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....

2007 Vol.20 Issue.6, Published 2007-12-25 Papers and Reports    Researches and Applications        Papers and Reports 727 ModelLikelihood Based SuperParentOneDependence Estimator Ensemble Method LI Nan, JIANG Yuan, ZHOU ZhiHua Averaged onedependence estimators (AODE) is an important Bayesian learning method. However, in AODE all outputs of the superparentonedependence estimators (SPODEs) are equally considered, which may bring bad influences to the final results. In this paper, every SPODE is viewed as a generative model, and the model likelihood is used to measure its performance. Then, a new approach, named modellikelihood based superparentone dependence estimator (LODE), is proposed which integrates the SPODEs based on model likelihood. Compared with AODE, LODE significantly improves the classification performance with only a slight increase in computation. 2007 Vol. 20 (6): 727-731 [Abstract] ( 285 ) [HTML 1KB] [ PDF 316KB] ( 505 ) 732 Fuzzy Robust Tracking Control for Aerospace Vehicle's Reentry Attitude Based on Fuzzy Feedforward WANG YuHui, WU QingXian, JIANG ChangSheng, HUANG GuoYong A method for fuzzy robust tracking control is proposed, which is used for solving aerospace vehicle's (ASV’s) attitude tracking problem during reentry. Based on TS fuzzy model of uncertain reentry attitude dynamics with external disturbance, the tracking errors of reference signals by attitude angle tracking are studied. The constraint of the exponential stability of the tracking error is obtained by introducing fuzzy feedforward. With the premise of stabilization control being the basis of tracking control, a fuzzy feedforward tracking controller and a H∞ fuzzy stabilization controller with pole constraints are designed. The control problem can be solved by using LMI and FLC tools of Matlab. The simulation results demonstrate the effectivity of the proposed method. 2007 Vol. 20 (6): 732-739 [Abstract] ( 271 ) [HTML 1KB] [ PDF 449KB] ( 401 ) 740 Solution to Large Scale Extraction of Social Relations of Persons Based on Web YAO CongLei, DI Nan Web information about social relations of persons is an important type of information on the Web. A lightweight method for extracting largescale information of social relations of persons is proposed. The minimum descriptive patterns which are used to describe the social relations in web pages are mined from the web with the help of the simulated annealing method. The descriptive patterns are also used to extract more social relations of persons from the web by the redundancy of the web. Six types of social relations are defined to test the proposed method, and each type of the relations is extracted from a specified person name list, which is created from the web. The experimental result shows the average precision and recall of the proposed method are 84.79% and 81.69% respectively. 2007 Vol. 20 (6): 740-744 [Abstract] ( 262 ) [HTML 1KB] [ PDF 323KB] ( 1115 ) 745 Solutions of Nonlinear Multilevel Programming Based on Particle Swarm Optimization ZHANG GuoFu, JIANG JianGuo, QI MeiBin, SU ZhaoPin An optimization algorithm for nonlinear multilevel programming problems is presented based on the analysis of the standard particle swarm optimization. The search for StackelbergNash equilibrium of nonlinear multilevel programming problems is implemented. The dynamic region is used to search the whole solution space, therefore, the algorithm has good performance to achieve the global convergence. An adaptive disturbance factor is adopted to make swarms jump out of local optimums, and a constrained fitness value is added to ensure the feasibility of the solutions. The effectiveness of the algorithm has been proved by experiments. 2007 Vol. 20 (6): 745-750 [Abstract] ( 258 ) [HTML 1KB] [ PDF 357KB] ( 411 ) 751 Expected Distribution Discriminant Analysis Based on Similarity of Sample Distribution GUO ZhiBo , YANG JingYu , ZHENG YuJie , YAN YunYang Principal component analysis (PCA) and linear discriminant analysis (LDA) are two kinds of popular feature extraction methods for pattern recognition. A new method, expected distribution discriminant analysis (EDDA), is proposed based on the similarity of sample distribution after some disadvantages of PCA and LDA are indicated. The distribution of extracted features is mostly close to the expected distribution such as idealized distribution by using EDDA. Based on EDDA, the small sample size problem (SSSP) does not occur any more. The dimension of discrimination feature is very low and the recognition performance is enhanced. Some experimental results on ORL and Yale face database demonstrate that the proposed method has higher recognition rate than PCA and LDA. 2007 Vol. 20 (6): 751-756 [Abstract] ( 284 ) [HTML 1KB] [ PDF 372KB] ( 458 ) 757 A Web Stream Algorithm for Mining Web Access Patterns GUO JianKui, HUANG ZhenHua, RUAN BeiJun, ZHU YangYong Discovering interesting web access patterns from web logs is a web usagemining problem with many practical applications. Some conventional algorithms, such as GSP, HPrefix and WAPmine have low efficiency on low support thresholds. An algorithm based on the topdown manner is proposed for mining web access pattern. Instead of stubbornly building intermediate data for each step of mining process, it selectively builds intermediate data according to the features of current area. The experimental results on various datasets show that the proposed algorithm has better performance than WAPmine. 2007 Vol. 20 (6): 757-762 [Abstract] ( 291 ) [HTML 1KB] [ PDF 474KB] ( 537 ) 763 Analysis and Extraction of Fingerprint Minutiae Based on Improved GPL Principal Curve Algorithm ZHANG HongYun, MIAO DuoQian, FU WenJie Based on the analysis of GPL principal curve algorithm and the specialties of fingerprint, the principal curve is used to reflect the structure of fingerprint. The original GPL principal curve algorithm is improved to get better skeletonization of fingerprint and the minutiae extraction based on the principal curves is applied. The experimental results show that the improved principal curve algorithm is more efficient and its skeletonization is much better than those of the original ones. The accuracy of minutiae extraction based on the principal curves is also higher than some traditional methods, which has higher value in application. 2007 Vol. 20 (6): 763-769 [Abstract] ( 332 ) [HTML 1KB] [ PDF 774KB] ( 507 ) 770 Affine Gaussian Descriptors and Their Application in Pattern Recognition LIU YiShu Finding efficient features which are invariant under affine transformation is the key to recognizing the images captured at different angles. Affine invariants, such as affine Gaussian descriptors, are derived in this paper. The covariance matrix of the given image is computed at first. Then its eigenvalues and eigenvectors are calculated, by which a set of ellipses with the same center are generated. Finally, affine Gaussian descriptors are derived based on 2dimensional Gaussian function and image compactification theory. Numerical experiments are carried out and the results show that the performance of affine Gaussian descriptors is high. 2007 Vol. 20 (6): 770-775 [Abstract] ( 237 ) [HTML 1KB] [ PDF 529KB] ( 582 ) 776 Image Recognition Based on MultiScale Geometric Analysis and Kernel Matching Pursuit GOU ShuiPing, JIAO LiCheng A method for image feature extraction and recognition is proposed. Abundant contour feature information of the image is expressed by Contourlet transform while flatness feature and texture of the image are described by Brushlet transform in the multiscale geometric analysis. The two types of feature information make up feature matrix. And fuzzy Cmean clustering is selected to compress the feature matrix and gain distributed datasets. Then object images are recognized with kernel matching pursuit classifier. The proposed method describes variant image information with different characterization transform then keeps useful original image information. Compared with Contourlet transform and Brushlet transform, the proposed method obtains higher accurate recognition rate with shorter running time. 2007 Vol. 20 (6): 776-781 [Abstract] ( 302 ) [HTML 1KB] [ PDF 847KB] ( 665 ) 782 Color Texture Image Segmentation Using Chromatic Statistical Landscape Features CHEN YunWen , XU CunLu Texture analysis is a nodus in computer vision and pattern recognition. A novel approach for color texture segmentation is proposed, called chromatic statistical landscape features (CSLF). An equiangular mapping process is used to transform an HSI space image into some pseudogray images, which can be regarded as the surfaces of the corresponding 3D landscapes. Variable horizontal planes are used to slice them. The statistic features are extracted from the set of solids which are introduced by the slicing. A supervised estimation algorithm based on Mahalanobis distance is employed for segmentation. Experimental results obtained on VisTex dataset, synthetic image and remote sensing image are presented and evaluated, which demonstrate the feasibility of the proposed method. 2007 Vol. 20 (6): 782-787 [Abstract] ( 303 ) [HTML 1KB] [ PDF 857KB] ( 664 ) 788 An Improved Binary Particle Swarm Optimizer XU YiChun, XIAO RenBin To solve 01 integer programming, an improved optimizer is proposed based on Kenney and Eberhart's binary particle swarm optimizer (BPSO). In the proposed optimizer (IBPSO) the probability calculation is simplified. The probability of the position vector depends on the values of the global best position and the private best position, and the current position does not affect the next position any more. On the test functions of De Jong, IBPSO performs better than BPSO. The results of IBPSO on the knapsack problem show IBPSO has higher convergence speed than the genetic algorithm. 2007 Vol. 20 (6): 788-793 [Abstract] ( 372 ) [HTML 1KB] [ PDF 549KB] ( 1067 ) Researches and Applications 794 Gait Recognition Based on Shape Context Descriptor CHEN Shi , MA TianJun , HUANG WanHong , GAO YouXing A method for gait recognition using human silhouette analysis is proposed. For each image sequence, the binary silhouettes of a walking figure are firstly obtained by background subtraction. An intensity difference algorithm over all color channels is introduced for shadow removal. The shape context descriptors are then employed in silhouette analysis for providing the shape feature. Modified Hausdorff distance gives an idea of the similarity measure between feature vectors of silhouettes. A set of key stances that occur during the walk cycle of an individual is chosen. Combined with window shift technology, distances of silhouettes between key stance sets are computed for subject classification and recognition. The proposed approach is applied to CASIA gait database and Soton database. The correct classification rates of 91.25% and 86.97% are achieved respectively, which illustrate that the proposed method outperforms the existing methods. Experimental results also indicate that the recognition rate maximizes when the number of sample points is 200. 2007 Vol. 20 (6): 794-799 [Abstract] ( 338 ) [HTML 1KB] [ PDF 709KB] ( 527 ) 800 A Partly Input-Adjusting Neural Network and Its Application in Nonlinear Data Reconstruction ZHAO Zhong-Gai, LIU Fei The reconstruction problem of missing data cannot be solved by the existing linear data methods in nonlinear industrial process. In order to realize reconstruction of nonlinear data, a novel neural network named partly input-adjusting neural network is proposed, whose missing variables are selected as the inputs. Different from the conventional network, the weights and threshold of the novel network have already been obtained by other network training. By back-propagation algorithm, the reconstruction is achieved. The inputs of the network are adjusted by backpropagation algorithm, then the reconstruction is achieved and the training is completed. The simulation result proves the validity of the proposed method. 2007 Vol. 20 (6): 800-804 [Abstract] ( 271 ) [HTML 1KB] [ PDF 639KB] ( 801 ) 805 Image Segmentation Based on Subsethood Measure Theory WU ChengMao In this paper, an image segmentation method based on subsethood measure theory of fuzzy set is proposed. Because of the subjective of human vision and the uncertainty of image structure, fuzzy techniques are used in image segmentation. Firstly, the subsethood measure theory of fuzzy set is introduced. Then the new criteria function of image segmentation threshold choosing is defined on the basis of fuzzy subsethood theory. Finally, the optimal method for choosing parameters of fuzzy membership is presented based on mutual information and Chaos theory. Experimental results show that the image thresholding segmentation method based on subsethood measure theory is feasible, and its segmentation performance is obviously better than that of thresholding method based on fuzzy entropy or similarity measure. 2007 Vol. 20 (6): 805-814 [Abstract] ( 287 ) [HTML 1KB] [ PDF 2166KB] ( 482 ) 815 A Multi-Valued Attribute and Multi-Labeled Data Decision Tree Algorithm LI Hong, CHEN Song-Qiao, ZHAO Rui, GUO Yue-Jian Multi-valued and multi-labeled classifier (MMC) and multi-valued and multi-labeled decision tree (MMDT) are two existing decision tree algorithms for dealing with multi-valued and multi-labeled data . Based on the two algorithms, formula sim3 is put forward to calculate the similarity between two label sets. By amending the measuring formula of samebased similarity of labelsets in MMDT, a new decision tree algorithm, similarity of same and consistent in constructing same in predicting (SCC_SP) is proposed with comprehensive consideration of both similarity and appropriateness of the label set. Results of contrast experiments with the same prediction mechanism show that SCC_SP has higher accuracy rate than MMDT. 2007 Vol. 20 (6): 815-820 [Abstract] ( 275 ) [HTML 1KB] [ PDF 323KB] ( 541 ) 821 A Color Quantization Algorithm Based on Human Visual Perception SHEN Xiang-Jun, WANG Zeng-Fu A color quantization algorithm is proposed in this paper. By this algorithm, an image is classified as edge, smooth and texture regions, and different weight strategies are assigned to them based on different degrees of perceptions. Thus, the relatively important perceptual regions, such as edge and smooth regions are strengthened and the relatively unimportant ones are weakened, such as complex texture regions. Moreover, to reach a compromise between color quantization results and time performance, the cellular color decomposition algorithm which fixes the V value is improved and the quantization algorithm is fulfilled, which could decompose the whole color space adaptively. The algorithm reduces the error of color quantization while the time performance is improved a little. The experimental results show that good quantization results are obtained by using a few colors. The proposed algorithm is especially suitable for contentbased image retrieval. 2007 Vol. 20 (6): 821-826 [Abstract] ( 323 ) [HTML 1KB] [ PDF 1212KB] ( 417 ) 827 An Automatic Keyword Extraction of Chinese Document Algorithm Based on Complex Network Features ZHAO Peng , CAI QingSheng ,WANG QingYi , GENG HuanTong Automatic keyword extraction is one of the most important techniques in natural language processing. In this paper, features of complex networks composed of Chinese are studied. A novel automatic keyword extraction algorithm for Chinese document is proposed which is based on the features of the complex networks according to the small world structure in language networks and the theoretical achievements in complex networks. It extracts keyword based on the feature values of the word nodes in a documental language network. Experimental results show the proposed algorithm obtains higher average precision compared with the keyword extraction algorithm based on TFIDF. 2007 Vol. 20 (6): 827-831 [Abstract] ( 314 ) [HTML 1KB] [ PDF 394KB] ( 851 ) 832 A Balance Principle Based SelfOrganizing Method for Task Allocation in MultiRobot System DONG YanBin, JIANG JingPing, HE Yan To overcome the problems of uncertainty in model and the difficulties in optimizing the quality function in control process of multiple robotic system, a method for decision control based on balance principle is proposed, which is put forward according to the balanced phenomena in nature. Furthermore, an algorithm of adjusting the strategies for selecting task is designed by applying the proposed method to the problem of task allocation. Simulation result demonstrates the validity of this algorithm on selforganizing task allocation in a team of robots. 2007 Vol. 20 (6): 832-837 [Abstract] ( 367 ) [HTML 1KB] [ PDF 366KB] ( 602 ) 838 Mining Web User Browsing Patterns in Fuzzy Environment WU Rui In this paper, the concepts of weighted support and preference are proposed to reflect user interests accurately. Linguistic evaluations on a web page given by experts are characterized as fuzzy linguistic variables. These fuzzy linguistic variables can be transformed as an importance weight of a web page by fuzzy simulation. To avoid loss of important user browsing information, FLAAT (frequent link and access tree) is built to save all user browsing information. Then user preferred browsing patterns can be mined from the FLAAT. In addition, time duration on web page is another important factor reflecting user interests and preference, which is also denoted by a corresponding fuzzy linguistic variable. Experimental results show the gained user preferred patterns with fuzzy time duration reflect user interests and preference more accurately. 2007 Vol. 20 (6): 838-842 [Abstract] ( 232 ) [HTML 1KB] [ PDF 406KB] ( 501 ) 843 AttributeOriented Induction Algorithm Based on Quantitative Extended Concept Lattice WANG DeXing , HU XueGang , L IU XiaoPing , HUANG DongMei In knowledge discovery in databases (KDD), users show much interest in highlevel, general and reductive information. Attribute oriented induction (AOI), which generally takes the statistical information from original data into account, has been commonly used in data reduction. However, attributeoriented algorithm based on quantitative concept lattice can finish induction with multilevel and multiattribute by using concept ascension according to the Hasse diagram of the quantitative extended concept lattice. Compared with AOI, the generalization path of the proposed algorithm is not unique. The proper generalization paths and thresholds on the Hasse diagram of quantitative extended concept lattice could be found easily. The required reasonable results are gotten, and different granular knowledge is provided for users. 2007 Vol. 20 (6): 843-848 [Abstract] ( 269 ) [HTML 1KB] [ PDF 406KB] ( 424 ) 849 UKF Algorithm Based on MEstimators and Its Application in Motion Estimation ZHOU LuPing , WANG ZhiLing , CHEN ZongHai To solve the problems on nonlinear of motion model and robustness of motion estimation, a robust unscented kalman filter (UKF) method called MUKF is proposed, which is combined with the principle of the equivalent weight of Mestimate. In this method, the UKF is used to get the initial estimation of the parameters of motion. Then more accurate estimations are obtained by MUKE. The combination of Mestimators and UKF not only solves the nonlinear problems but also conquers the inflation of outliers, which has greatly improved the robustness of estimation. Finally, experimental results demonstrate the validity of the proposed method. 2007 Vol. 20 (6): 849-854 [Abstract] ( 265 ) [HTML 1KB] [ PDF 828KB] ( 374 ) 855 A Fast and Applicable Detection of Scattered Chinese Italic Characters XIA Yong , XIAO BaiHua , WANG ChunHeng , DAI RuWei Various algorithms of detecting italic characters are introduced and analyzed in detail. An algorithm of vertical and horizontal weighted normalized comparison is presented, which can find Chinese italic characters scattered in the documents rapidly by enlarging the differences of image features between normal and italic characters. Three collections, including character, string and document, are used to evaluate the algorithm. Various algorithms of detecting italic characters are tested and compared based on these collections. The experimental results demonstrate the proposed method is effective and applicable. 2007 Vol. 20 (6): 855-860 [Abstract] ( 250 ) [HTML 1KB] [ PDF 472KB] ( 553 ) 861 Research on PolicyBased Decision Support in CRM WU XiaoDong, CHEN Chun Aiming at customer maintenance and detainment items of CRM business in operation, a solution of policybased decision support is proposed. Firstly, the concept of policybased principium is described. Then, the meaning of policybased decisionsupporting application in CRM architecture is illustrated, and its working mechanism in CRM system is studied including policycase modeling, policy database regulation and correlative policycase matching algorithm. Based on the research, a system framework of policybased decision support solution is presented, and an application case is described in detail combined with some actual business procedures. 2007 Vol. 20 (6): 861-866 [Abstract] ( 253 ) [HTML 1KB] [ PDF 570KB] ( 417 )

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