With the advent of intelligent technologies, miscellaneous data containing sensitive information are explosively generated and shared. Compressive sensing methods are naturally suitable for such scenarios due to their joint compression and encryption capabilities. However, data users of most existing compressive sensing methods need to reconstruct original images before use, which brings two disadvantages. First, indiscriminately requiring every data user to reconstruct images without considering their exact requirements is neither advisable nor efficient. Second, allowing all data users to reconstruct original images may cause private or confidential information exposure. To address these issues, in this paper, a novel image sharing method is proposed, which realizes efficient multilevel privacy preservation. Specifically, data owners compress the original images with designed measurement matrices through the proposed Tℓ1-B2DLDA algorithm, which outputs dimension-reduced data with the ability to simultaneously support the subsequent classification tasks for level I data users and reconstruction tasks for level II data users. Therefore, low level data users could achieve their goals without obtaining any private or confidential information in the original images. Experiments are conducted to verify the feasibility, performance and robustness of the proposed method. Furthermore, the security of the proposed method is analyzed both theoretically and practically. The source code of the proposed method is publicly available at https://github.com/xchuxiao23/TL1-B2DLDA.

We release a large-scale dataset that captures interactions between human users and CommentRobert, an LLM-based social media agent on Weibo. The dataset contains Weibo posts in which users actively mention the LLM agent account @CommentRobert, indicating that the users are interested in interacting with the platform-empowered LLM agent. The dataset contains 557,645 interactions from 304,400 unique users over 17 months. We detail our data collection methodology, user attributes, and content characteristics, underscoring the dataset's value in examining real-world human-LLM agent interactions. Our analysis offers insights into the demographic and behavioral traits of users interested in the selected LLM agent, interaction dynamics between humans and the agent, and linguistic patterns in comments. These interactions provide a unique lens through which to explore how humans perceive, trust, and communicate with LLMs. This dataset enables further research into modeling human intent understanding, improving LLM agent design, and studying the evolution of human-LLM agent relationships. Potential applications also include long-term user engagement prediction and AI-generated comment detection on social platforms. This constructed dataset is available at https://zenodo.org/records/16921462.

With the rapid growth of IoT technology, Visual IoT (VIoT) plays a key role in areas like security surveillance and intelligent transportation, where large volumes of sensitive data are collected and transmitted. Many applications face limitations in computing and storage, such as challenges in managing traffic data. Recently, Compressed Sensing (CS) theory has been applied to improve data acquisition and processing efficiency. However, existing CS-based privacy methods focus mainly on protecting single privacy zones through key-based control and lack support for hierarchical protection across multiple sensitive regions. To address this, we propose a multi-level privacy-preserving scheme for VIoT data using CS, offering full, partial, and no access levels to meet varying security needs. We also implement watermarks to prevent key-sharing attacks among partially authorized users. Experimental results demonstrate that the scheme ensures data security while minimizing time and space overhead, making it suitable for resource-limited VIoT scenarios.

Recent end-to-end image compressive sensing networks primarily use Convolutional Neural Networks (CNNs) and Transformers, each with distinct limitations: CNNs struggle with global feature capture, while Transformers lack local feature extraction. We propose a novel network, DCCT-Net, which combines Dynamic CNN (DCNN) and Transformer. This integration leverages DCNN’s local feature strengths and the Transformer’s global representation capabilities, resulting in superior image reconstruction quality. To further enhance the network’s performance, we propose a Feature Dynamic Augment Module (FDAM), which dynamically extracts features based on the saliency of segmented image regions, thereby amplifying the CNN’s local feature expression. Additionally, we design a Weighted Fusion Module (WFM), which optimizes the combination of local and global features extracted by the DCNN and Transformer, respectively. Extensive experiments demonstrate that our proposed DCCT-Net significantly outperforms most existing state-of-the-art methods in the field.

Breast diseases pose a significant threat to women’s health. Automatic lesion segmentation in breast ultrasound images (BUSI) plays a crucial role in fast diagnosis. While various enhanced U-Net-based models have achieved success in multi-scale feature analysis and handling blurred boundaries, two key challenges persist that could guide the improvement of BUSI segmentation networks: 1) significant fluctuations in pixel intensity distribution similarity between the lesion and surrounding tissues, and 2) inconsistent transmission of spatial detail due to multi-scale lesion sampling. These issues highlight the necessity of semantic elasticity understanding and consistency control. To this end, we propose ElaD-Net, an Elastic Semantic Decoupling Network for lesion segmentation in BUSI. This network uses the pre-trained EfficientNet-B2 for multi-scale encoding of BUSI. The decoding stage features two key modules: Elastic Semantic Decoupling (ESD) and Spatial Semantic Reconstruction (SSR). ESD learns and decouples multi-frequency semantics in multi-scale channels with a self-calibration mechanism, enabling dynamic adjustment of receptive depth to resist similarity fluctuations. SSR further optimizes ESD outputs via feature branching, compression, and excitation to ensure spatial semantic consistency, thereby separately reconstructing edge and body.

The rising popularity of large language models (LLMs) has raised concerns about potential abuse and harmful content. As a result, developing a highly generalizable and adaptable machine-generated text (MGT) detection system has become an urgent priority. Given that LLMs are most commonly misused in academic writing, this work investigates the generalization and adaptation capabilities of MGT detectors in three key aspects specific to academic writing: First, we construct MGT-Academic, a large-scale dataset comprising over 336M tokens and 749K samples. MGT-Academic focuses on academic writing, featuring human-written texts (HWTs) and MGTs across STEM, Humanities, and Social Sciences, paired with an extensible code framework for efficient benchmarking. Second, we benchmark the performance of various detectors for binary classification and text attribution tasks in both in-domain and cross-domain settings. This benchmark reveals the often-overlooked challenges of text attribution tasks. Third, we introduce a novel text attribution task in which models must adapt to new classes over time, with little or no access to prior training data, spanning both few-shot and many-shot scenarios. We implement a range of adaptation techniques to enhance performance across these settings. Our findings provide new insights into the generalization ability of MGT detectors and lay the foundation for building robust, adaptive detection systems. The code framework is available at https://github.com/Y-L-LIU/MGTBench-2.0.

Current vulnerability detection methods encounter challenges, such as inadequate feature representation, constrained feature extraction capabilities, and coarse-grained detection. To address these issues, we propose a fine-grained hybrid semantic vulnerability detection framework based on Transformer, named TFHSVul. Initially, the source code is transformed into sequential and graph-based representations to capture multilevel features, thereby solving the problem of insufficient information caused by a single intermediate representation. To enhance feature extraction capabilities, TFHSVul integrates multiscale fusion convolutional neural network, residual graph convolutional network, and pretrained language model into the core architecture, significantly boosting performance. We design a fine-grained detection method based on a self-attention mechanism, achieving statement-level detection to address the issue of coarse detection granularity. In comparison to existing baseline methods on public data sets, TFHSVul achieves a 0.58 improvement in F1 score at the function level compared to the best performing model. Moreover, it demonstrates a 10% enhancement in Top-10 accuracy at the statement-level detection compared to the best performing method.

The advancement of cloud computing and Internet of Things (IoT) has driven progress in image-based searchable encryption technology, which meets the escalating security demands of outsourced multimedia data in IoT scenarios. However, existing encrypted image retrieval schemes still face critical challenges, such as lack of cross-domain support, low retrieval efficiency and accuracy, and absence of reliable verifiability. To address these issues, this paper proposes a verifiable and privacy-preserving cross-domain image retrieval scheme (VPCDIR) in IoT. In our scheme, the re-encryption and key transformation technologies are implemented to achieve the availability of cross-domain image retrieval, and the learning with errors (LWE)-based enhanced secure k-nearest neighbor (kNN) algorithm is used to preserve the privacy of image features. Furthermore, the hybrid index mechanism that combines clustering and locality-sensitive hashing (LSH) is designed to enhance retrieval efficiency and accuracy, and the Merkle hash tree (MHT) with the short signature realizes reliable authenticity verification of the retrieval result. Finally, formal security analysis confirms the security of our scheme. Extensive experiments on the real dataset demonstrate the efficiency and practicability of VPCDIR for cross-domain image retrieval in IoT.

Currently, the popularity of social networks has brought us rich and colorful displays and pleasant experiences. However, social networking is also a double-edged sword. While pleasing us, it also raises the issue of privacy disclosure of social media images. How to protect the privacy of media images has become a significant issue for social networks. When we post large quantities of images on social networks to share our daily lives, sometimes we only want to share them with specific friends or want friends with different permissions to see different image content, which involves hierarchical privacy protection. In particular, when the image to be shared contains multiple privacy-sensitive areas of different levels, and we only want to protect the privacy-sensitive areas rather than the whole image, how to protect each privacy-sensitive area is a major problem. Aiming at the above problems, a multilevel privacy protection scheme for image sharing in social networks based on 2-D compressive sensing is proposed. This scheme has the advantages of compressive sampling, privacy protection and controllable access. In addition, we propose a 2-D projected gradient algorithm with accompanying privacy region decryption (2DPG-APRD) for implementing the hierarchical privacy-preserving function of the proposed scheme. Experimental results show that our scheme has multilevel reconstruction quality, high-security intensity for different authorized users, and can well protect the privacy information of images. Therefore, the proposed scheme suits for many practical multilevel encryption situations.

Privacy preservation and low-cost data processing have become two critical issues in the era of Internet of Things (IoT) due to the widespread deployment of lightweight smart surveillance and sensors. In this article, we propose a multitiered reversible data privacy preservation system based on compressive sensing (CS) and watermarking. The system anonymizes the region of interest (ROI) using an obfuscation matrix while compressing and encrypting the entire document. CS provides the first-tier encryption for data documents, and the obfuscation matrix provides the second-tier encryption for sensitive parts of data documents. The system offers a multitiered privacy protection scheme where restricted-authorized users can only access nonsensitive data while fully authorized users can access the entire document. To implement the reversible elimination of the obfuscation matrix, two watermark embedding methodologies are proposed in the CS domain in order. In both methods, the watermark generated by the obfuscation matrix is embedded in the encrypted CS measurement values, with the first methodology concentrating on the optimal data reconstruction quality and the second methodology working to balance storage space and data restoration quality. Extensive experimental results indicate the superiority of the proposed methodologies over other conventional reversible data privacy preservation schemes.

With the rapid advancement of smart cities, the utilization of digital images has become widespread, particularly in services such as urban traffic management and public space security surveillance. However, the acquisition, transmission and sharing of digital images inevitably raise concerns about privacy disclosure. To address this challenge, we propose a lightweight image encryption scheme based on data hiding and compressive sensing (CS). Specifically, during the CS sampling and compression stages, we employ data-hiding techniques to embed information from the confusion matrix and coordinates of sensitive regions into CS ciphertext, ensuring the secure transmission of encryption keys for sensitive regions. Additionally, our solution can provide personalized access control mechanisms based on the permission levels of different authorized users and offer customized image recovery quality according to their specific requirements. This effectively addresses the potential privacy leakage risks associated with cross-departmental image sharing, ensuring the security of data transmission and sharing processes. Under consistent experimental conditions, our proposed solution demonstrates a minimum 1.5% improvement in reconstruction quality compared to existing methods. In the security analysis, we further demonstrate that the proposed scheme provides differential reconstruction quality and high-security strength for staff members with different permissions. We believe the proposed solution can suit many practical applications in smart cities.

Deep network-based image compressed sensing (CS) methods have attracted much attention in recent years due to their low reconstruction complexity and high reconstruction quality. However, the existing methods usually use one or multiple convolution layer(s) consisting of convolutional kernels with the same size to extract image features in image sampling, which results in incomplete feature extraction. Besides, the existing models usually focus on the extraction of deep features in image reconstruction, while ignoring the influence of shallow features. To overcome these issues, this paper proposes a multi-scale characteristic residual learning network (dubbed MSCRLNet) for image CS. In this network, convolutional kernels with different sizes are used to capture multi-level spatial features in image sampling, and a multi-scale residual network with channel attention is used to speed up network convergence in image reconstruction. Experiments show that the proposed MSCRLNet outperforms many existing state-of-the-art methods.

Nowadays, the requirements for image data sharing among participant nodes of the Internet of Things (IoT) are constantly emerging. And thanks to the popularity of digital cameras and the development of digital photography technologies, the processes of data sharing commonly concern substantial amounts of data that may contain critical or private information. So, how to design a secure and efficient secret image sharing (SIS) method suitable for IoT apparatuses is attracting ever-increasing attention. With the aim of simultaneously realizing SIS, image data compression, and critical information or privacy protection, this article proposes a multigroup SIS method based on the model of compressed sensing (CS) and chaos theory. In the proposed method, participants of SIS are classified into two groups with different authorization levels. Solely members of full-authorized groups could reconstruct secret images with visible critical sectors. Members of restricted-authorized groups, however, could merely obtain reconstruction results with critical sectors concealed. The CS model is introduced to the proposed method to accomplish image data compression and applications of the chaos theory contribute to the secure data transmission. Experimental simulations and theoretical analyses are performed to discuss the feasibility, flexibility, and security of the proposed method.

The Internet of Things (IoT) is a large-scale network of various sensing devices connected via the Internet to achieve intelligent functions. Efficient secure data transmission is a significant guarantee for specific functions of IoT. In recent years, compressive sensing (CS) has been applied to the field of the IoT and a substantial number of CS-based IoT data transmission schemes have been proposed. While existing IoT data transmission schemes based on CS could mostly accomplish signal sampling, data compression, and encrypted transmission, the goal of privacy protection has rarely been achieved. Based on chaos theory and semi-tensor product CS (STP-CS), this article proposes two novel secure data transmission methods for different scenarios of IoT: 1) multiple concealing method (MC method) and 2) precise concealing method (PC method). Both methods can provide different levels of reconstruction results for different receivers with various authorization levels. The feasibility, storage requirements, robustness, and security of the proposed methods are theoretically analyzed and experimentally simulated. The results show that the proposed methods ensure both the stability and security of data transmission, save storage space of sensors, and flexibly protect the privacy of the content of data transmitted.

Compressed sensing is an emerging theory of signal processing and it has wide applications in many frontier fields. The construction of the measurement matrices is still a central problem in compressed sensing. In this paper, two types of deterministic constructions of binary measurement matrices are presented via unitary geometry. Then, the lower bounds of the spark of unitary geometry measurement matrices are theoretically analyzed, and an asymptotic comparison between unitary geometry measurement matrices and projective geometry measurement matrices is given via the worst-case recovery capability. After that, a clipping-embedding operation is proposed for binary matrices to generate measurement matrices with more sizes, which can strongly extend the applicability of the deterministic binary matrices in practice. Finally, simulation results demonstrate that the performance of our measurement matrices is comparable to, sometimes even better than, that of the corresponding Gaussian random matrices under OMP and BP.