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.3, Published 2007-06-25 Papers and Reports    Researches and Applications    Surveys and Reviews        Papers and Reports 295 Minimum Attribute Reduction Algorithm Based on Binary Particle Swarm Optimization YE DongYi, LIAO JianKun Based on binary particle swarm optimization, a minimum attribute reduction algorithm for a decision table is presented. A proper fitness function is defined. Thus, the minimum attribute reduction problem is equivalently transformed into a binary combinatorial optimization problem without additional nonlinear constraints. The concept of a seed particle is introduced with its protection strategy. Finally, an improved binary particle swarm optimization algorithm is proposed to solve the transformed problem. Experimental results show the effectiveness of the presented algorithm. 2007 Vol. 20 (3): 295-300 [Abstract] ( 242 ) [HTML 1KB] [ PDF 336KB] ( 465 ) 301 A Fast MultiClass Support Vector Machine LI JianWu, LU Yao A binary encoding based fast multiclass support vector machine(SVM) is introduced. How to avoid the uneven class size of each SVM in the multiclassification system is discussed based on the encoding method. Then the strategy of searching the optimal division of different classes is proposed. Thus, with little loss of accuracy the system has a higher classification speed than the traditional ones. Therefore, the classifier is suitable for real time or online systems. Finally, the introduced classification system is evaluated by experiments. 2007 Vol. 20 (3): 301-307 [Abstract] ( 246 ) [HTML 1KB] [ PDF 353KB] ( 628 ) 308 Interactive MultiAgent Genetic Algorithm HUANG YongQing , HAO GuoSheng , LIANG ChangYong , YANG ShanLin A interactive multiagent genetic algorithm (IMAGA) is proposed. Every agent fixed on a latticepoint in IMAGA interoperates with their neighbors, and the optimal one carries out selflearning to increase the energy. Hence the abilities of global convergence and local search of the algorithm are improved. In every generation, users only need to select the interested individuals instead of evaluating every individual, which simplifies the users' evaluation. The simulations of function optimization and fashion design shows that the proposed algorithm with higher convergence velocity reduces the total times of users' evaluation so as to alleviate user fatigue. 2007 Vol. 20 (3): 308-312 [Abstract] ( 245 ) [HTML 1KB] [ PDF 663KB] ( 503 ) 313 Contour Extraction of Moving Objects Based on PCA/Snake Mixture Model LI ChunMing , LI YuShan , ZHANG DaPu , LIU Yang Based on a Principal Component Analysis (PCA)/Snake mixture model, an algorithm for contour extraction of moving objects is proposed. Firstly, the moving objects in the image sequences are detected based on an extended PCA model and eigenforground. The extended PCA model can be used directly to process image sequences. Secondly, the precision contour of moving object is extracted based on a Snake model according to the detection result of the moving object. The experimental results show that the proposed algorithm extracts the moving object precisely with a high speed. 2007 Vol. 20 (3): 313-318 [Abstract] ( 233 ) [HTML 1KB] [ PDF 1460KB] ( 620 ) 319 General Ant Colony Algorithm and Its Applications in Robot Formation ZHANG Ying , CHEN XueBo A general ant colony algorithm is proposed. In this algorithm, ants are supposed to be divided into several swarms and each swarm possesses its own nest and food at different places. During a preset period, ants from the same swarm increase the strength of pheromone on the shortest path which they have found between a food source and the nest. In the meanwhile, they adjust the strength of pheromone on other paths to zero. Each swarm moves on its own path and collision never occurs. When environment changes, the swarms crawl to their food along the increased pheromone. The general ant algorithm combined with the affine transformation is applied to the robot formation, and the results are effective. 2007 Vol. 20 (3): 319-324 [Abstract] ( 312 ) [HTML 1KB] [ PDF 410KB] ( 471 ) 325 Two Point Matching Algorithms Based on Point Spatial Features TAN ZhiGuo, SUN JiXiang The fundamental theory and the antinoise problem of the traditional eigenvector approach (EA) are discussed and analyzed. Two matching algorithms are proposed, namely weighted eigenvector approach (WEA) and sorting approach (SA). WEA decomposes the intraset distance matrices of point sets and gets the feature vectors of the points. Then, the feature vectors are weighted by the eigenvalues of matrices. The algorithm gets the matching map by comparing the similarity of the weighted feature vectors. Without the decomposition of matrices, SA acquires the characteristics of the point by sorting the distance matrices, and obtains the matching in the same way above. The two algorithms solve the choosing problem of Gauss parameter and have better antinoise ability than EA. Experimental results show the practicability of the algorithm and the better performance than that of EA. 2007 Vol. 20 (3): 325-330 [Abstract] ( 319 ) [HTML 1KB] [ PDF 1452KB] ( 442 ) 331 Constructing Feedforward Neural Networks to Approximate Polynomial Functions CAO FeiLong , ZHANG YongQuan , PAN Xing It is investigated that multivariate polynomial functions with n order are approximated by feedforward neural networks with three layers. Firstly, for a given polynomial function with n order, a feedforward neural network with three layers is designed by a constructive method to approximate the polynomial with any degree of accuracy. The number of hidden layer nodes of the constructed network only depends on the order and dimension of approximated polynomial. Then, an algorithm to realize the approximation is given. Finally, two numerical examples are given for further illustration of the results. The obtained results have a guidance signification to construct feedforward neural networks with three layers to approximate the class of polynomial functions and realize the approximation. 2007 Vol. 20 (3): 331-335 [Abstract] ( 266 ) [HTML 1KB] [ PDF 550KB] ( 454 ) 336 Identification of Outlier Patterns in Multivariate Time Series WENG XiaoQing , SHEN JunYi Multivariate time series (MTS) is widely available in many fields including finance, medicine, science and engineering. An approach for identifying outlier patterns in MTS is proposed. By using bottomup segmentation algorithm, MTS is divided into nonoverlapping subsequences. An extended Frobenius norm is used to compare the similarity between two MTS subsequences. Kmeans algorithm is employed to cluster MTS subsequences into some classes. According to the definitions of outlier patterns, the outlier patterns in MTS can be identified from the classes. Experiments are performed on two realworld datasets: stock market dataset and brain computer interface dataset. The experimental results show the effectiveness of the algorithm. 2007 Vol. 20 (3): 336-342 [Abstract] ( 282 ) [HTML 1KB] [ PDF 837KB] ( 484 ) 343 Fuzzy MultiAttribute Group Decision Making Methods Based on Structured Element LIU HaiTao, GUO SiZong The purpose of this paper is to introduce the structured element theory into fuzzy multiattribute decision making, and to simplify the complex operations of traditional fuzzy decision making. The compromised group decision making method is one of the most commonly methods in classical multiattribute group decision making. According to the principle of this method , two kinds of fuzzy compromised group decision making methods based on the structured element theory are put forward. The decision making is carried out by using the example from Transformation Group of Monotone Functions with Same Monotonic Formal on [-1,1] and Operations of Fuzzy Numbers by Sizong Guo. Finally, the proposed methods are compared with the traditional decision making methods. The proposed methods have a reference value to the fuzzy multiattribute group decision making problems. 2007 Vol. 20 (3): 343-348 [Abstract] ( 274 ) [HTML 1KB] [ PDF 336KB] ( 436 ) Surveys and Reviews 349 Survey of Particle Swarm Optimization Algorithm NI QingJian, XING HanCheng, ZHANG ZhiZheng, WANG ZhenZhen, WEN JuFeng The particle swarm optimization (PSO) algorithm is an evolutionary algorithm that simulates the mechanism of biological swarm social behavior. The models of bird flocking and swarm actions are firstly introduced, and the fundamental characteristics and the working mechanisms of PSO algorithm are also analyzed. Then the recent progress in theory of PSO algorithm is reviewed, which are related to the improvement of PSO algorithm, the parameter selection in PSO algorithm, the convergence features of PSO algorithm, and the merging mechanism to other metaheuristic optimization algorithms. In addition, several typical application areas of PSO algorithm are surveyed respectively, which include continuous function optimization, neural network training, optimization of power system and optimization in electromagnetics. Finally, some suggestions on future trends and existing problems related to PSO algorithm are discussed and concluded. 2007 Vol. 20 (3): 349-357 [Abstract] ( 313 ) [HTML 1KB] [ PDF 472KB] ( 937 ) Researches and Applications 358 Incremental Learning Algorithm of a Modified MinimumDistance Classifier SANG Nong, ZHANG Rong, ZHANG TianXu An incremental learning algorithm of a modified minimumdistance classifier is proposed to eliminate the inter disturbance of the classifier during the incremental learning process. It enables the classifier to remember the old knowledge and learn the new one at the same time. Incremental learning requires the modification of the classifier structure, and certain number of old samples must be reserved to help to review old knowledge while learning the new one. In terms of normal distributed sample sets, a new filtering algorithm is proposed. A few samples are preserved which are representative and greatly reduced the cost of storing as well as retraining. Experimental results show that the algorithm gives high rate of recognition correctness. It makes the old samples remain high rate of recognition correctness while the new samples are also effectively recognized with less storing space. 2007 Vol. 20 (3): 358-364 [Abstract] ( 276 ) [HTML 1KB] [ PDF 374KB] ( 435 ) 365 Gait Recognition Based on Independent Component Analysis and Information Fusion LU JiWen, ZHANG ErHu, XUE YanXue A gait recognition method is proposed based on independent component analysis/support vector machine (ICA/SVM) and information fusion from multiple views. Human silhouette extraction is obtained by background subtraction and shadow elimination. Wavelet descriptor is applied to describe these silhouettes. Then, independent component analysis is employed to compress and extract their features, and gait classification is performed by support vector machine. The gait features from multiple views are fused, and recognition is finished. The method is evaluated on the National Laboratory of Pattern Recognition (NLPR) and Xi’an University of Technology (XAUT) gait database and the correct recognition rate is relatively high. The experimental results show that the proposed method has good recognition performance. 2007 Vol. 20 (3): 365-370 [Abstract] ( 259 ) [HTML 1KB] [ PDF 962KB] ( 509 ) 371 An Improved Spatial Clustering Algorithm HU CaiPing, QIN XiaoLin Spatial clustering is one of the most important spatial data mining techniques. In this paper, an improved spatial clustering algorithm (AISCA) based on DBSCAN is proposed. In order to cluster largescale spatial databases effectively, the proposed algorithm adopts a new sampling technique. In addition, it considers not only spatial attributes but also nonspatial attributes by introducing the concept of the matching neighborhood. Experimental results of 2D spatial datasets show that the proposed algorithm is feasible and efficient. 2007 Vol. 20 (3): 371-376 [Abstract] ( 233 ) [HTML 1KB] [ PDF 321KB] ( 415 ) 377 Face Recognition Based on Wavelet Transform, TwoDimensional Principal Component Analysis and Independent Component Analysis GAN JunYing , LI ChunZhi Combined with wavelet transform (WT), twodimensional principal component analysis (2DPCA) and independent component analysis (ICA), a method for face recognition is presented. Firstly, the original images are decomposed into highfrequency and lowfrequency components by using WT. The horizontal and vertical highfrequency components are ignored, and the noise is eliminated. Then, dimension reduction is performed by 2DPCA, and a whitened matrix is obtained. The independent components of training samples are acquired by ICA. Meanwhile, an independent basis subspace is constructed by the independent basis of training samples. Finally, the projected features of training and the testing samples on the independent basis subspace are gained, therefore face recognition can be realized according to the nearest neighbour rule. Experimental results on Olivetti Research Laboratory (ORL) and Yale face database show that the recognition rate by the proposed method is higher than that by 2DPCA, 2DPCAICA, and WT2DPCA respectively. 2007 Vol. 20 (3): 377-381 [Abstract] ( 234 ) [HTML 1KB] [ PDF 333KB] ( 404 ) 382 Image Retrieval Algorithm Based on Salient Points ZHAO Shan, CUI JiangTao, ZHOU LiHua An image retrieval algorithm based on salient points is proposed. Firstly, a robust and selfadaptive extraction algorithm of salient points is introduced based on the block difference of inverse probabilities model image which was built by an improved block difference of inverse probabilities model. According to the distribution of salient points, the colorspatial feature and the shape feature are extracted to represent image properties for retrieval. The algorithm avoids the defects of interest points in the image retrieval. Furthermore, it reduces the computational complexity of traditional extraction algorithm of salient points. The experimental results demonstrate the proposed method has better performance and higher accuracy than other algorithms. 2007 Vol. 20 (3): 382-387 [Abstract] ( 300 ) [HTML 1KB] [ PDF 1014KB] ( 417 ) 388 Vehicle Tracking Algorithm Based on Modified GVFSnake Model ZHANG Hui , DONG YuNing , XIA Yang A vehicle tracking algorithm based on modified Gradient Vector Flow (GVF)Snake model is proposed. In this algorithm, the original contours of vehicles are automatically acquired by using the framedifference method. The outline of vehicles can be obtained accurately from video stream by means of the modified GVFSnake model. To Compute fast, the modified GVFSnake model adopts the greedy algorithm to control the points according to the distances among them so as to adapt to the change of vehicle targets. Thus, the forecast algorithm is applied to track vehicles. Experimental results testify the effectiveness of the proposed algorithm. 2007 Vol. 20 (3): 388-393 [Abstract] ( 285 ) [HTML 1KB] [ PDF 804KB] ( 506 ) 394 Improved Kernel Minimum Squared Error Method and Its Implementations XU Yong , LU JianFeng , JIN Zhong , YANG JingYu On the basis of the fact that the discriminant vector of the feature space associated with the kernel minimum squared error (KMSE) model can be expressed in terms of a linear combination of samples selected from all the training samples, the idea of variable selection can be exploited to improve the KMSE model. To improve the classification efficiency, an algorithm based on the minimum square error criterion is proposed. It classifies test samples efficiently. Experiments show that the proposed method also has good classification performance. 2007 Vol. 20 (3): 394-398 [Abstract] ( 230 ) [HTML 1KB] [ PDF 328KB] ( 364 ) 399 Research on Gene Expression Programming Algorithm Based on Virus Evolution YANG Jie, LI DeHua, WANG ZuXi, CHEN Lei Combined with the concept of the virus evolution principles, an algorithm, virusgene expression programming (VGEP), is proposed to solve the function fitting problem and the timeseries optimization problem. It improves searching efficiency and decreases the probability of premature phenomena by constructing a new update and infect mechanism for virus. Theoretical analysis has proved that VGEP converges to the global optimum. The simulation results indicate that VGEP performs better than SGEP both in quality of solution and speed of convergence. The results from the project of image calibration show the effectiveness of the propose algorithm. 2007 Vol. 20 (3): 399-405 [Abstract] ( 332 ) [HTML 1KB] [ PDF 1023KB] ( 492 ) 406 Individual Spam Filtering Algorithm Based on Immune Principles ZHANG ZeMing, LUO WenJian, WANG XuFa With the inspiration from selfprotection mechanism of biological immune system, an individual spam filtering algorithm based on immune principles is proposed. Firstly, the spam communities are defined according to the users’ interests and the email features. Then all spams are classified into different spam communities. Secondly, the community features are extracted and represented by a set of feature detectors. Finally, the identification of a spam depends on whether the email can be classified into any spam community. The proposed algorithm is an incremental learning algorithm and it can continuously filter spam without retraining. The immune learning and immune memory mechanisms adopted in this algorithm improve not only the detectable rate and the accuracy rate but also the filter speed. Experimental results show that the algorithm is better than the AISEC algorithm and the Nave Bayesian algorithm. 2007 Vol. 20 (3): 406-414 [Abstract] ( 271 ) [HTML 1KB] [ PDF 682KB] ( 380 ) 415 Online Segmentation Algorithm for Time Series Based on Hierarchical Clustering DU Yi, LU DeTang , LI DaoLun, ZHA WenShu How to segment sequential data in realtime is becoming one of the most important tasks in the time series mining domain. A new online segmentation algorithm called online segmentation algorithm for time series based on hierarchical clustering (OSHC) is presented. According to the order characteristics of sequence data, a novel Segment Feature List (SFList) is developed to save segmentation information. In the algorithm, time series are segmented effectively with one scan of the database and the time complexity is O(n). Historical information can also be inquired quickly by using the SFList. Experimental results show that the algorithm is efficient. 2007 Vol. 20 (3): 415-420 [Abstract] ( 392 ) [HTML 1KB] [ PDF 458KB] ( 497 ) 421 Time Series Pattern Representation Based on Interpolated Edge Operator ZHAN YanYan, XU RongCong, CHEN XiaoYun In terms of the basic principle of edge operator in digital image field, a time series pattern representation based on interpolated edge operator(IEO representation) is put forward.This method, according to a measurement standard, chooses the edge point (end point) of each subpattern of time series pattern representation. The measurement standard is the combination of two submeasurement of interpolated edge operator: edge intensity and interpolation error. The IEO representation of time series can not only compress data, but also effectively restrain the influence of noise. Therefore its adaptability is relatively strong, and it can adapt different data feature environment. 2007 Vol. 20 (3): 421-427 [Abstract] ( 274 ) [HTML 1KB] [ PDF 410KB] ( 703 ) 428 Algorithm for Constrained 3D Object Pose Determination from Single Intensity Images CHENG Lan , XING Zhe , TIAN Yuan Objects, such as road vehicles, are often constrained on a ground plane. The groundplane constraint significantly reduces the pose redundancy of 2D image and 3D line matching. A method is proposed to locate 3D objects on a known plane from a single calibrated intensity image. Firstly, all possible rotation angles of a given model are computed and clustered to obtain some of the most probable rotation angles. Then all possible positions are calculated and clustered for each rotation angle. The candidate rotation angles and positions are filtered and refined to obtain the final results. The proposed method is capable of coping with moderate occlusion. The effectiveness of the method has been verified by experiments. 2007 Vol. 20 (3): 428-434 [Abstract] ( 248 ) [HTML 1KB] [ PDF 751KB] ( 487 ) 435 AdaboostBased Face Detection Method in Hierarchical Feature Spaces LI SanPing, WEI ZhenHan, ZHANG YuSen In the face detection method of Viola and Jones, the weak classifiers based on the feature of Haarlike are very weak in the late stages of the cascade classifier training. Aiming at this problem, a new face detection method based on Adaboost in hierarchical feature spaces is developed. In this method, the performance of the weak classifiers is improved through the training in local and global feature spaces. The experimental results show that the performance of this modified method is better than that of the current stateoftheart system for the high accurate detection rates and the small false alarm number of the system. 2007 Vol. 20 (3): 435-438 [Abstract] ( 250 ) [HTML 1KB] [ PDF 655KB] ( 522 )

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