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....
 
 
2014 Vol.27 Issue.1, Published 2014-01-30
Papers and Reports    Researches and Applications    Surveys and Reviews   
   
Papers and Reports
1 Knowledge Guided Particle Swarm Optimization for Feature Selection
GONG Dun-Wei, HU Ying, ZHANG Yong
Feature selection is one of important data processing methods in pattern classification. A method of selecting features is proposed based on knowledge guided particle swarm optimization. The problem of feature selection which contains discrete variables is converted to an optimization one with continuous variables by encoding the particles with the selected probabilities of features. The type and its updated probability of the feature are determined by the particle′s fitness and the selection frequency of the particle component in order to speed up the convergence of the swarm. The experimental results on 10 typical test datasets and a clinical diagnosis dataset of hepatitis show that the proposed method improves the classification accuracy on the premise of reducing the number of features.
2014 Vol. 27 (1): 1-10 [Abstract] ( 488 ) [HTML 1KB] [ PDF 636KB] ( 630 )
11 Research on Fast Dense Stereo Matching Technique Using Adaptive Mask
ZHOU Jia-Li, WU Min, ZHOU Hua-Ping
Based on the classical phase-only correlation algorithms, a fast stereo matching method is proposed for dense point cloud. The adaptive discrete mask is used to the matching weight of the similar dense fields by estimating the gradient of matching points, thus the reconstruction precision and reliability are improved. Moreover, the proposed method also improves computational efficiency via storing and reusing the intermediate data of 2D DFT. Since the proposed algorithm satisfies SIMD model, the GPU parallel computing makes the matching process basically reach real-time . The experimental results show that the proposed fast phase-only correlation algorithm performs well in the surface reconstruction of smooth irregular diffuse objects obtained from short-baseline parallel-axis binocular stereo device, therefore it can be widely used in 3D face recognition and other fields.
2014 Vol. 27 (1): 11-20 [Abstract] ( 357 ) [HTML 1KB] [ PDF 1558KB] ( 793 )
21 A Hierarchical Model-Set Adaptation Method for Multiple Model Filter
XU Xue-Song
Model-set adaptation method is important for VSMM. In this paper, a hierarchical model-set adaptation method is proposed, which mimics the working principle for B cells recognizing antigens in immune system. According to this method, the system model-set is decomposed into base grid and moving grid. The base grid is used to detect the distribution of system model by rough searching and the moving grid is used to obtain the final filtering results by fine searching the area indicated by base grid. The calculation time of interactive filtering algorithm is effectively reduced, since the whole large model-set is decomposed into two small hierarchical model-sets. The simulation results show that higher filtering accuracy can be obtained with less calculation time.
2014 Vol. 27 (1): 21-27 [Abstract] ( 311 ) [HTML 1KB] [ PDF 898KB] ( 530 )
28 Weighted Frequent Pattern Tree Structure Algorithm Based on Information Entropy
ZHAO Xu-Jun, CAI Jiang-Hui, MA Yang
In association rule mining, the importance of items is different and can not be subjectively given, which affects the mining result. The weighted items and weighted association rules are given, in which the weights of single attribute are determined by information entropy and the weights of items are determined by the compromise method between geometric mean and maximum weight value. Thus, the important projects are highlighted and the overall weights are balanced at the same time. On the basis of all above factors, weighted frequent patterns are extracted by using weighted frequent pattern tree, and the structure method of weighted frequent pattern tree is given. Finally, the experimental results on the spectral data of celestial body and the mechanical equipment EDEM verify the high efficiency of the proposed algorithm.
2014 Vol. 27 (1): 28-34 [Abstract] ( 464 ) [HTML 1KB] [ PDF 377KB] ( 645 )
Surveys and Reviews
35 A Survey of Human Body Action Recognition
LI Rui-Feng, WANG Liang-Liang, WANG Ke
Action recognition has become a hotspot in the fields of video surveillance, virtual reality, human-computer interaction and others recently. In this paper, action recognition is comprehended as a process of detecting action data, called symbols of action message, and distinct actions based on action feature extraction and reception are further classified. On the basis, an overview of vision-based full-body action recognition techniques is presented within the domain of moving object detection, action feature extraction and action feature perception, and the corresponding methods are classified. Besides, the research trend of action recognition is discussed.
2014 Vol. 27 (1): 35-48 [Abstract] ( 751 ) [HTML 1KB] [ PDF 649KB] ( 3429 )
Researches and Applications
49 Informative Gene Selection for Tumor Classification Based on Iterative Lasso
ZHANG Jing, HU Xue-Gang, LI Pei-Pei, ZHANG Yu-Hong
Tumor classification based on gene expression profiles, which is of tremendous convenience for cancer accurate diagnosis and subtype recognition, has drawn a great attention in recent years. Due to the characteristics of small samples, high dimensionality, much noise and data redundancy for gene expression profiles, it is difficult to mine biological knowledge from gene expression profiles profoundly and accurately, and it also brings enormous difficulty to informative gene selection in the tumor classification.Therefore, an iterative Lasso-based approach for gene selection,called Gene Selection Based on Iterative Lasso(GSIL), is proposed to select an informative gene subset with fewer genes and better classification ability. The proposed algorithm mainly involves two steps. In the first step, a gene ranking algorithm, Signal Noise Ratio, is applied to select top-ranked genes as the candidate gene subset, which aims to eliminate irrelevant genes. In the second step, an improved method based on Lasso, Iterative Lasso, is employed to eliminate the redundant genes. The experimental results on 5 public datasets validate the feasibility and effectiveness of the proposed algorithm and demonstrate that it has better classification ability in comparison with other gene selection methods.
2014 Vol. 27 (1): 49-59 [Abstract] ( 442 ) [HTML 1KB] [ PDF 603KB] ( 1256 )
60 A Geometrical Judgment Method for Linear Separability Based on Pseudo-Separating Hyperplane and Its Application
ZHANG Yin-Chuan, HAN Li-Xin, ZENG Xiao-Qin
Aiming at the problem of linear separability in pattern classification, the patterns are taken as points in Euclidean space, the geometric properties including the relationship between points and planes in Euclidean space are studied, and the pseudo-separating hyperplane is defined based on the general separating hyperplane. By analyzing linear separability equivalence, the mapping from a higher dimensional space to a lower dimensional space is developed when spatial dimension reduction is required. The method for finding pseudo-separating hyperplane is studied and a judgment method for linear separability is presented with obvious geometric meaning. A classification complexity measure is proposed based on this method. The experimental results show that the proposed method reflects the complexity of data classification well.
2014 Vol. 27 (1): 60-69 [Abstract] ( 343 ) [HTML 1KB] [ PDF 882KB] ( 819 )
70 Robust Occlusion Pattern Recognition Algorithm Based on Block Sparse Recursive Residuals Analysis
HU Zheng-Ping, ZHAO Shu-Huan, LI Jing
A robust occlusion pattern recognition algorithm is proposed which considers how to detect occluded region automatically with unknown occlusion pattern and conquers the influence of occluded region to improve the robustness of the recognition algorithm. Firstly, the test image is divided into up module and down module. Next, the sparse representations are computed respectively. Then, the module with higher sparsity and the corresponding sparse solution are found. The test image is reconstructed using this module and the N largest coefficients. According to the residual of original test image and reconstruction image, the occluded pixels can be confirmed coarsely. Considering the continuity of occluded region, the coarsely confirmed occluded region is regularized by morphological operation and gets the weighting matrix. Finally, the test and training set are weighted and normalized by using this weighting matrix and then the final decision is made by using global sparse representation. The experimental results on AR, Yale B and MNIST databases verify that the proposed method can detect the occluded region roughly, and the effectiveness and the robustness of the proposed method can be observed obviously.
2014 Vol. 27 (1): 70-76 [Abstract] ( 339 ) [HTML 1KB] [ PDF 1228KB] ( 745 )
77 Classification Probability Preserving Discriminant Analysis and Its Application to Face Recognition
YANG Zhang-Jing, LIU Chuan-Cai, HUANG Pu, ZHU Jun
To solve the problems in feature extraction algorithms, an algorithm based on linear discriminant analysis (LDA), called classification probability preserving discriminant analysis (CPPDA), is proposed for face recognition. Firstly, the classification probability of each sample is computed by CPPDA, and both the between-class scatter matrix and the within-class scatter matrix are redefined by the classification probability. Secondly, through maximizing the between-class scatter and minimizing the within-class scatter simultaneously, an optimal projection matrix can be preserved in the low-dimensional feature space, such as the distribution information contained in the original data. Finally, the experimental results on the ORL,Yale and FERET face databases demonstrate the superiority of the proposed algorithm compared with other algorithms.
2014 Vol. 27 (1): 77-81 [Abstract] ( 386 ) [HTML 1KB] [ PDF 438KB] ( 587 )
82 Research on Dataset Feature Structure Based on Interaction Information
LIU Juan, ZHU Xiang-Ou, LIU Wen-Bing
In machine learning area, classification algorithms are widely studied and a large number of different types of algorithms are proposed. How to select appropriate ones from so many classification algorithms for the datasets becomes a crucial problem. Recently, a new method in reference [8] is proposed to characterize datasets and achieve better results in algorithm recommendation. In this paper, two methods are presented to characterize datasets under the theory of interaction information. The performance of 12 different types of classification algorithms on the 98 UCI datasets illustrates that both two-variable and three-variable interaction information methods can improve the precision and the hit rate of recommended algorithms. Furthermore, the latter performs even better under datasets with poor adaptability.
2014 Vol. 27 (1): 82-88 [Abstract] ( 430 ) [HTML 1KB] [ PDF 1159KB] ( 829 )
89 Vehicle Tracking Based on Improved Deterministic Data Association
ZHOU Jun-Jing, DUAN Jian-Min, Yang Guang-Zu

The existing deterministic data association defines the optimal track set by global optimization algorithms, which can only work with prior knowledge of the objects number. Therefore, it is limited to vehicle detection in driver assistance system. A local optimization algorithm is proposed to implement deterministic correspondence in vehicle detection. With the aid of a proper tracker management strategy, the algorithm can handle object entries, object exits and occlusions. To improve the accuracy of correspondence, the cost function is defined by fusing multiple features. The motion feature and the figuration features are taken as main constraint condition and minor ones respectively. The validity of the tracking algorithm in vehicle detection system is verified by experiments.
2014 Vol. 27 (1): 89-96 [Abstract] ( 317 ) [HTML 1KB] [ PDF 572KB] ( 616 )
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