Home      About Journal      Editorial Board      Instructions      Ethics Statement      Contact Us                   中文

Judgement and Disposal of Academic Misconduct Article Copyright Transfer Agreement Proof of Confidentiality Requirements for Electronic Version More.... Chinese Association of Automation National ResearchCenter for Intelligent Computing System Institute of Intelligent Machines,Chinese Academy of Sciences More....

2025 Vol.38 Issue.7, Published 2025-07-25 Papers and Reports    Researches and Applications        Papers and Reports 577 Underwater Image Enhancement Network Based on Semantic Collaborative Perceptual Attention YANG Jing, LIANG Hui, ZHU Wenhan, YANG Shuo, WU Zhize To address the image degradation caused by light attenuation and scattering in underwater imaging, an underwater image enhancement network based on semantic collaborative perceptual attention is proposed. First, a dual-path competitive perceptual attention mechanism is designed. Sliding window attention and pooling attention are integrated into Softmax to synchronously capture both coarse-grained and fine-grained features of images, thereby enabling multi-scale feature perception. Second, a convolutional gated linear unit is introduced to realize the attentionalization of the channel mixer. Based on the above, a cascaded perceptual attention network is constructed by integrating the perceptual attention mechanism and convolutional gated linear unit to capture and fuse local and global information of underwater degraded images. Finally, a feature dominant module is developed to embed and collaboratively integrate semantic features, and the capability of the model to understand and represent underwater scene semantics is enhanced. This elevates the enhancement process from mere pixel-level restoration to semantic-level scene reconstruction. Experiments demonstrate that the proposed network exhibits superior generalization ability and significant application value in downstream visual engineering tasks. 2025 Vol. 38 (7): 577-595 [Abstract] ( 207 ) [HTML 1KB] [ PDF 8095KB] ( 150 ) 596 Cross-Modal Interactive Image Editing Based on Bidirectional Collaboration with Large Language Models SHI Hui, JIN Conghui Diffusion models exhibit high visual fidelity in image generation tasks. However, they are confronted with critical challenges in image editing, such as ambiguity in user intent interpretation, insufficient control over local details, and lag in interactive response. To address these issues, a cross-modal interactive image editing method based on bidirectional collaboration with large language models(BiC-LLM) is proposed. A bidirectional collaboration mechanism is introduced as its core. The top-down semantic guidance from large language models is combined synergistically with bottom-up direct interaction from users. Therefore, controllability and precision in image editing are fundamentally enhanced by employing semantic enhancement, feature decoupling and a dynamic feedback mechanism. First, a hierarchical semantic-driven module is designed. The user-input text is decoupled and reasoned by the large language model, and fine-grained semantic vectors are generated to interpret user intent precisely. Second, a dynamic control module for vision-structure decoupling is constructed. Multi-level visual feature extractors and object-level modeling are combined to achieve independent control over global structure and local appearance. Finally, a real-time interaction mechanism is introduced to enable users to dynamically intervene in the editing process through mask annotations and parameter adjustments, thereby supporting iterative optimization. Experiments on LSUN, CelebA-HQ, and COCO datasets demonstrate that BiC-LLM significantly outperforms baseline models in terms of textual consistency, structural stability, and interactive controllability. Moreover, BiC-LLM effectively enables multi-object semantic editing in complex scenes while preserving the integrity of unedited regions, demonstrating its robustness and effectiveness in image editing tasks. 2025 Vol. 38 (7): 596-612 [Abstract] ( 101 ) [HTML 1KB] [ PDF 5040KB] ( 95 ) 613 Hierarchical Topic Model Based on Chain of Thought and Semantic Decoupling WANG Zhihua, LI Yang, LI Deyu, WANG Suge Hierarchical topic models can uncover latent topics in documents and model the hierarchical relationships between topics, providing technical support for applications such as data governance, information retrieval, content classification, and knowledge management. A hierarchical topic model based on chain of thought and semantic decoupling(CoT-SDHT-M) is proposed in this paper. First, a hierarchical topic generation module based on a chain of thought is established. An initial hierarchical topic structure is generated by a large language model(LLM) under the guidance of hierarchical topic generation chain of thought. Then, a topic similarity discrimination mechanism based on LLM is introduced to generate refined topics and to guide the LLM in merging topics through examples, thereby improving the quality of the generated topics. Finally, a hierarchical topic optimization module based on transport planning and semantic decoupling is designed. It incorporates the initial hierarchical structure as a topic prior for downstream modeling. The relationships between topics are modeled as an optimal transport problem, and parent-child topic decoupling is performed based on the keywords of upper-layer and lower-layer topics to optimize the hierarchical topic structure. The experiments on various standard public datasets, including NeurIPS, ACL and 20 Newsgroups, demonstrate that CoT-SDHT-M significantly outperforms existing baseline models in terms of topic quality metrics and hierarchical metrics. 2025 Vol. 38 (7): 613-626 [Abstract] ( 96 ) [HTML 1KB] [ PDF 890KB] ( 203 ) Researches and Applications 627 Dynamic Update of Attribute Reduction Based on Property Pictorial Diagrams BAI Pu, WAN Qing, MA Yingcang, WEI Ling Attribute reduction is a prominent research focus in formal concept analysis, and exploring its dynamic update methods is crucial for knowledge discovery. The property pictorial diagram, a Hasse diagram representation of a formal context, can be employed to derive attribute reducts that preserve the concept lattice structure. In this paper, dynamic update methods for attribute reduction are investigated under the changes in the attribute set of a formal context by analyzing the update rules of the property pictorial diagram. First, a relation matrix for the property pictorial diagram is defined using the upper(lower) neighborhood relations among attributes, and its properties are studied. Then, update methods for the property pictorial diagram are proposed based on the relation matrix for two cases: attribute deletion and attribute addition. Finally, based on the update rules of the property pictorial diagram, change rules of attribute characteristics are given, and then dynamic update methods for attribute reduction are developed. The proposed methods further enrich the theoretical foundation of attribute reduction and numerical experiments demonstrate their effectiveness. 2025 Vol. 38 (7): 627-640 [Abstract] ( 76 ) [HTML 1KB] [ PDF 785KB] ( 89 ) 641 Graph Anomaly Detection Based on Local Differential and Global Spectral Fusion XU Dengbin, YUAN Lining, WU Peichen, LIU Zhao Graph neural networks(GNNs) are widely recognized for their effectiveness in graph anomaly detection tasks. However, existing methods commonly struggle to identify camouflaged anomalies and exhibit poor performance under label scarcity. To address these issues, a method for graph anomaly detection based on local differential and global spectral fusion(GAD-LDSF) is proposed. Feature embeddings are employed in both spatial and spectral domains to enhance performance under camouflaged anomalies and label scarcity. First, probability-adaptive continuum embedding based on soft quantization is utilized to achieve data augmentation. A graph differential attention network is employed to accurately capture subtle differences between nodes and generate node representations in the spatial domain. Then, a globally homogenized dynamic spectral enhancement module is constructed. Chebyshev polynomials are leveraged to efficiently capture global anomaly signals and generate node representations in the spectral domain. Finally, spatial domain and spectral domain node representations are dynamically fused, and a predictor is employed to achieve anomalous node detection. Experiments on three benchmark datasets validate that GAD-LDSF achieves overall superior performance, particularly demonstrating its strong robustness and generalization in handling challenges such as camouflaged anomalies and label scarcity. 2025 Vol. 38 (7): 641-654 [Abstract] ( 100 ) [HTML 1KB] [ PDF 3183KB] ( 94 ) 655 Image Tampering Detection Network Based on Edge Information and Contrastive Learning WANG Yiqun, GAO Yancheng To address the insufficient utilization of edge information and high-frequency features in image tampering detection under complex scenarios, an image tampering detection network based on edge information and contrastive learning(EICL-Net) is proposed. First, a dynamic weight update strategy is designed to enhance the feature extraction capability for high-frequency image information. Next, by integrating edge detection algorithms with tampered region detection algorithms, the edge features of images are extracted and enhanced, and the saliency of anomalous information is improved. Finally, a contrastive learning mechanism is introduced to optimize the ability to distinguish pixel distribution differences by constructing positive and negative sample pairs for feature comparison, thereby achieving precise localization of tampered regions. Experiments on multiple public datasets demonstrate that EICL-Net exhibits strong generalization performance and the ability to identify subtle tampering traces under complex scenarios. Therefore, EICL-Net offers a solution to image tampering detection. With its high practical application value, EICL-Net can be widely applied in the fields such as information security and digital forensics. 2025 Vol. 38 (7): 655-667 [Abstract] ( 105 ) [HTML 1KB] [ PDF 3612KB] ( 102 )

Supervised by China Association for Science and Technology Sponsored by Chinese Association of Automation NationalResearchCenter for Intelligent Computing System Institute of Intelligent Machines, Chinese Academy of Sciences Published by Science Press

Copyright © 2010 Editorial Office of Pattern Recognition and Artificial Intelligence Address: No.350 Shushanhu Road, Hefei, Anhui Province, P.R. China Tel: 0551-65591176 Fax:0551-65591176 Email: bjb@iim.ac.cn Supported by Beijing Magtech 　Email:support@magtech.com.cn