during compression. AJF, with its broad scope, supports diverse applications such as tampering detection, image recovery, and anti-forensics. We also examine the interrelationships between these tasks and discuss the challenges that hinder the field’s progress, including issues related to multi-compression scenarios, quality factor dependence, and the growing need for robust, generalizable methods. Finally, we propose benchmarks for evaluating the robustness of forensic models against adversarial attacks and multi-format compression schemes, while highlighting emerging trends and future research directions, including the integration of multi-modal information and advancements in deep learning-based solutions.

Robust image steganography in social networks presents significant challenges for steganalysis. Existing methods exploit low payload visibility and resistance to evade conventional passive steganalysis, leading to high false-positive and missed detection rates, which undermines reliable verification and poses a significant security threat to cyberspace. Active steganalysis is designed to eliminate hidden information through controlled modifications. However, existing methods usually encounter challenges such as a decline in the visual quality of processed images, excessive training costs of the models, and dependence on the image size. All these have limited the practical application of active steganalysis. To address this issue, we propose a block-based multi-objective analysis framework, which integrates dual-block constraints with multi-objective loss optimization. This framework achieves adaptive image partitioning by dividing the image into non-overlapping 2×2 pixel blocks. The partitioning simultaneously satisfies distortion boundary constraints and luminance preservation requirements. A multi-objective loss function is proposed to eliminate redundant information effectively while maintaining image visual clarity. The weights of loss function are set based on evolutionary optimization. Furthermore, the performance of the proposed method is enhanced by a linear-time complexity depth-first search algorithm, which minimizes the PSB loss under block-level constraints. The optimization process incorporates perturbation-based refinement to facilitate global convergence. Extensive experiments conducted on the BOSSBase-1.01 and CelebA datasets validate the effectiveness of the proposed framework in defending against robust image steganography prevalent in social networks. Evaluations across different payload settings indicate that the proposed method achieves an average PSNR improvement of approximately 34% over AO-Net across both BOSSBase-1.01 and CelebA datasets. Furthermore, the framework processes standard test images with a latency of 6.4ms, without requiring any preset image size.

Machine learning is a rapidly evolving field with applications in all aspects of human life. Utilizing the decentralized computing architecture can alleviate the high training and computational burden of central servers and improve service accuracy. However, the inherent properties of decentralized networks pose great challenges to communication security. It is urgent to design novel and appropriate security schemes, as malicious adversaries are curious about user private information, sensing data, and service demands. Furthermore, end devices always cooperate to accomplish the service targets, which means that group security schemes are needed to protect transmissions among them. In this paper, a certificateless-based group authentication and key agreement (CL-GAKA) scheme is proposed, named 2PCLGA, for distributed learning-based mobile edge computing (DL-MEC) networks. The proposed scheme establishes a session key among the end device group with the group leader MEC server based on the elliptic curve cryptography. Besides, the 2PCLGA scheme adopts dynamic pseudonym identity technology to realize the anonymity. The provable security analysis under the random oracle model, the formal analysis tool, and the informal analysis are adopted. The performance of 2PCLGA is also evaluated with the benchmarks, and the results show that the 2PCLGA scheme is greatly applicable to the resource-constrained circumstance.

Joint Photographic Experts Group (JPEG) quantization steps estimation aims to reveal the compressed history of the images, which can serve as an essential component and powerful technique to support forensics. Nowadays, various deep learning-based estimation methods have been proposed to achieve higher accuracy. However, due to supposing the ideal secure conditions of estimation, their robustness against deliberate attacks (especially adversarial attacks) has not been thoroughly studied, which poses a significant threat to their reliability. To address this issue, as the first attempt, we investigate the robustness of deep learning-based estimation methods against adversarial attacks, which can significantly deteriorate estimation accuracy without noticeable distortion. Specifically, we introduce a generation-based adversarial attack framework and propose two types of anti-forensic attacks, Direct Attack (DA) and Preemptive Attack (PA), to craft adversarial examples on double and single compressed images. To maximize the attack ability, we study the effect of regression and classification objectives on the adversarial property and design a joint loss function for stable and smooth optimization. Extensive experiments prove that the proposed DA and PA can achieve a high attack ability with low perturbation magnitude and satisfactory visual quality. More importantly, the generated adversarial examples present superior transferability across different estimation models and datasets, which proves the generality of the proposed method and also reveals the vulnerability of the existing deep learning-based estimation methods towards adversarial examples. Our code will be publicly available soon.

Diffusion Models (DMs) have achieved remarkable success in realistic voice cloning (VC), while they also increase the risk of malicious misuse. Existing proactive defenses designed for traditional VC models aim to disrupt the forgery process, but they have been proven incompatible with DMs due to the intricate generative mechanisms of diffusion. To bridge this gap, we introduce VoiceCloak, a multi-dimensional proactive defense framework with the goal of obfuscating speaker identity and degrading perceptual quality in potential unauthorized VC. To achieve these goals, we conduct a focused analysis to identify specific vulnerabilities within DMs, allowing VoiceCloak to disrupt the cloning process by introducing adversarial perturbations into the reference audio. Specifically, to obfuscate speaker identity, VoiceCloak first targets speaker identity by distorting representation learning embeddings to maximize identity variation, which is guided by auditory perception principles. Additionally, VoiceCloak disrupts crucial conditional guidance processes, particularly attention context, thereby preventing the alignment of vocal characteristics that are essential for achieving convincing cloning. Then, to address the second objective, VoiceCloak introduces score magnitude amplification to actively steer the reverse trajectory away from the generation of high-quality speech. Noise-guided semantic corruption is further employed to disrupt structural speech semantics captured by DMs, degrading output quality. Extensive experiments highlight VoiceCloak's outstanding defense success rate against unauthorized diffusion-based voice cloning. Additional audio samples of VoiceCloak are available in demo pages.