Latent diffusion models (LDMs) have become the dominant paradigm for AI image synthesis, while also posing unprecedented challenges to AI image attribution. To mitigate these risks, digital watermarking has emerged as a crucial technique for embedding traceable identity information into generated content. However, existing watermarking techniques for LDMs often suffer from degraded image quality, limited robustness, and substantial computational and storage overhead. To address these limitations, we propose Feature Watermarking, a novel watermarking paradigm in which pre-trained, randomly sampled feature vectors are leveraged to enable robust attribution capabilities. Subsequently, we employ a dedicated encoder-decoder architecture to generate initial noise and reconstruct features, respectively. Image attribution is performed by measuring the similarity between the original and reconstructed features. Experimental results demonstrate that the proposed method matches or outperforms existing state-of-the-art approaches in image quality and attribution accuracy, while significantly reducing resource consumption.

Reversible data hiding (RDH) for JPEG images remains relatively underexplored, with key challenges lying in coefficient selection and modification strategies. Existing methods select coefficients for embedding through block-wise or frequency-band-based operations, resulting in coarse-grained decisions that constrain embedding performance. In this paper, a novel RDH scheme for JPEG images based on gap-driven histograms generation with coefficient-wise selection is proposed. First, a multi-metric weighted complexity and coefficient-wise selection approach is proposed, integrating four local feature criteria to assess each coefficient individually, enabling more precise per-coefficient selection. Then, a gap-driven adaptive multi-histogram generation strategy is introduced, leveraging gap pairs to minimize shifting distortion by segmenting histograms via bisection and avoiding modifications to high-magnitude coefficients. Experimental results confirm that the proposed method achieves improved visual quality and more efficient file size control compared to existing state-of-the-art approaches.

With the advent of sixth-generation (6G) technologies and growing communication demands, Low Earth Orbit (LEO) satellite networks have become essential in modern communications. However, due to the dynamic topology and complex network state of LEO environments, existing routing methods often fail to make effective decisions, limiting transmission performance. This paper proposes a LEO satellite routing method based on incremental evolutionary graph reinforcement learning (IEGRL). To address network state perception challenges, we introduce a topological learning model using deep graph attention (DGA), which captures complex inter-satellite connectivity and resource states. Additionally, by integrating incremental evolution strategies (IES) into deep reinforcement learning (DRL), we replace sequential interactive proximal policy optimization (PPO) with global parallel ES, achieving efficient routing convergence in the highly dynamic LEO environment. Experimental results demonstrate that our IEGRL approach enhances LEO network load balancing by reducing end-to-end (E2E) network latency, decreasing packet loss, and improving throughput compared with the benchmark approaches.

Low Earth orbit (LEO) satellite networks are an important component of future 6G. However, due to the unique characteristics of the space environment—such as the complexity in modeling network states and the rapid dynamics of the network topology—existing routing methods often struggle to make appropriate routing decisions in the LEO satellite network context, which significantly limits network transmission performance. In this paper, we propose a dynamic satellite routing method based on deep graph attention and incremental evolution strategy (DGA-IES). Firstly, to address the challenge of accurately perceiving satellite network information, we introduce a topological perception learning model based on deep graph attention. By combining an enhanced message passing process with a self-attention mechanism, this model effectively captures complex features of the LEO network state, including inter-satellite connectivity relationships, as well as the resource states of satellites and links. Secondly, to tackle the problem of inefficient routing re-convergence in rapidly changing topologies, this paper integrates evolution strategies (ES) into deep reinforcement learning (DRL) approaches. We use the global parallel processing capabilities of ES to replace the sequential interactive proximal policy optimization (PPO) strategy in existing DRL. Moreover, we design an incremental evolutionary process based on satellite motion patterns, facilitating efficient routing convergence in highly dynamic satellite environments. Experimental results demonstrate that our DGA-IES approach enhances LEO network load balancing by reducing end-to-end (E2E) network latency by 10.3% 58.1%, decreasing packet loss by 3.8% 20.0%, and improving throughput by 11.1% 57.0% compared with the benchmark approaches.

Satellite edge computing (SEC) extends the capabilities of edge computing technology to satellite networks, facilitating rapid local processing of global task requirements. Deep reinforcement learning (DRL) has emerged as a promising approach for SEC scenarios due to its inherent dynamic adaptability, complex state modeling capability, and long-term optimization potential. However, existing DRL-based computing offloading techniques continue to encounter challenges including low sample efficiency, poor decision quality, and insufficient long-term stability, which constrain their performance in real satellite network environments. To address these challenges, this study proposes a diffusion and DRL-based approach for computation offloading in SEC networks called the generative artificial intelligence-DRL (GenAI-DRL). First, by implementing the cooperative computing model of the multi-SEC, this study comprehensively considers the heterogeneous computing and communication capabilities of satellite nodes, diversity of task types, and dynamic distribution of resources in an offloading strategy, thereby ensuring long-term system sustainability under dynamic resource constraints and provides a solid foundation for computation offloading in satellite networks with time-varying resource. Second, we integrate generative diffusion modeling (GDM) into the DRL framework to enhance policy generation by producing contextually relevant and high-quality action samples. This not only reduces the dependence on large-scale training data but also improves decision precision and generalization in complex, high-dimensional environments. Finally, a Lyapunov optimization framework is introduced to transform the offloading problem into an online per-slot optimization process, thereby ensuring the long-term stability of the SEC system under dynamic and unpredictable task arrivals and environmental conditions. The experimental results demonstrate that the method proposed offers significant advantages over the existing approaches in reducing task latency and enhancing system stability.

Pixel value ordering (PVO) is an efficient method for implementing reversible data hiding, which can achieve embedding based on overlapping pixel blocks when combined with the flexible patch moving (FPM) mode, especially the two-dimensional (2D) FPM mode. However, the existing 2D FPM mode, whose pairing way of prediction error is not conducive to generating more pixel blocks available for embedding, and whose movement rules are too inefficient to fully exploit the potential of the PVO, results in wasting many available blocks. Therefore, in this paper, a two-stage embedding mechanism is proposed for the 2D FPM mode, in which the combination of prediction errors is adjusted to improve the possibility of generating available blocks and the two-stage embedding doubles the number of pixel blocks available for embedding. Furthermore, an FPM mode selection is proposed, where four novel 2D FPM modes are designed to efficiently exploit the potential of the PVO according to the different directional gradients. Lastly, a set of efficient 2D mappings is well-designed for multiple histograms to achieve lower embedding distortion. The extensive experimental results show that the proposed method outperforms other state-of-the-art methods in terms of embedding capacity and image fidelity. The average peak signal-to-noise ratio for the Kodak image dataset is as high as 63.62 dB after embedding 10,000 bits.

In recent years, reversible data hiding in encrypted domain (RDH-ED) has garnered considerable interest among researchers, resulting in the development of high-performance methods based on various carriers. However, the challenge of enhancing the data embedding capacity while ensuring reversibility becomes increasingly pronounced when the carrier is a three-dimensional (3D) model. In this paper, a high capacity RDH-ED method based on dynamic prediction and virtual connection for 3D models is proposed. Unlike existing methods that partition the vertices in the model into embeddable and prediction sets, where each vertex can only serve one function, the proposed dynamic prediction mechanism constructs a data embedding order set by leveraging the connectivity relationships between vertices. This allows each vertex within the set to both embed data and provide predictions, significantly increasing the proportion of embeddable vertices. Moreover, the proposed method is the first work to consider independent vertices within the model and integrates a novel virtual connection approach with the dynamic prediction process, enabling all independent vertices to participate in data embedding and prediction, thereby further enhancing the data embedding capacity. Experimental results demonstrated that the proposed method significantly outperforms other state-of-the-art methods in terms of data embedding capacity while ensuring reversibility.

Federated learning (FL) has attracted widespread attention because it supports the joint training of models by multiple participants without moving private dataset. However, there are still many security issues in FL that deserve discussion. In this paper, we consider three major issues: 1) how to ensure that the training process can be publicly audited by any third party; 2) how to avoid the influence of malicious participants on training; 3) how to ensure that private gradients and models are not leaked to third parties. Many solutions have been proposed to address these issues, while solving the above three problems simultaneously is seldom considered. In this paper, we propose a publicly auditable and privacy-preserving federated learning scheme that is resistant to malicious participants uploading gradients with wrong directions and enables anyone to audit and verify the correctness of the training process. In particular, we design a robust aggregation algorithm capable of detecting gradients with wrong directions from malicious participants. Then, we design a random vector generation algorithm and combine it with zero sharing and blockchain technologies to make the joint training process publicly auditable, meaning anyone can verify the correctness of the training. Finally, we conduct a series of experiments, and the experimental results show that the model generated by the protocol is comparable in accuracy to the original FL approach while keeping security advantages.

Currently, deep learning-based video inpainting algorithms can fill in a specified video region with visually plausible content, usually leaving imperceptible traces. Since deep video inpainting methods can be used to maliciously manipulate video content, there is an urgent need for an effective method to detect and localize deep video inpainting. In this paper, we propose a dual-stream video inpainting detection network, which includes a ConvNeXt dual-stream encoder and a multi-scale feature cross-fusion decoder. To further explore the spatial and temporal traces left by deep inpainting, we extract motion residuals and enhance them using 3D convolution and SRM filtering. Furthermore, we extract filtered residuals using LoG and Laplacian filtering. These residuals are then entered into ConvNeXt, thereby learning discriminative inpainting features. To enhance detection accuracy, we design a top-down pyramid decoder that aims at deep fusion of multi-dimensional multi-scale features to fully exploit the information of different dimensions and levels in detail. We created two datasets containing state-of-the-art video inpainting algorithms and conducted various experiments to evaluate our approach. The experimental results demonstrate that our approach outperforms existing methods and attains a competitive performance despite encountering unseen inpainting algorithms.

Glyph perturbation adjusts the characters’ structures and strokes to make the original characters change subtly, which cannot be detected by the naked eye. These generated variants with different glyph perturbation can represent different status of secret messages, which can be used to embed information in Chinese text documents. However, Chinese characters have characteristics in large numbers, complex structures, and diverse fonts, which limit the generation of glyph perturbation and make the design of Chinese characters time-consuming and laborious. Many font style transfer methods for Chinese characters have been proposed to improve the efficiency of Chinese character generation based on deep learning. At present, there are few studies on efficient font style transfer for glyph perturbation of Chinese characters. In this paper, a stylized glyph perturbation method based on style extractor and attention augmented convolution is proposed. It adopts a multi-head attention mechanism to enhance convolution in the font transfer, which concatenates the convolution feature maps and the self-attention activation maps to weaken the limitations of ordinary convolution in processing images. The extracted style features are sent into the decoder of the font transfer network so as to improve the stylized ability. Particularly, the impact of style extractor and attention augmented convolution on the glyph perturbation generation is addressed. The extraction accuracy and embedding capacity are tested in our experiments. The embedding capacity of secret message can achieve around 1.8 bit/character.

With the increasing popularity of Online Social Networks (OSNs), covert communication is rapidly shifting from lossless channels like email to lossy channels, specifically social networks. In response to this trend, robust adaptive steganography has emerged as a powerful technique for concealing information in lossy transport channels. Previous approaches have aimed to address the challenge of JPEG image compression during transmission by utilizing static compression-resistant domains, Syndrome-Trellis Codes (STC), and Error Correction Codes (ECC). However, reliance on a significant number of ECC check codes to ensure robustness could inadvertently affect security. In response to this challenge, we introduce the “Adaptive STC-ECC” strategy, which enhances security by minimizing the number of check codes without compromising robustness. We further improve the robustness by simulating the embedding process and strategically placing the wet point in unstable cover elements. Furthermore, we exploit the residual information between the pre-cover and cover images to adjust the distortion and accurately determine the direction of the dither modulation, thus improving the overall security. Extensive experiments have been conducted to evaluate the performance of our proposed approach, and the results demonstrate its superior robustness and security compared to existing state-of-the-art approaches.

Recently, reversible data hiding in encrypted images (RDHEI) has received widespread attention from researchers. To embed high payload into encrypted images while maintaining sufficient security, a novel RDHEI algorithm in combination with consecutive zero-valued high bit-planes compression, bit-plane swapping as well as block rearrangement is proposed in this article. The proposed method is the first work to compress global zero-valued high bit-planes in a block-wise manner and adaptively allocate different Huffman indicators based on the occurrence frequency of zero-valued bit-planes so that a higher embedded payload is greatly provided. Unlike existing RDHEI methods embedded with unencrypted auxiliary information, resulting in low security, the bit-plane swapping and block rearrangement are subtly designed to cluster together all embeddable bit-planes, which enables most auxiliary information to be encrypted, largely enhancing the security and facilitating data embedding and data extraction. The experiment results demonstrate that the proposed method outperforms some state-of-the-art RDHEI methods in terms of security and payload. The average payload of the proposed method for two publicly-used datasets including BOSSbase and BOWS-2, are 3.793 bpp and 3.705 bpp, respectively.

Currently, health management driven by intelligent means is a general demand of social systems. Although a number of researchers have paid attention to such areas, they have primarily focused on improving the performance of intelligent algorithms. Such intelligent algorithms are mostly based on the central computing mode, where all the user data are aggregated together in a central cloud to implement computing tasks. This poses a great threat to personal privacy due to exposure to the outside world. To address this challenge, this work uses a federated learning mechanism and proposes a privacy-preserving social computing framework for health management. User data are deposited in different user terminals to prevent exposure. A group of parameters are pretrained for each terminal in an iteration and are then transferred to the center cloud for updating. After multiple rounds of interactive training between the center cloud and the terminals, a recognition model finishes training for each terminal without direct access to data from other sources. Finally, this work also conducts experiments on a real-world dataset to assess the overall performance of the proposed approach.

With the development of video and image processing technology, the field of video tampering forensics is facing enormous challenges. Specifically, as the fundamental basis of judicial forensics, passive forensics for object removal video forgery is particularly essential. To extract tampering traces in video more sufficiently, the author proposed a spatiotemporal trident network based on the spatial rich model (SRM) and 3D convolution (C3D), which provides three branches and can theoretically improve the detection and localization accuracy of tampered regions. Based on the spatiotemporal trident network, a temporal detector and a spatial locator were designed to detect and locate the tampered regions in the temporal and spatial domains of videos. For the temporal detector, 3D CNNs were employed in three branches as the encoders and a bidirectional long short-term memory (BiLSTM) as the decoder. For the spatial locator, a backbone network named C3D-ResNet12 was designed as the encoder of the three branches, and the region proposal networks (RPNs) were employed as the decoders in three branches. In addition, we optimized the loss functions of the above two algorithms based on focal loss and GIoU loss. The experimental results revealed the effectiveness of spatiotemporal detection and localization algorithms: for temporal forgery detection, the accuracy of the frame classification increased to 99+%; for spatial forgery localization, the successful localization rate of the tampered regions in forged frames reached 96+%, and the mean intersection over union of the located tampered regions and the real tampered regions reached 62+%.

Traffic flow prediction is an important research issue to avoid traffic congestion in transportation systems. Traffic congestion avoiding can be achieved by knowing traffic flow and then conducting transportation planning. Achieving traffic flow prediction is challenging as the prediction is affected by many complex factors such as inter-region traffic, vehicles' relations, and sudden events. However, as the mobile data of vehicles has been widely collected by sensor-embedded devices in transportation systems, it is possible to predict the traffic flow by analysing mobile data. This study proposes a deep learning based prediction algorithm, DeepTFP, to collectively predict the traffic flow on each and every traffic road of a city. This algorithm uses three deep residual neural networks to model temporal closeness, period, and trend properties of traffic flow. Each residual neural network consists of a branch of residual convolutional units. DeepTFP aggregates the outputs of the three residual neural networks to optimize the parameters of a time series prediction model. Contrast experiments on mobile time series data from the transportation system of England demonstrate that the proposed DeepTFP outperforms the Long Short-Term Memory (LSTM) architecture based method in prediction accuracy.