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.5, Published 2007-10-25 Papers and Reports    Researches and Applications    Surveys and Reviews        Papers and Reports 583 An Orthogonal Immune Clone Particle Swarm Algorithm with Quantization for Numerical Optimization CONG Lin, JIAO LiCheng, SHA YuHeng In order to overcome prematurity and low searching speed of PSO algorithm, an orthogonal immune clone particle swarm algorithm with quantization (OICPSO/Q) is proposed according to the immune clone selection theory. An orthogonal subspace division method is presented and the orthogonal crossover strategy is used to increase the uniformity of solution. To avoid losing the optimal solution in neighborhood of individuals, a self-learning operator is presented. The global convergence of OICPSO/Q has been proved by theoretical analysis. In experiments, OICPSO/Q is tested on unconstrained benchmark problems with 20~1000 dimensions, and is compared with five methods. The effects of parameters on computational cost of the algorithm are analyzed. The results indicate that OICPSO/Q is capable of solving complex problems and preserving the diversity of population. To some extent, it avoids prematurity and improves the convergence speed. 2007 Vol. 20 (5): 583-592 [Abstract] ( 231 ) [HTML 1KB] [ PDF 637KB] ( 370 ) 593 Complexity Analysis of Partial Implication Semantics ZHOU Yi, CHEN XiaoPing Minimal model and partial implication semantics play important roles in the subfields of artificial intelligence. In this paper, the complexity issues of minimal model and partial implication semantics are analyzed when the antecedent and the consequent are literals, literal sets and formulas respectively. The results show that the complexities of minimal model and partial implication semantics increase when the antecedent and consequent become more complex. Moreover, the complexities of all these decision problems lie in the first two layers of polynomial hierarchy. 2007 Vol. 20 (5): 593-598 [Abstract] ( 339 ) [HTML 1KB] [ PDF 332KB] ( 350 ) 599 Mechanism of Granular Computing Based on Information System MENG Zu-Qiang , SHI Zhong-Zhi Algebraic system of decision logic and topological algebraic system of universe are set up based on information system. Then relation theorem of the two systems is discovered. With the two algebraic systems, granular world models are established, in which the mechanism of granular computing (GrC) is studied by analyzing concept learning and rule acquiring. In the meantime, the model provides coherent explanation of all granular computing methods, which leads to unification of set based granular calculation and decision logic based granular calculation to some degree. 2007 Vol. 20 (5): 599-605 [Abstract] ( 247 ) [HTML 1KB] [ PDF 385KB] ( 491 ) 606 A Particle Swarm Algorithm for MultiObjective Optimization Problem JIANG Hao, ZHENG JinHua, CHEN LiangJun A new multiobjective particle swarm optimization (MOPSO) based on enhanced εdominance is proposed, which keeps good diversity. A new idea named guide mutation is introduced to select global guide from the archive. Then a population updating strategy and selfadaptive mutation operation are shown to speed up convergence. Experimental results show that the proposed approach has effective and steadystate performance and is simple to implement. 2007 Vol. 20 (5): 606-611 [Abstract] ( 330 ) [HTML 1KB] [ PDF 466KB] ( 991 ) 612 A Mean Shift Tracking Algorithm Based on Texture Model NING JiFeng , WU ChengKe In Mean Shift tracking algorithm, model representation of the tracked target has great influence on the results of tracking. By analyzing image features of nine uniform texture models of Local Binary Pattern (LBP), five uniform modes related to the edge and the corner are used to represent the object model, which is called FLBP8,1, and FLBP8,1 is put into Mean Shift algorithm for object tracking. The edge is combined with its texture effectively by FLBP8,1. The key pixels of object mode are extracted automatically, and a few of them are used to represent the accurate object. Thus low computational complexity is obtained. Experimental results show that the proposed algorithm gets better performance in accuracy and speed of tracking than RGB based method under complex conditions. 2007 Vol. 20 (5): 612-618 [Abstract] ( 412 ) [HTML 1KB] [ PDF 1154KB] ( 504 ) 619 A Triangle Matrix Feature Transformation Method ZHAO Ying, LIU Hong-Xing, GAO Dun-Tang Based on the optimization of the initial basis of the feature space, the Triangle Matrix Feature Transformation method is proposed to calculate the transformation matrix 公式 in feature extraction. In this method, the number of the unknown parameters in the transformation matrix is only half of that in the currently used methods. Thus it reduces the calculation pressure a lot. This method supports various feature transformation criterions, and it is of good flexibility. 2007 Vol. 20 (5): 619-623 [Abstract] ( 226 ) [HTML 1KB] [ PDF 381KB] ( 400 ) 624 An Evaluation Method for Agent Coalition Based on D-S Evidence Theory SU Zhao-Pin, JIANG Jian-Guo, XIA Na, ZHANG Guo-Fu Agent coalition is an important cooperative manner in multi-agent system. Its performance counts for much in task accomplishment. However, the existing methods have a lack of performance evaluation on agent coalition after coalition formation. In this paper, D-S evidence theory is adopted to make uncertain and incomplete evaluation on agent coalition according to agent ability, harmonious performance, communication cost, familiarity and continuable expansibility. A two-layered evaluation method on agent coalition is proposed, and its evaluation accords with process of human thinking and judgement. The agility, validity and rationality of the method have been illustrated by the example. 2007 Vol. 20 (5): 624-629 [Abstract] ( 241 ) [HTML 1KB] [ PDF 398KB] ( 371 ) 630 Computation of Document Structural Similarity Based on PartWhole Matching MA Jun , CHEN ZhuMin , ZHAO Yan , LEI JingSheng Traditional algorithms for document structural similarity (DSS) are based on either tree edit distances or Fourier transformation. New ways for the computation of DSS are presented based on partwhole matching between the tree Q corresponding to the structural question description and the tree T corresponding to structural description of a document. A way is provided to label above trees by strings then the DSS between two trees is calculated based on string matching operations. Experimental results show the proposed algorithms are better than those based on tree edit distance in terms of recall and precision. 2007 Vol. 20 (5): 630-635 [Abstract] ( 295 ) [HTML 1KB] [ PDF 402KB] ( 555 ) 636 Concept Lattice Based DataDriven Uncertain Knowledge Acquisition WANG Yan , WANG Guo-Yin , DENG Wei-Bing Uncertain knowledge acquisition is a problem when no prior domain knowledge is available. The relationship of knowledge uncertainties among three different knowledge presentation models, i.e. decision table, decision rule, and concept lattice, is discovered through analyzing their knowledge presentation styles. A datadriven automatic uncertain knowledge acquisition algorithm based on concept lattice is developed by using this relationship. Experimental results show that this algorithm is valid for acquiring uncertain knowledge. 2007 Vol. 20 (5): 636-642 [Abstract] ( 298 ) [HTML 1KB] [ PDF 371KB] ( 437 ) 643 Image Compression Based on Data Domain Description SHE Qing-Shan, SU Hong-Ye, ZHANG Ying, CHU Jian An image compression approach is proposed based on support vector domain description (SVDD) and adaptively weighted support vector machine (w-SVM). An intensity image is divided into some non-overlapped rectangular subblocks and transformed from spatial domain to frequency domain via discrete cosine transform (DCT). On each segmented subblock, its corresponding weight function model is obtained according to the distance from each DCT coefficient to the center of the smallest enclosing hypersphere in high dimension feature space. The established model is eventually embedded into the w-SVM based image compression scheme. Simulation results show that the proposed approach is superior in prediction performance and compression effect to general SVM-based image compression algorithms. 2007 Vol. 20 (5): 643-648 [Abstract] ( 248 ) [HTML 1KB] [ PDF 972KB] ( 524 ) 649 Feature Extraction Based on K-Nearest Neighbor Decision Boundary HAO Hong-Wei, SU Rong-Wei A modified K-nearest neighbor based decision boundary analysis (KNN-DBA) method is proposed for improving the classification performance. The decision boundary is determined by K-nearest neighbor classifier which is simple and fast. The extracted feature dimensionality is not limited by class number. Experimental results on the USPS handwritten digit dataset using nearest neighbor and support vector classifiers show that the DBA method outperforms principal component analysis (PCA). 2007 Vol. 20 (5): 649-653 [Abstract] ( 285 ) [HTML 1KB] [ PDF 416KB] ( 1126 ) 654 A Chaos Quantum Immune Algorithm for Continuous Space Optimization LI Pan-Chi , LI Shi-Yong By integrating the ergodicity of chaos searching and the high efficiency of quantum computation into immune optimization, a novel chaos quantum immune algorithm for continuous space optimization is presented. In this algorithm, antibodies in colonies are coded by quantum bits and updated by quantum rotation gates. To change phase of qubit, two different chaos variables are introduced into the quantum rotation gate. The one with the relatively small amplitude performs the cloning of excellent individuals to implement the local searching, and the other one with the relatively large amplitude performs the mutation of inferior individuals to realize the global searching. The convergence of the proposed algorithm has been proved. The experimental results indicate that the algorithm remarkably improves the convergence performance and the search efficiency of the immune optimization algorithm. 2007 Vol. 20 (5): 654-660 [Abstract] ( 247 ) [HTML 1KB] [ PDF 488KB] ( 415 ) 661 An Algorithm for Mining Maximum Frequent Itemsets LI QingFen , WANG Li , ZHOU WeiLin , CHEN HuoWang Mining complete set of frequent patterns remains a key problem to the application of association rules. Up to date, the most commonly used methods are Apriori algorithm and FPTREE algorithm.In this paper, a high efficient algorithm, minimal support minimal combination algorithm (MSMCA), is proposed. It is completely different from the two existing methods. The candidate set of frequent itemsets are not produced by using MSMCA, thus the cost of computer reduces largely. In addition, a subproject, minimal support minimal combination in repeat array, is proposed in the course of studying MSMCA. 2007 Vol. 20 (5): 661-666 [Abstract] ( 326 ) [HTML 1KB] [ PDF 353KB] ( 448 ) Surveys and Reviews 667 Survey of ART-2 and Its Improvement ZHOU Xin-Ran , TENG Zhao-Sheng , LIU Xiao-Bo ART-2 based on adaptive resonance theory is a kind of self-organizing neural network and usually utilized in pattern clustering and recognition, etc. In order to satisfy some specific requirements of certain applications or to simplify the hardware implementation,some improved versions of ART-2 have been put forward in recent years. In this paper, the original ART-2 is briefly introduced, its training algorithm is firstly analyzed, and its inherent limitations are explored. The background, objects and implementation of typical improved versions are summarized and generalized, and their properties and suitabilities are remarked on. Finally, the theoretical value and some rules pointing to future application and improvement of ART-2 are shown. 2007 Vol. 20 (5): 667-674 [Abstract] ( 239 ) [HTML 1KB] [ PDF 483KB] ( 375 ) Researches and Applications 675 A Robust Iris Localization Algorithm for NonIdeal Capturing Conditions ZHANG Wen-Cong , YE Xue-Yi , LI Bin , YAO Peng , ZHUANG Zhen-Quan The existing iris localization algorithms have fine performance for the iris images captured on ideal conditions. However, the captured iris images are greatly influenced by bad conditions, such as different luminance, eyelashes, hair, and glasses frame, which make precise localization be a hard task. To improve the robustness of iris localization, a novel localization algorithm is proposed based on the radial symmetry transform, in which the radial symmetry characteristic of the pupil is fully used to realize iris localization. Experimental results show that the proposed algorithm is robust to the influence of luminance and other complex conditions. Moreover, it realizes precise localization in a real-time system. 2007 Vol. 20 (5): 675-680 [Abstract] ( 266 ) [HTML 1KB] [ PDF 785KB] ( 510 ) 681 A Sparse Least Squares Support Vector Machine Classifier LIU XiaoMao , KONG Bo , GAO JunBin , ZHANG Jun Support Vector Machine (SVM) has to solve the quadratic programming problem, while least squares support vector machine (LSSVM) only needs to deal with the linear equations. However the defect of LSSVM is the lack of sparseness. In this paper, a method named sparse least squares support vector machine classifier (SLSSVM) is presented to remedy the defect of the LSSVM. It is carried out by preextracting margin vectors using center distance ratio method as original training samples and putting those which have not been classified correctly in the first training together as new training samples. The proposed method not only remedies the defect of LSSVM, but also speeds up training and classifying. Furthermore, it can rectify the deviation of the classifier for unbalanced training data and the classifying ability is not affected. The good performance of SLSSVM is verified on several data sets. 2007 Vol. 20 (5): 681-687 [Abstract] ( 290 ) [HTML 1KB] [ PDF 369KB] ( 685 ) 688 An Adaptive MaxMin Ant Colony Algorithm SU Chang , TU Jun The structure and principle of ant colony algorithm are introduced. Its excellence and deficiency are analyzed, and several important improved models are reviewed. Based on MaxMin Ant Systems (MMAS), an adaptive improved model is put forward. To achieve adaptive adjustment of parameters and enhance the performance of the proposed algorithm, the weighting coefficient, state transferring rule and pheromone increment mode are improved. To testify the performance of the improved algorithm, numerical experiment is made and the result shows the improved algorithm is effective. 2007 Vol. 20 (5): 688-691 [Abstract] ( 279 ) [HTML 1KB] [ PDF 395KB] ( 1153 ) 692 Finger Vein Recognition Based on Wavelet Moment Fused with PCA Transform WANG KeJun, YUAN Zhi As a new kind of identity authentication technology, finger vein recognition has more merits than other biometric feature authentication system. Therefore, it has a vast application prospect. The algorithm based on wavelet combined with Principal Component Analysis (PCA) transformation and LDA transformation is proposed. It not only overcomes the disadvantage of the single feature recognition, but also solves the lowspeed problem of common template matching. Experimental results indicate that the proposed method can provide fast and accurate identification, and the results are satisfactory. 2007 Vol. 20 (5): 692-697 [Abstract] ( 331 ) [HTML 1KB] [ PDF 633KB] ( 715 ) 698 A Generalized Method for Unsupervised Text Clustering Using Finite Mixture Models ZHANG Liang, LI Min-Qiang A generalized method is presented for unsupervised text clustering. The relevance of the features to the mixture components is introduced to the mixture model as a set of latent variables. Then the model selection, feature selection and parameter estimation of the mixture model are integrated into one general framework. Experimental results on four large scale document datasets show that the proposed method achieves fine results in model selection, feature selection and clustering performance. 2007 Vol. 20 (5): 698-703 [Abstract] ( 238 ) [HTML 1KB] [ PDF 371KB] ( 593 ) 704 Fuzzy k-Plane Clustering Algorithm WANG Ying, CHEN Song-Can, ZHANG Dao-Qiang,YANG Xu-Bing A clustering algorithm named Fuzzy k-Plane Clustering (FkPC) is proposed by introducing fuzzy membership into the prevalent k-Plane Clustering (kPC). From the view of prototype selection, FkPC substitutes hyperplanes for points as the prototype, which is similar with kPC. Meanwhile, FkPC represents the membership between the points and its central hyperplanes much more clearly than kPC, due to the introduction of fuzzy membership into its objective function. Experimental results of both artificial and UCI datasets have proved the clustering validity of FkPC, and they also reveal that besides the similarity metric, the expression of prototype also plays a crucial role in clustering. 2007 Vol. 20 (5): 704-710 [Abstract] ( 482 ) [HTML 1KB] [ PDF 593KB] ( 886 ) 711 An Improved Method of D-S Evidence Theory Based on Pretreatment Mode SHE Er-Yong , WANG Run-Sheng , XU Xue-Wen Dempster-Shafer (D-S) theory of evidence is one of the main methods of decision fusion, but typical D-S theory is sensitive to high conflict evidences. A method based on the pretreatment mode is presented. The basic probability assignments of conflict focal elements are partly transferred to the union of focal elements before the Dempster combination rule is used. The combination order is determined based on the conflicted measurement. Therefore, D-S theory could deal with the cases with high conflict evidences since conflict information is translated into unknown knowledge representation. 2007 Vol. 20 (5): 711-715 [Abstract] ( 265 ) [HTML 1KB] [ PDF 293KB] ( 462 ) 716 A ShrinkingClustering Method for High Dimensional Data Using Flexible Size Grid ZHANG JianYe , PAN Quan , LIANG JianHai A shrinkingclustering method using flexible size grid is proposed to solve the clustering problem of high dimensional data in data mining. The data bins are arranged according to their density span, and the data points are moved along the direction of the density gradient. Thus the condensed and widelyseparated clusters are generated. Then the connected components of dense cells are detected using a sequence of grids with flexible size. Finally, the best clustering result is obtained when the borderline does not change again. The simulation result shows that the method could detect clusters effectively and efficiently in both low and high dimensional data. 2007 Vol. 20 (5): 716-721 [Abstract] ( 239 ) [HTML 1KB] [ PDF 532KB] ( 470 ) 722 Infrared Target Tracking Method Based on Hierarchical Matching and Background Patching HUANG Fei, LI DeHua In infrared image sequence, it is difficult to describe the moving change of target using model on account of its anomalous moving. An infrared target precise tracking method is proposed based on hierarchical matching and background patching. The sequence image difference, which is caused by anomalous tingling of the camera, is wiped off by using background patching method. The tracking precision is improved greatly using the integration of hierarchical matching and background patching. The experimental results show the proposed method is efficient and robust for the infrared target tracking in the clutter background. 2007 Vol. 20 (5): 722-726 [Abstract] ( 249 ) [HTML 1KB] [ PDF 975KB] ( 456 )

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