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....

2006 Vol.19 Issue.4, Published 2006-08-25 Papers and Reports    Researches and Applications        Papers and Reports 433 Semantic Analyses for Formulas in Rough Logic and Reasoning Study Based on the Semantic Analysis YAN Lin , WANG QuanRui , LIU Yan In the first place, a discussion is made to construct a kind of formulas called nary formulas in an approximate space of rough set theory. These formulas are an expansion of the formulas in Pawlak rough logic, so that the domains of the nary formulas are extended from subsets of U to subsets of Un(=U×U×…×U) . Then based on Pawlak rough logic, five logical values are efined for the nary formulas, and through these logical value operations the study about rough logical reasoning in semantics is discussed. Some properties indicate that some forms of logical reasoning in classical logic are also true in rough logic for some rough logical values. However, because 5value rough logic is different from classical 2value logic, some new properties are naturally obtained. 2006 Vol. 19 (4): 433-438 [Abstract] ( 286 ) [HTML 1KB] [ PDF 312KB] ( 412 ) 439 Dynamic Fuzzy QLearning and Its RealTime Application in Embedded System LU YongKui , XU Min, LI YongXin, DU HuaSheng, WU YueHua, YANG Jie A new dynamic fuzzy Qlearning (DFQL) method is presented in this paper which is capable of tuning fuzzy inference systems (FIS) online. In DFQL system, the generation of continuous actions depends upon a discrete number of actions of every fuzzy rule and the vector of firing strengths of fuzzy rule. In order to explore the set of possible actions and acquire experiences through the reinforcement signals, the actions are selected using an explorationexploitation strategy based on the expended greedy algorithm. A function Q that gives the action quality with eligibility trace and meta learning rule is used here to speed up learning. εcompleteness of fuzzy rules criterion and temporaldifference (TD) error criterion are considered for rule generation. The DFQL approach has been applied to a realtime control caterpillar robot for the wall following task. Experimental results and comparative studies with the fuzzy Qlearning and continuousaction Qlearning in the wallfollowing task of mobile robots demonstrate that the proposed DFQL method is superior. 2006 Vol. 19 (4): 439-444 [Abstract] ( 409 ) [HTML 1KB] [ PDF 622KB] ( 587 ) 445 A Dependency Analysis Based Algorithm for Learning Bayesian Networks HU XueGang, HU ChunLing Bayesian network is a powerful knowledge representation and reasoning tool under uncertain conditions. Current algorithms for learning Bayesian networks structures are inefficient to a certain degree. Therefore,an efficient and reliable algorithm, ISOR, is proposed in this paper. Firstly, all the potential edges of the underlying network are produced by the maximum weight spanning tree algorithm and heuristic cut  set searching algorithm. Then, methods based on identifying colliders and scoring  search methods are integrated to orient all the edges in the network. Finally, redundant edges in the network are removed. Compared with other current algorithms based on dependency analysis, the proposed algorithm greatly reduces the number and the order of conditional independence tests. Algorithm analysis and experimental results on Alarm network show algorithm ISOR has good performance. 2006 Vol. 19 (4): 445-449 [Abstract] ( 296 ) [HTML 1KB] [ PDF 310KB] ( 669 ) 450 HighOrder Iterative Learning Control of NonLinear System with Disturbance LI HongSheng The basic idea of iterative learning control (ILC) is to make use of the repetitiveness within the system to achieve a better performance. Highorder ILC convergence sufficient condition is provided for a non  linear time varying dynamic system with uncertainties and disturbances. Convergence speed of the typical PD highorder ILC scheme is also discussed in comparison with Dtype ILC algorithm. Simulation examples are provided to illustrate that highorder ILC scheme can be better than Dtype ILC in terms of convergence speed and the convergence is guaranteed if the proposed conditions are satisfied and items of uncertainties and disturbances are bounded. 2006 Vol. 19 (4): 450-454 [Abstract] ( 275 ) [HTML 1KB] [ PDF 519KB] ( 393 ) 455 Objects Detection and Classification Based onGabor Features and Enhanced Fisher Discriminant Model HE Yi, YANG Xin An approach for detection and classification of objects based on Gabor features and enhanced fisher discriminant model (EFM) is presented in this paper. Decomposed by Gabor filters, the dimensions of Gabor features of object images and models are very large. Principal component analysis (PCA) is used to extract the master components and reduce dimensions of Gabor features. Whether there are vehicle objects or not is primarily justified by the magnitude of the Gabor features. If candidate object is detected, EFM is carried out to compare its features to those of models to determine which one it belongs to-vehicles or back ground. The experiments prove the proposed arithmetic can get good results while reducing the feature dimensions. Furthermore, arithmetics for determining vehicle's number and positions are also discussed. And the experimental results also validate their feasibility. 2006 Vol. 19 (4): 455-461 [Abstract] ( 261 ) [HTML 1KB] [ PDF 581KB] ( 623 ) 462 An Invariance Algorithm of Synergetic Pattern Recognition SHAO Jing, GAO Jun, XU XiaoHong A synergetic invariance algorithm is proposed in this paper. Through the influential dynamic theory of synergetic pattern recognition, the parameters of the affine transform are gotten. The right pattern can be gotten by the dynamic evolvement of the order parameters and the autoadaptive or the nationalization of the prototype to testing patterns. This method avoids the processing of frequency field utilized by the Fourier transform. And it is similar to the recognition of human being. The validity and robustness of the algorithm are demonstrated by the experiments. 2006 Vol. 19 (4): 462-468 [Abstract] ( 299 ) [HTML 1KB] [ PDF 831KB] ( 438 ) 469 Extended Tree Augmented Naive Bayesian Classifier LI XuSheng, GUO YaoHuang Tree Augmented Naive Bayesian Classifier (TAN) often outperforms Naive Bayesian, yet at the same time maintains the computational simplicity and robustness that characterize Naive Bayesian. But TAN often requires a prior discretization of continuous variables. It is important to investigate mixedmode data, in order to represent data distributions well and avoid the problem of information loss. In this paper, the maximum likelihood function of hybrid data is deduced, and a new classifier called Extended Tree Augmented Naive Bayesian Classifier (ETAN) is put forward. The proposed classifier breaks through the restriction that continuous variables must be discretized, and it can deal with hybrid variables in the framework of TAN. Experiments show that this classifier has a good accuracy of classification. 2006 Vol. 19 (4): 469-474 [Abstract] ( 346 ) [HTML 1KB] [ PDF 335KB] ( 1092 ) 475 Pareto Archive MultiObjective Particle Swarm Optimization LEI DeMing, WU ZhiMing Pareto archive multiobjective particle swarm optimization (PAMOPSO) is designed, in which the improved method of strength Pareto evolutionary algorithm 2 (SPEA2) is used to maintain external archive and the global best location for each particle is selected in the procedure of archive maintenance. PAMOPSO is applied to five test instances and compared with other three multiobjective optimization algorithms. The computational results demonstrate that PAMOPSO has good performance in multiobjective optimization. 2006 Vol. 19 (4): 475-480 [Abstract] ( 408 ) [HTML 1KB] [ PDF 440KB] ( 681 ) 481 The Constructional Research of Fuzzy Entropy of Vague Sets ZHU LiuBing, YANG Bin, CHEN JiDong In this paper, three existing constructional methods of fuzzy entropy of vague sets are analyzed and their deficiencies are pointed out. A new method is proposed through analyzing the basic properties of fuzzydegree of vague sets. In addition, a constructional method of probability fuzzy entropy of vague sets is also provided. 2006 Vol. 19 (4): 481-484 [Abstract] ( 329 ) [HTML 1KB] [ PDF 254KB] ( 346 ) 485 The Learning and Optimizing of Markov Network Classifiers Based on Dependency Analysis WANG ShuangCheng, LIU XiHua, TANG HaiYan To decomposable probability model, it is proved that the Markov network classifier is optimal under zeroone loss. At present, the algorithms of learning the structure of Markov network classifier are inefficient and unreliable. In this paper, a new method of learning the structure of Markov network classifier is presented. The classifier structure is built by combining basic dependency relationship between variables, basic structure between nodes and the idea of dependency analysis. And Markov network classifier is optimized by removing unrelated and redundancy attribute variables to improve the ability of withstanding noise and predicting. A contrast experiment about the accuracy of classifiers is done by using artificial and real data. Experimental results show high classing accuracy of optimized Markov network classifier. 2006 Vol. 19 (4): 485-490 [Abstract] ( 230 ) [HTML 1KB] [ PDF 346KB] ( 592 ) Researches and Applications 491 Research on European Character Recognition WANG Kai, SHI GuangShun, WANG QingRen The main difference between English OCR system and other European OCR systems is character set. Therefore, European OCR system construction mainly depends on European character recognition. European character set is divided into two parts in this paper: English characters and special characters. Two key problems are considered, i.e. how to decrease the misclassification rate between English characters and special characters, and how to improve the recognition accuracy for special characters. Experimental result shows that the new system is more effective than the previous ones. Furthermore, the ideas proposed in this paper can be generalized to distinguish any similar symbols. 2006 Vol. 19 (4): 491-496 [Abstract] ( 303 ) [HTML 1KB] [ PDF 482KB] ( 477 ) 497 A Feature Matrix Similarity Measure Method and Its Application to Image Retrieval LIU YueHu, WANG Fei, LIU XiaoDong, YUAN ZeJian A novel feature matrix similarity measure method, which is suitable for two dimensional object recognition and matching, is presented. In this method, a similarrow vector is produced by comparing dynamic programming (DP) matching distances, which describe the similarity between the row of a query matrix and those of a sample matrix. And the similarrow vector is used to represent the query matrix. Then the DP matching is again performed to obtain a similarity measure. The proposed method is employed in an image retrieval system using a dominant color feature matrix representation. The experimental results show that the method is efficient. 2006 Vol. 19 (4): 497-502 [Abstract] ( 438 ) [HTML 1KB] [ PDF 676KB] ( 1773 ) 503 An Integrated Approach for Perspective Distortion Correction of SmallSquare Document Images MA YiChao, DAI RuWei, WANG ChunHeng In this paper, an integrated perspective distortion correction method is proposed aiming at the smallsquare document images acquired by the camera. The characteristics of smallsquare documents-small area, fewer words and complicated layout, boundaries of smallsquare document region extracted by color image segmentation methods as well as the text information in the document are used to correct the distortion. Experimental results show that this method can rectify the smallsquare document image effectively. 2006 Vol. 19 (4): 503-508 [Abstract] ( 238 ) [HTML 1KB] [ PDF 1082KB] ( 692 ) 509 Left Ventricle MRI Image Segmentation by Unifying Statistic Model and Curves Evolving ZHOU ZeMing,CHEN Qiang, Pheng Ann Heng , XIA DeShen An MRI image segmentation algorithm is proposed by unifying region statistic model and image gradient information. Due to cardiac deformation and blood flowing, weak edges, local gradient maximum regions and artifacts often can be found in the MRI images. The level set method which constructs stopping term with image gradient intensity cannot segment those cardiac MRI images accurately. A twostage algorithm is thus proposed to address the difficulties. Firstly, incorporating prior knowledge about the cardiac MRI and the image histogram, the populations of pixels are given. The prior probabilities of those classes and the parameters of the Gaussian distributions are estimated with Maximumlikelihood principle. With the posterior probability of pixel belonging to ROI, the velocity function of level set is constructed to search for the rough boundary of ROI. Next, another velocity function based on the gradient vector flow is designed to locate the edges accurately. The experimental results demonstrate the effectiveness of the segmentation algorithm. 2006 Vol. 19 (4): 509-514 [Abstract] ( 246 ) [HTML 1KB] [ PDF 857KB] ( 421 ) 515 A RuleBased Expert System with Multiple Knowledge Databases SONG LiangTu, LIU XianPing, BI JinYuan, ZHA JinShui In this paper, a new rulebased expert system with multiple knowledge databases is designed according to the characteristics of the domain expert knowlege and its representation.Compared with classic expert system ,the new system has a number of corresponding knowledge bases according to the subproblems of a complex problem. And each knowledge base includes a shallow knowledge base of the empirical knowledge and a deep knowledge base of the scientific knowledge. The inference engine of the system adopts twolevel inferring structure, the inference at system level and the one of subproblem, which can reduce the searching space quickly and work efficiently.The system runs well over internet. 2006 Vol. 19 (4): 515-519 [Abstract] ( 253 ) [HTML 1KB] [ PDF 359KB] ( 420 ) 520 Matching of 3D Object Based on Genetic Algorithm and Model Projection ZHANG Yu A 3D matching method based on genetic algorithm (GA) and model projection is presented, which is a process of searching for the best parameters of projective transform. Coarse to fine strategy is adopted. Firstly, based on the outermost boundary of model all the edges are matched. Then the correct initial values of parameters are obtained quickly to improve the efficiency of the algorithm. The fitness function of GA is based on 34 chamfer distance transform. Finally, label mapping is proposed to deal with the noise in distance map. Meanwhile, features of edge pixels, which are difficult to be applied directly to the distance map, are available to label map. Experimental results show that the proposed method is effective in 3D modelbased matching. 2006 Vol. 19 (4): 520-525 [Abstract] ( 328 ) [HTML 1KB] [ PDF 875KB] ( 469 ) 526 ManMade Objects Detection from Aerial Images Based on Level Set Method CAO Guo, YANG Xin A novel method for detecting manmade objects in aerial images is described. The method is based on a simplified MumfordShah model. It applies fractal error metric and additional constrainttexture edge descriptor on the image to get a preferable segmentation. Manmade objects and natural areas are optimally differentiated by evolving the partial differential equation for MumfordShah model. The method avoids selecting a threshold, which, if improperly selected, often results in great segmentation errors to separate the fractal error image. Experiments of the segmentation show that the proposed method is efficient. 2006 Vol. 19 (4): 526-530 [Abstract] ( 240 ) [HTML 1KB] [ PDF 945KB] ( 421 ) 531 Document Feature Selection Based on the Minimum Term Frequency Threshold CHEN XiaoYun, LI RongLu, HU YunFa In this paper, a novel method of feature evaluation function based on document frequency with the minimum term frequency threshold (DFn) is presented. To decrease the influence of the unrelated features on the system of text categorization, the attribute of the unrelated features is analyzed and the term frequency of the unrelated feature is commonly low. By applying minimum term frequency to filter the low frequency features, the unrelated features are obviously decreased. The experimental results validate the proposed method greatly reduces the number of the unrelated features and effectively improves the accuracy of the text categorization. The improvement to Mutual Information(MI) is very obvious, the Macroaverage F1 value based on DFn is 40% higher than that of Term Frequency, and 15～30% higher than that of Document Frequency(DF). 2006 Vol. 19 (4): 531-537 [Abstract] ( 267 ) [HTML 1KB] [ PDF 431KB] ( 732 ) 538 A New Particle Dynamical Evolutionary Algorithm for Solving Complex Constrained Optimization Problems LI KangShun, LI YuanXiang, KANG LiShan, LI BangHe In this paper a particle dynamic evolutionary algorithm(CPDEA) for solving constrained problems efficiently is presented according to the equation of particle transportation, the principle of energy minimizing and the law of entropy increasing in phase space of particles based on transportation theory. A fitness function of constrained optimization problems is defined based on the theory in which particle systems in phase space reach equilibrium from nonequilibrium. The energy of particle systems minimizes and the entropy of particle systems increases gradually in the evolving process of particles in order that all the individuals have chance to crossover and mutate. Finally all the optimal solutions are obtained quickly. In the numerical experiments, precise optimal solutions of the constrained problems are gotten by using this algorithm. Compared with traditional evolutionary algorithms, the experiments show that not only all the global solutions of complex constrained optimization problems can be solved in an easy and quick way, but also premature phenomenon can be avoided. 2006 Vol. 19 (4): 538-545 [Abstract] ( 263 ) [HTML 1KB] [ PDF 496KB] ( 521 ) 546 An Approach for Ontology Version Mapping ZHAO SiYang, ZHU Yun, ZHOU XueHai, Brendan Tierney Singledirectional mapping between ontology versions can not satisfy the requirement of data sharing and reusing. In this paper a new approach, bidirectional mapping, for mapping ontology versions is introduced. It can process mappings between two ontology versions in two directions at the same time and produce the bidirectional mapping which can relate and transform the similar elements in different ontology versions. So it can greatly improve the ability of data sharing between data resources that use different versions of the same ontology. Four elements of the bidirectional mapping between ontology versions are described in details, including mapping element, bidirectional mapping relation, bidirectional transformation and additional metadata. Finally, an example of bidirectional mapping between ontology versions is presented. 2006 Vol. 19 (4): 546-551 [Abstract] ( 235 ) [HTML 1KB] [ PDF 333KB] ( 411 ) 552 Fast Recognition of Harbor Target in Large Scale Remote Sensor Images ZHU Bing, LI JinZong, CHEN AiJun A fast recognition method of the harbor target in large gray remote sense image is presented. Multiresolution processing is used to detect big, middle and small harbor respectively, and threshold processing method is introduced to segment image into seawater and land. Then a fast method of extracting the candidate harbor region is established based on the statistic representation of block. Finally fast recognition of harbor is implemented according to its inherent feature (half close of seawater). Eighteen large images are used to test the proposed algorithm, and the result shows that harbor images with 10000 by 10000 pixels can be detected in three seconds and recognition rate is 93.9%. 2006 Vol. 19 (4): 552-556 [Abstract] ( 253 ) [HTML 1KB] [ PDF 1061KB] ( 464 ) 557 The Metamer Number Prediction Based on Improved SVM WANG DeJi, XIONG FanLun, WANG RuJing, ZHA ShiHong The relation between the temperature and the metamer is very important for the virtual plant growth model. However, it is difficult to predict it just by SVM because there are too many noises in the raw data. In this paper, a new kernel function based on the information geometry is established to overcome the high noise and nonlinear data. The relation between number of metamer and temperature can thus be gotten precisely. The method is applied to the cotton growth model. Compared with the methods of least square and SVM, the improved SVM can predict the number of metamer more precisely. 2006 Vol. 19 (4): 557-560 [Abstract] ( 240 ) [HTML 1KB] [ PDF 299KB] ( 316 )

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