Table of Content

30 May 2025, Volume 43 Issue 3

Previous Issue   

Digital Media Forensics and Security

AIGC Users Traceability Technology Based on Text Watermarking

SONG Yimin, LIU Gongshen

2025, 43(3):  361-369.  doi:10.3969/j.issn.0255-8297.2025.03.001

Asbtract ( 23 )   PDF (563KB) ( 8 )  

References | Related Articles | Metrics

This study addresses the limitations of text watermarking technology in the Chinese language context, and proposes both modified watermarking and generative watermarking schemes for implementation in English and Chinese. Using the Bert model for English and the WoBert model for Chinese, this study designs a portable word substitution watermarking module, which embeds watermarking information by replacing the specified lexical elements in the source text. For generative watermarking, this study adopts the adversarial generative text watermarking model with targeted modifications and migrations on the Chinese corpus, ensuring compatibility with Chinese semantic structures and linguistic conventions of Chinese text. Experiments are conducted using a human-ChatGPT comparison corpus in both Chinese and English. The effectiveness of the proposed watermarking schemes is evaluated based on text watermarking evaluation metrics in terms of both accuracy and semantics. Results demonstrate the proposed methods’ effectiveness in enhancing watermark robustness and traceability in multilingual text.

Robust Watermarking Algorithm Guided by Invisibility Under Mesh Spectral Coefficients

WU Xiao, HUANG Ying, SONG Chunhua, GUAN Hu, NIU Baoning

2025, 43(3):  370-386.  doi:10.3969/j.issn.0255-8297.2025.03.002

Asbtract ( 32 )   PDF (4696KB) ( 7 )  

References | Related Articles | Metrics

In the field of copyright protection, mesh watermarking is a key technology for protecting 3D models from malicious attacks during transmission. However, existing watermarking algorithms for 3D meshes often fail to effectively connect the spatial and frequency domains and the setting rule of watermark embedding strength is not clear enough. To solve this problem, we propose a robust mesh watermarking algorithm guided by invisibility (RWGI). Taking the Laplacian transform of a mesh as an example, we establish, for the first time during the watermark embedding process, a model that characterizes the relationship between the spatial-domain invisibility evaluation index and the spectraldomain embedding strength. In this way, the watermark embedding can be adaptively guided by the invisibility index, ensuring the quantification of the watermark embedding strength on demand. Additionally, we design a spectral coefficient optimization strategy based on absolute value, along with a simple yet efficient watermarking segmentation algorithm. Experimental results show that the proposed algorithm achieves strong invisibility, robustness against common attacks, and adaptive control of watermark embedding intensity with respect to model scale.

Multi-pass PVO Reversible Data Hiding Based on Spatial Location Optimization

ZHOU Tongyang, TANG Xin, XU Yichen, SONG Chuqiao, BAI Jing, ZOU Yifei

2025, 43(3):  387-402.  doi:10.3969/j.issn.0255-8297.2025.03.003

Asbtract ( 27 )   PDF (2111KB) ( 9 )  

References | Related Articles | Metrics

In this paper, an optimized multi-pass pixel value ordering (PVO) algorithm based on spatial position is proposed. The target image is first divided into multiple subblocks of 3 ￡ 3 pixels. By fully integrating the spatial positions of each pixel, all pixels are partitioned into two groups for two rounds of embedding. In the first round, the edge pixels of the pixel block are predicted and embedded based on their spatial positions. In the second round, some pixel values are regenerated first, and then the middle pixels are sorted and embedded according to the multi-pass pixel value ordering algorithm, thereby fully exploiting the correlation between spatial position and pixel value size. Experimental results show that the regeneration of pixel values improves the embedding efficiency in the multi-pass sorting algorithm. While ensuring reversibility, the proposed algorithm not only enhances the embedding capacity, but also achieves excellent peak signal to noise ratio performance, ensuring the image quality after embedding and meeting practical application requirements.

Robust Text Steganography Based on Cross-Modal Learning

MA Ting, TAN Yun, QIN Jiaohua, XIANG Xuyu

2025, 43(3):  403-414.  doi:10.3969/j.issn.0255-8297.2025.03.004

Asbtract ( 18 )   PDF (1545KB) ( 5 )  

References | Related Articles | Metrics

This paper proposes a robust text steganography method based on cross-modal learning, embedding secret information by generating sentences consistent with the semantics of image. To improve text generation quality, both semantic and regional features of the image are integrated. Moreover, a random word deletion attack layer is designed during training to further enhance the robustness of the steganographic texts. Experiments evaluate the model’s robustness against both text and image attack. The results demonstrate that the proposed method achieves superior text generation quality and robustness, effectively improving the cognitive concealment of steganographic text.

Computer Science and Applications

A Path Planning Algorithm for Mobile Robots Based on an Improved Deep Deterministic Policy Gradient

ZHANG Qingling, NI Cui, WANG Peng, GONG Hui

2025, 43(3):  415-436.  doi:10.3969/j.issn.0255-8297.2025.03.005

Asbtract ( 32 )   PDF (2864KB) ( 8 )  

References | Related Articles | Metrics

The deep deterministic policy gradient (DDPG) algorithm utilizes an actorcritic framework to ensure smooth motion of mobile robots. However, the critic network tends to fail to distinguish effectively between different states and actions, leading to inaccurate Q-value estimates. Additionally, the sparse reward function in DDPG slows down convergence during model training, while the random uniform sampling approach utilizes the sample data inefficiently. To address these challenges, this paper introduces dueling networks to improve Q-value estimation accuracy within DDPG framework. The reward function is optimized to guide the mobile robot toward more efficient and effective movement. Furthermore, the single experience replay buffer is split into two parts, and a dynamic adaptive sampling mechanism is adopted to enhance replay efficiency. Finally, the proposed algorithm is evaluated in a simulation environment built with the robot operating system (ROS) system and Gazebo platform. Experimental results demonstrate that compared to the standard DDPG algorithm, the proposed approach reduces training time by 17.8%, improves convergence speed by 57.46%, and increases the success rate by 3%. Moreover, the proposed method outperforms other algorithms in terms of stability during model training, significantly improving the efficiency and success rate of mobile robot path planning.

Semi-supervised Encrypted Traffic Classification Model Based on Contrastive Learning

JIN Yanliang, FANG Jie, GAO Yuan, ZHOU Jiahao

2025, 43(3):  437-450.  doi:10.3969/j.issn.0255-8297.2025.03.006

Asbtract ( 19 )   PDF (1092KB) ( 10 )  

References | Related Articles | Metrics

To address the performance degradation of most encrypted traffic classification (ETC) models due to scarce labeled data, this paper proposes a semi-supervised encrypted traffic classification model based on contrastive learning (SSETC-CL). By comparing the similarities and differences between samples, SSETC-CL is capable of learning useful representations from large amounts of unlabeled data, thereby obtaining a versatile and effective feature encoding network, and reducing dependence on labeled data for downstream tasks. The performance of SSETC-CL is evaluated on the public dataset ISCXVPN2016 as well as two self-collected datasets. Compared to other baseline models, SSETC-CL achieved a maximum accuracy improvement of 8.92% on the specified task, showing its superior performance. Experimental results clearly demonstrate that SSETC-CL not only achieves high accuracy on traffic seen during pretraining but also exhibits the ability to transfer the knowledge gained from pretraining to unknown traffic.

Frequency-Domain Multi-feature Fusion for Deepfake Video Detection Based on Key Frames

WANG Jinwei, ZHANG Meigui, ZHANG Jiawei, LUO Xiangyang, MA Bin

2025, 43(3):  451-462.  doi:10.3969/j.issn.0255-8297.2025.03.007

Asbtract ( 22 )   PDF (1379KB) ( 7 )  

References | Related Articles | Metrics

To avoid data redundancy and save computing resources, most of the existing Deepfake video detection methods select multiple frames or partial segments of videos as the detection objects. However, this selection strategy compromises the representation ability of the detection objects and limits the performance. Moreover, while the existing algorithms perform well on individual datasets, their performance degrade seriously when detecting across datasets, highlighting the need for improved generalization. To address these challenges, we propose a frequency domain multi-feature fusion algorithm for Deepfake video detection based on key frames. The mean square error in frequency domain is used to extract the key frames as the detection objects. Then the artifact features of the main frame and temporal inconsistency features between the key frames are learned in frequency domain. These features are fused and passed through a fully connected layer to obtain the final detection results. Experimental results show that our algorithm achieves superior performance in cross-dataset detection compared to existing methods, showcasing strong generalization capabilities.

Verifiable Privacy-Preserving Personalized Federated Learning

YANG Zhe, REN Yanli, ZHONG Yuege, FENG Guorui

2025, 43(3):  463-474.  doi:10.3969/j.issn.0255-8297.2025.03.008

Asbtract ( 17 )   PDF (631KB) ( 6 )  

References | Related Articles | Metrics

To address privacy leakage and performance degradation in federated learning with heterogeneous data, we propose a verifiable privacy-preserving personalized federated learning scheme. In the scheme, the privacy of users is guaranteed through homomorphic encryption. Personalized model customization is enabled by calculating similarities over ciphertexts. Based on the ring learning with errors problem, users can verify the correctness of personalized updates. Theoretical and experimental analysis shows that the proposed scheme effectively preserves user privacy, ensuring that neither the server nor the user can access others’ local or personalized updates. Furthermore, the additional computational and communication overhead incurred by privacy preservation remains within acceptable limits. Experimental results on two public datasets show that the proposed scheme achieves higher accuracy than federated averaging and other personalized schemes under both independently and non-independently distributed data settings.

Coupling Analysis of Urbanization and Ecological Environment in Beijing-Tianjin-Hebei Urban Agglomeration from 2000 to 2020

ZHANG Yongbin, LI Chunyu, LIU Mingyue, MAN Weidong, SONG Tanglei, LIU Yahui

2025, 43(3):  475-490.  doi:10.3969/j.issn.0255-8297.2025.03.009

Asbtract ( 20 )   PDF (3632KB) ( 5 )  

References | Related Articles | Metrics

Taking the Beijing-Tianjin-Hebei urban agglomeration as the research object, this study constructed a comprehensive light index representing urbanization by integrating night light data, and a remote sensing ecological index representing ecological environment quality by integrating daylight remote sensing MODIS data. The spatial-temporal evolution of urbanization intensity and ecological environment in the Beijing-Tianjin-Hebei urban agglomeration from 2000 to 2020 was studied. Combined with the coupling coordination degree model, the synergistic development relationship between the two was analyzed to provide theoretical support and scientific decision-making for the harmonious coexistence of urbanization and ecological environment in the region. The results show that: 1) Urbanization development exhibited spatial differences across cities, showing a distinct spatial pattern of high levels in the central south and low levels in the northwest. From 2000 to 2010, urbanization developed slowly, while 2010 to 2020 marked a period of rapid development. 2) From 2000 to 2020, the remote sensing ecological index (RSEI) of the Beijing-Tianjin-Hebei urban agglomeration fluctuated greatly. However, on the whole, the RSEI increased from 0.48 in 2000 to 0.58 in 2020, indicating an improvement in the ecological environment quality. In terms of the change in ecological grade, there was a transformation mainly from the poor and relatively poor ecological grades to the medium ecological grade, from the medium ecological grade to the good ecological grade, and from the good ecological grade to the excellent ecological grade. 3) Over the past two decades, the coupling coordination type of the Beijing-Tianjin-Hebei urban agglomeration has transitioned from near-maladjustment with urbanization lag to moderate coordination with ecological environment lag, under the precondition of transformation development. Strengthening ecological protection remains essential as urbanization continues to advance.

Signal and Information Processing

Entity Relationship Extraction Method Based on Bidirectional Decoding

LIU Hui, ZHANG Zhi, CHEN Yupeng

2025, 43(3):  491-503.  doi:10.3969/j.issn.0255-8297.2025.03.010

Asbtract ( 23 )   PDF (655KB) ( 6 )  

References | Related Articles | Metrics

To address the challenges of error propagation, overlapping triple problem and subject-object alignment in the existing relationship triplet extraction methods, this study proposes a novel bidirectional translation and decoding model. The model reframes the extraction process into three sub-tasks: entity extraction, subject-object alignment and relationship judgment. The bidirectional structure effectively alleviates error propagation, while the translation and decoding method based on the attention mechanism deals with the overlapping triple problem and aligns the subject and object. Finally, a bipartite entity-torelationship diagram fully explores the relationship between entity pairs, enabling accurate relationship judgment. Experimental results on public datasets have validated the performance of the proposed model.

Virtual Blocks Based Reversible Data Hiding in Encrypted Domain for Images

TANG Xin, FU Yaowen, ZHANG Yiwei, CHEN Haixin, ZHOU Yiteng

2025, 43(3):  504-518.  doi:10.3969/j.issn.0255-8297.2025.03.011

Asbtract ( 29 )   PDF (4544KB) ( 5 )  

References | Related Articles | Metrics

Adaptive most significant bit prediction (AMP) is an important technique to achieve reversible data hiding in encrypted images. AMP typically predicts pixel values by comparing the first pixel within a block with the remaining ones, extracting the longest common most significant bits to create space for embedding secret information. However, the significant difference between the first pixel and others can limit the embedding capacity. To solve this problem, this paper aggregates the unchanged pixels after data embedding within blocks and constructs a virtual pixel block. By applying the AMP algorithm again to the virtual pixel block, the proposed scheme enhances the embedding capacity. In order to further increase the number of virtual pixel blocks, this paper proposes a filling strategy. Taking a 2×2 pixel block as an example, when the block’s embedding capacity is large enough, the pixel correlation is increased by filling fixed bits. This ensures the first two pixels remain unchanged during embedding and can all be used to construct a virtual block. Because of the T field constructed by the filling strategy and the new 32 bits pixel structure proposed, pixel correlation is enhanced and embedding capacity is increased as a result. Experimental results on real-world datasets demonstrate that the proposed algorithm significantly outperforms existing AMP algorithms in terms of embedding capacity while maintaining data reversibility.

Prediction of Soil Organic Carbon for Cultivated Lands in Jianyang District of Nanping City Based on Soil Texture

YE Qing, XU Yehui, LI Huichuan, MA Dan, ZHANG Liming

2025, 43(3):  519-529.  doi:10.3969/j.issn.0255-8297.2025.03.012

Asbtract ( 23 )   PDF (1644KB) ( 4 )  

References | Related Articles | Metrics

The prediction accuracy of soil organic carbon contents in cultivated lands across hilly and mountainous areas is relatively low. This paper proposes a digital soil mapping method based on multi-temporal remote sensing and hybrid random forest (RF) models for soil texture. A total of 979 soil samples collected in 2008, Landsat 5 TM images with 30-meter spatial resolution during 2007—2010, a digital elevation model (DEM) with 12.5-meter spatial resolution and meteorological data with 1-kilometer spatial resolution were used as data sources. Remote sensing factors, relief factors and meteorological factors were extracted from those data sources. Then these factors were used to construct hybrid RF models for soil texture types and for classification probability of soil texture, respectively. The SOC predicting accuracy of the global RF models were analyzed compared with single-temporal and multi-temporal remote sensing factors. Furthermore, the accuracy of two hybrid RF models were also compared against that of the global RF model. Finally, the best-performing model was employed to predict SOC contents for cultivated lands in Jianyang District of Nanping City. The results showed that SOC prediction accuracy was higher for the multi-temporal synthetic Landsat 5 TM images (2007—2010) compared to single-temporal images. Specifically, the R² of hybrid RF model for classification probabilistic of soil texture improved by 53.57%, while the RMSE decreased by 11.20% relative to the global RF model. The spatial distributions of SOC contents generally exhibited higher levels in western regions and lower levels in central-eastern regions. The SOC map in study area became much smoother and more continuous in boundary regions. This study demonstrates that hybrid RF model for classification probabilistic of soil texture combined with multi-temporal synthetic Landsat 5 TM image can significantly improve SOC mapping accuracy in hilly and mountainous areas.

Control and System

Predefined-Time Trajectory Tracking Control for Dual-Arm Space Robot

HONG Mengqing, WANG Qun, LIU Jiayin

2025, 43(3):  530-540.  doi:10.3969/j.issn.0255-8297.2025.03.013

Asbtract ( 14 )   PDF (1082KB) ( 5 )  

References | Related Articles | Metrics

A predefined-time nonsingular sliding mode control method is proposed to address the issue of end-effector trajectory tracking for a free-floating dual-arm space robot subjected to unknown external disturbances, ensuring the end-effectors reach a stable state within predefined time. Firstly, the dynamic model of the dual-arm space robot is established using the Lagrange method. Furthermore, based on the theory of predefined time stability, a predefined-time nonsingular sliding mode surface is designed. The sliding mode surface not only allows for the pre-setting of system stabilization time, but also effectively avoids the singularity issues associated with predefined-time control. The stability of the closed-loop control system has been proven through the Lyapunov function method. Finally, simulation experiments conducted on a planar two-joint dual-arm space robot validate the effectiveness of the proposed control algorithm.