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....

2023 Vol.36 Issue.6, Published 2023-06-25 Papers and Reports    Researches and Applications        Papers and Reports 483 Modified Moth Search Algorithm Based on Adaptive ε-Constrained Method FENG Yanhong, WANG Gaige, LI Mingliang, LI Xi The multidemand multidimensional knapsack problem includes two types of inequality constraints with conflicts, making the search for the feasible solution region exceptionally difficult. Therefore, a modified moth search algorithm(MMS) based on adaptive ε-constrained method is proposed in this paper. In the Lévy flight phase, the step is adjusted according to the current iteration. In the straight flight phase, the mutation rate is introduced to increase the diversity of the population. Finally, the uniform mutation operator is applied to the whole population to improve the global search capability of the algorithm. The space mapping method is utilized to transfer the search space to the problem space, and the adaptive ε-constrained method is adopted. Experiments on classic 96 benchmark instances show that adaptive lévy flight operator, mutation straight flight operator and uniform mutation operator contribute significantly to the solution accuracy of the algorithm and the proposed algorithm performs better on the majority of instances. Furthermore, orthogonal experimental design method is utilized to analyze the influence of parameters on the ε-constrained method. 2023 Vol. 36 (6): 483-494 [Abstract] ( 241 ) [HTML 1KB] [ PDF 796KB] ( 512 ) 495 Optimal Scale Selection and Attribute Reduction of Multi-scale Multiset-Valued Information Systems Based on Entropy WANG Leixi, WU Weizhi, XIE Zhenhuang Existing information systems are difficult to reflect and deal with the data duplication in the process of data fusion. In this paper, the concept of multi-scale multiset-valued information systems is introduced and the optimal scale selection and attribute reduction in these systems are discussed. Firstly, a similarity relation on the universe of discourse from any attribute subset in a multi-scale multiset-valued information system is defined by employing the Hellinger distance on multi-sets of the domain of any attribute. Then, information granules in the form of similarity classes are constructed. Knowledge rough entropy is further introduced in the context of multi-scale multiset-valued information systems. Optimal scales based on the similarity relation and the knowledge rough entropy are defined in a multi-scale multiset-valued information system, respectively. It is examined that the optimal scale based on the similarity relation and entropy optimal scale are equivalent. Finally, reducts and entropy reducts based on the optimal scale are discussed in the multi-scale multiset-valued information system, and algorithms for calculating the entropy optimal scale and an entropy reduct are also designed in a multi-scale multiset-valued information system. 2023 Vol. 36 (6): 495-510 [Abstract] ( 181 ) [HTML 1KB] [ PDF 754KB] ( 220 ) 511 Transformer-Based Multi-scale Optimization Network for Low-Light Image Enhancement NIU Yuzhen, LIN Xiaofeng, XU Huangbiao, LI Yuezhou, CHEN Yuzhong Enhancing low-light images with high quality is a highly challenging task due to the features of low-light images such as brightness, color, and details in the information of different scales. Existing deep learning-based methods fail to fully utilize multi-scale features and fuse multi-scale features to comprehensively enhance the brightness, color and details of the images. To address these problems, a Transformer-based multi-scale optimization network for low-light image enhancement is proposed. Firstly, the Transformer-based multi-task enhancement module is designed. Through multi-task training, the Transformer-based enhancement module gains the ability to globally model brightness, color, and details. Therefore, it can initially cope with various degradation challenges commonly found in low-light images, such as insufficient brightness, color deviation, blurred details and severe noises. Then, the architecture combining global and local multi-scale features is designed to progressively optimize the features at different scales. Finally, a multi-scale feature fusion module and an adaptive enhancement module are proposed. They learn and fuse the information association among different scales, while adaptively enhancing images in various local multi-scale branches. Extensive experiments on six public datasets, including paired or unpaired images, show that the proposed method can effectively solve the problems of multiple degradation types, such as brightness, color, details and noise in low-light images. 2023 Vol. 36 (6): 511-529 [Abstract] ( 442 ) [HTML 1KB] [ PDF 14392KB] ( 412 ) Researches and Applications 530 An Alpha Matting Algorithm Based on Micro-scale Searching for High-Resolution Images FENG Fujian, YANG Yuan, TAN Mian, GOU Hongshan, LIANG Yihui, WANG Lin High-resolution image matting is essentially a large-scale combinatorial optimization for foreground/background pixel pairs. However, there is few research achievements on this issue. Alpha matte inverse extraction occurs when the foreground and the background in an image are highly similar. To address this problem, a decision set decomposition strategy is designed to effectively decompose high-resolution image matting problems. Moreover, a optimized information transmission strategy is designed, the weight relationship between sub-problems is obtained, and the optimization sequence of the image matting problem is established. Based on the optimized information transmission strategy, an alpha matting algorithm based on micro-scale searching(MS-AM) is proposed. MS-AM effectively solves the issue of alpha matte inverse extraction in high-resolution image matting problems by searching through effective decision subsets instead of the entire decision set, providing insights for the analysis of large-scale combinatorial optimization problems. The alphamatting benchmark dataset is selected as testing data, and MS-AM is compared with typical matting optimization algorithms. Results demonstrate that MS-AM can address alpha matte inverse extraction problem when the foreground is similar to the background and improve the alpha matting accuracy with significantly reduced dimension of high-resolution image matting problem. 2023 Vol. 36 (6): 530-543 [Abstract] ( 240 ) [HTML 1KB] [ PDF 3169KB] ( 314 ) 544 Pneumonia Classification and Recognition Method Based on Multi-resolution Attention Dense Network ZHOU Tao, YE Xinyu, LU Huiling, CHANG Xiaoyu, LIU Yuncan X-ray film of pneumonia suffers from inconspicuous imaging features, low contrast between lesions and surrounding tissues, and blurred edges. Therefore, a pneumonia classification and recognition method based on multi-resolution attention dense network is proposed. Shallow localization information and deep semantic information are deeply fused. A multi-resolution spatial attention gate is constructed to enhance semantic interaction between deep and shallow information at different resolutions, establishing interdependency for lesion information in deep and shallow information. In addition, the coordinate frequency attention is designed to adaptively enhance the representation of pneumonia features in a complementary manner of orientation and location. Experiments on five pneumonia X-ray datasets including ChestXRay2017 show that the proposed method achieves better performance in pneumonia classification and recognition task with robustness on the public pneumonia dataset. 2023 Vol. 36 (6): 544-555 [Abstract] ( 228 ) [HTML 1KB] [ PDF 2626KB] ( 447 ) 556 Non-negative Orthogonal Matrix Factorization Based Multi-view Clustering Image Segmentation Algorithm ZHANG Rongguo, CAO Junhui, HU Jing, ZHANG Rui, LIU Xiaojun Multi-view graph clustering shows some advantages in dealing with nonlinear structured data, but it exhibits drawbacks such as the need of post-processing and low time efficiency. To solve this problem, a non-negative orthogonal matrix factorization based multi-view clustering image segmentation algorithm(NOMF-MVC ) is proposed. Firstly, multi-view data of an image is extracted, and the manifold learning nonlinear dimensionality reduction method is employed to obtain the spectral embedding matrix of each view. Corresponding spectral block structure is constructed and it is fused into a consistency graph matrix via designed adaptive weights. Secondly, the non-negative embedding matrix is obtained by the non-negative orthogonal matrix factorization of the consistency graph matrix. Finally, the clustering of multi-view features is performed by the non-negative embedding matrix, and thereby image segmentation results are yielded. Comparative experiments on five datasets show certain improvements in segmentation accuracy and time efficiency achieved by NOMF-MVC. 2023 Vol. 36 (6): 556-571 [Abstract] ( 317 ) [HTML 1KB] [ PDF 4207KB] ( 266 )

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