模式识别与人工智能
Home      About Journal      Editorial Board      Instructions      Ethics Statement      Contact Us                   中文
Pattern Recognition and Artificial Intelligence
22 Judgement and Disposal of Academic Misconduct Article
22 Copyright Transfer Agreement
22 Proof of Confidentiality
22 Requirements for Electronic Version
More....
22 Chinese Association of Automation
22 National ResearchCenter for Intelligent Computing System
22 Institute of Intelligent Machines,Chinese Academy of Sciences
More....
 
 
2016 Vol.29 Issue.2, Published 2016-02-29
Papers and Reports    Researches and Applications   
   
Papers and Reports
97 Semi-supervised Projection Twin Support Vector Machine via Manifold Regularization
CHEN Weijie,SHAO Yuanhai, LI Chunna,DENG Naiyang

Projection twin support vector machine (PTSVM) is a supervised learning method and its performance deteriorates when supervised information is insufficient. To resolve this issue, a semisupervised projection twin support vector machine (SPTSVM) is proposed inspired by the manifold regularization. Both supervised (labeled) and unsupervised (unlabeled) information are utilized to build a more reasonable semisupervised classifier. Compared with PTSVM, SPTSVM takes the intrinsic geometric information into full consideration via manifold regularization. Furthermore, by selecting appropriate parameters, SPTSVM degenerates into either supervised PTSVM or projection twin support vector machine with regularization term. The effectiveness of the proposed approach is demonstrated by comparison on both artificial and realworld datasets.
2016 Vol. 29 (2): 97-107 [Abstract] ( 640 ) [HTML 1KB] [ PDF 551KB] ( 739 )
108 Optimal Granularity Selections in Consistent Incomplete Multi-granular Labeled Decision Systems
WU Weizhi, CHEN Ying, XU Youhong, GU Shenming
Aiming at knowledge acquisition in incomplete information systems with multi-granular labels, the concept of incomplete multi-granular labeled information systems is firstly introduced.Similarity relations of an incomplete multi-granular labeled information system are then defined.Representations of information granules with different levels of granulation as well as their relationships are also explored.Lower and upper approximations based on similarity relations with different levels of granulation are further defined and their properties are presented.Finally, by belief and plausibility functions in Dempster-Shafer theory of evidence, optimal granularity selections in consistent incomplete multi-granular labeled decision systems are discussed.
2016 Vol. 29 (2): 108-115 [Abstract] ( 485 ) [HTML 1KB] [ PDF 348KB] ( 449 )
116 Comparative Analysis of Quantum State Estimation Algorithm Based on Compressive Sensing
CONG Shuang , ZHANG Hui, LI Kezhi
The alternating direction method of multipliers (ADMM) is used to estimate quantum density matrix with 6 qubits based on the completed research on 5 qubits estimation. In addition, the comparison with least squares and Dantzig optimization method is studied under the situations with and without external interference. The optimization schemes are implemented in Matlab environment to realize the fast estimation of quantum pure state. The experimental results show that ADMM is superior to two other algorithms in estimation accuracy and resistance to external disturbances.
2016 Vol. 29 (2): 116-121 [Abstract] ( 452 ) [HTML 1KB] [ PDF 418KB] ( 501 )
122 Online Structure Learning Algorithm for Weighted Networks
JIANG Xiaojuan, ZHANG Wensheng
With continuous development of internet technology, the scope of network datasets increases massively. Analyzing the structure of network data is a research hotspot in machine learning and network applications. In this paper, a scalable online learning algorithm is proposed to speed up the inference procedure for the latent structure of weighted networks. Firstly, the exponential family distribution is utilized to represent the generative process of weighted networks. Then, using stochastic variational inference technique, the online-weighted stochastic block model (ON-WSBM) is developed to efficiently approximate the posterior distribution of underlying block structure. In ON-WSBM an incremental approach based on the subsampling method is adopted to reduce the time complexity of optimization, and then the stochastic optimization method is employed by using natural gradient to simplify the calculation and further accelerate the learning procedure. Extensive experiments on four popular datasets demonstrate that ON-WSBM can efficiently capture the community structure of the complex weighted networks, and can achieve comparatively high prediction accuracy in a short time.
2016 Vol. 29 (2): 122-130 [Abstract] ( 404 ) [HTML 1KB] [ PDF 582KB] ( 622 )
131 Object Tracking Model Based on Fusing Multiple Weighted Distribution Field Features
LUO Huilan, SHAN Shunyong, KONG Fansheng
Object tracking is difficult to be implemented by using single feature. Histogram is simple and convenient to describe the image features. However, the spatial information of the features can not be expressed by the statistical histograms, and the distribution field descriptors can reflect the spatial information of the features. Based on their advantages, an object tracking algorithm is proposed by fusing gray value features, texture features and edge features. Three kinds of features are combined through distribution field descriptors to form joint representations. And the distribution layers of dense distribution field features are multiplied by the corresponding weights to construct an efficient target model. An adaptive object model updating scheme is used to update the target model and adapt to varietiesof the background and the illumination. The experimental results on commonly used testing video sequences show that the proposed algorithm generates better performance in complicated situations, such as pose change, rotation, occlusion and illumination changes and it has stronger robustness.
2016 Vol. 29 (2): 131-142 [Abstract] ( 498 ) [HTML 1KB] [ PDF 1315KB] ( 442 )
143 Haze Forecast Method of Selective Ensemble Based on Glowworm Swarm Optimization Algorithm
NI Zhiwei, ZHANG Chen, NI Liping
Haze is a kind of serious environmental pollution. Therefore, haze weather forecast is an effective way to minimize the negative influence of haze. A selective ensemble learning based on glowworm swarm optimization algorithm is proposed. Firstly, some individual support vector machines are trained by the mixed kernel support vector machine independently, and then some classifiers with high precision and diversity are selected by the improved discrete glowworm swarm optimization algorithm. Finally, the classification results are obtained by majority voting. The proposed algorithm is utilized to forecast haze weather in China. Experimental results show that it has higher effectiveness and feasibility.
2016 Vol. 29 (2): 143-153 [Abstract] ( 549 ) [HTML 1KB] [ PDF 588KB] ( 655 )
Researches and Applications
154 Recognition of Complex Human Behavior Based on Key Frames
XIA Limin, SHI Xiaoting
A key frames based method is proposed for recognition of complex human behavior. The body contours are adopted to represent actions. The complex behavior fragment is divided into simple behaviors by the shot boundary detection algorithm. To improve the recall rate, double sampling is employed during the segmentation. The self-splitting competitive learning algorithm is utilized to acquire key frames of simple behaviors. Finally, the recognition of complex human behavior is achieved according to the similarity of the key frames. During the process, the visual factor, the order factor and the interference factor are taken into account. Thus, the calculations of the recognition are more rational and comprehensive. The proposed method is verified on UCF Sports database and self-built database and the results indicate the high recognition accuracy of the proposed method.
2016 Vol. 29 (2): 154-162 [Abstract] ( 510 ) [HTML 1KB] [ PDF 1594KB] ( 1143 )
163 Multi-view Clustering Based Natural Image Contour Detection
ZHANG Heng, TAN Xiaoyang, JIN Xin
The gradient feature gives an invariant description for linear lighting changes while sparse coding methods can exploit the data statistics from the image data point. In multi-view clustering algorithm, different attributes set in the same cluster are considered as different views, and the importance of different views is taken into account for co-clustering. An algorithm based on multi-view clustering for image contour detection is proposed and it integrates both features into a unified multi-view clustering framework to effectively improve the robustness of the detection system. The combination of image local features and sparse code features is utilized to train model, and the spatial information and curvature information of the image pixels are added to obtain the global features and ensure the accuracy of the contour detection and region consistency. Experiments on two large public available datasets show the feasibility and effectiveness of the proposed algorithm.
2016 Vol. 29 (2): 163-170 [Abstract] ( 632 ) [HTML 1KB] [ PDF 2024KB] ( 1069 )
171 Recommendation Algorithm of Collaborative Filtering Graph Model Based on Belief Network
ZHU Kunlei, HUANG Jiajin
Information retrieval model has been applied to the collaborative filtering algorithm now. The belief network model in information retrieval is used to describe user-based collaborative filtering and item-based collaborative filtering uniformly, and a recommendation algorithm of collaborative filtering graph model based on belief network is put forward. Due to the property that belief network is convenient to combine the information of additional sources, the expert information is added to the collaborative filtering model to provide decision support for the users, and consequently the data sparse problem of the recommendation system is solved. Experimental results show that the proposed algorithm improves the recommendation accuracy.
2016 Vol. 29 (2): 171-176 [Abstract] ( 509 ) [HTML 1KB] [ PDF 369KB] ( 514 )
177 Adaptive Thresholding Based Edge Detection Approach for Images
LI Minhua, BAI Meng, Lü Yingjun
To detect the edge of noisy image, an adaptive thresholding based edge detection approach is proposed. In this approach, the differential operators of the two-dimensional Gaussian function are used to design the multi-oriented edge detection filter. The image gradient is computed based on the designed filters. To reduce the effect of noise to the gradient image, an adaptive method is proposed to determine the filter size based on the candidate thresholds. After the filter size is determined, an adaptive thresholding method is proposed to select the hysteresis threshold. The proposed edge detection approach is evaluated under different noise conditions in experiments. The relationships among filter sizes, hysteresis thresholds and the proposed algorithm performance are studied. Experimental results demonstrate that the proposed approach determines the filter size and hysteresis threshold based on the image noise adaptively and it produces good anti-noise performance.
2016 Vol. 29 (2): 177-184 [Abstract] ( 636 ) [HTML 1KB] [ PDF 4986KB] ( 788 )
185 FCM Neural Network Classifier Using Density-Based Spatial Clustering of Applications with Noise
ZHANG Xiaoqian, YANG Bo, WANG Lin, LIANG Zhifeng
Arbitrary shape clusters are difficult to be found by the K-means algorithm adopted in the implementation of floating centroids method (FCM). Moreover, K-means algorithm is sensitive to outliers. Aiming at these problems, density-based spatial clustering of applications with noise (DBSCAN) algorithm is used to improve FCM neural network classifier in this paper. The outliers are considered as the points that can not be dealt with, and clusters of arbitrary shape can be found by DBSCAN algorithm. Thus, the color points in the partition space can be divided into several more accurate partitions. In addition, an optimization objective function is defined, and the particle swarm optimization algorithm is employed to optimize the parameters of the neural network to obtain an optimal classification model. Several commonly used datasets from UCI database are selected to conduct a comparative experiment. The experimental results show that the improved FCM method generates better performance on classification accuracy, robustness and running time than that of the original FCM method.
2016 Vol. 29 (2): 185-192 [Abstract] ( 569 ) [HTML 1KB] [ PDF 465KB] ( 621 )
模式识别与人工智能
 

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