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.2, Published 2007-04-25 Papers and Reports    Researches and Applications    Surveys and Reviews        Papers and Reports 145 Organizational Evolutionary Particle Swarm Optimization for Numerical Optimization CONG Lin, SHA YuHeng, JIAO LiCheng An organizational evolutionary particle swarm optimization (OEPSO) is presented. The evolutional operations are acted on organizations directly in the algorithm. The global convergence is gained through competition and cooperation among the organizations, and the mathematic convergence is given. In the experiments, OEPSO is tested on 12 unconstrained benchmark problems, and compared with FEP and three algorithms based on the PSO. In addition, the effects of parameters in the algorithm are analyzed. The results indicate that OEPSO performs better than other algorithms both in solution quality and computational complexity. The analyses of parameters show OEPSO has stable performance and high success ratio, and it is insensitive to parameters. 2007 Vol. 20 (2): 145-153 [Abstract] ( 269 ) [HTML 1KB] [ PDF 581KB] ( 420 ) 154 Study on Symbolization Analysis of Time Series XIANG Kui, JIANG JingPing Symbolization is an important method for time series analysis, but choosing appropriate symbolization strategy is very difficult. Finite statistic complexity (FSC) can calculate the information quantity contained in the symbol time series, so it is evaluation criterion of symbolizing process. In this paper, several symbolization methods are analyzed including static transformation method, dynamic method, wavelet space method, etc. Eight time series are transformed into the symbol series by different methods and the FSC of all the symbol series are compared from several aspects. These time series which come from different domains are nonlinear and nonstationary. Some meaningful empirical conclusions are thus drawn. All of the analyses imply that the dynamic transformation is the best, and then the integrated one and wavelet space one. Unexpectedly, the static transformation is the most commonly used but the worst. 2007 Vol. 20 (2): 154-161 [Abstract] ( 411 ) [HTML 1KB] [ PDF 835KB] ( 1051 ) 162 Image Registration Based on the Quadratic Sum of Phase Correlation LIAN Wei, LIANG Yan, CHENG YongMei, PAN Quan, ZHANG HongCai A method for twoimage registration with rigid transformation is proposed. It directly constructs an evaluation function of rotations, which comes from the quadratic sum of the one dimensional phase correlation between the Randon transforms in the two aligned images. The coordinate of the maximum value of this function yields the estimate for rotation angle. Experimental results show that the proposed method outperforms phaseonly bispectrum based method, and it can estimate both arbitrary rotations and medium translations robustly and accurately. 2007 Vol. 20 (2): 162-166 [Abstract] ( 345 ) [HTML 1KB] [ PDF 843KB] ( 650 ) 167 Brain Emotion Circuit Based Artificial Emotional Intelligence Model WANG ShangFei, WANG XuFa The emotion is the highlevel function of brain, which helps the organisms to survive and adapt to the environments. It is a strong faculty for learning, memorizing and making decision. A computational model, named brain emotion circuit based artificial emotional intelligence model (BEI), is presented to imitate the amygdale and the orbitofrontal cortex which are two key parts of the brain and responsible for emotion learning and emotion adjusting. A BEI controller is proposed and its effect is experimentally evaluated in controlling chemical process and inverse pendulum. The experimental results demonstrate the quick response and the robustness of proposed controller. 2007 Vol. 20 (2): 167-172 [Abstract] ( 306 ) [HTML 1KB] [ PDF 593KB] ( 1022 ) 173 An OrdinationFuzzy MinMax Neural Network Classifier on Unlabelled Pattern Classification HU Jing, YANG Jing, GAO Jun An ordinationfuzzy minmax neural network(OFMM) based on nonmetric multidimensional scaling (MDS) is proposed to solve the classification problems of unlabelled input pattern. Firstly, all the input patterns are sorted by MDS to get their similarity measures. Then these measures are used to supervise the following expansion and contraction stage of hyperboxes for classification. OFMM shows the improvements in the validity of unlabelled patterns classification, the network structure, and training time. The experimental results on standard dataset demonstrate that OFMM is a practical and effective classifier which is superior to the traditional generalfuzzy minmax neural network (GFMM). 2007 Vol. 20 (2): 173-179 [Abstract] ( 246 ) [HTML 1KB] [ PDF 487KB] ( 415 ) 180 An Object Detection Algorithm Based on Moving Shadow GUO LiSheng, GUO Li, JIAO RongHui, ZHENG Jun Moving shadow detection is critical for video surveillance system since shadow points are often misclassified as object points, which causes errors in segmentation and tracking. The conventional algorithms always need the scene characteristics, which makes the algorithms not widely used and automatically performed. An approach is presented that does not rely on any scene assumptions such as camera location and scene characteristics. Color information, texture information and spatial constraints are embedded to define the overall algorithm. Firstly, relevant areas are identified in each image. Then, the color distortion caused by shadow are calculated. Finally, the moving object and shadow are detected by using color distortion compensation and texture verification. The proposed algorithm is demonstrated on many kinds of outdoor video sequences. Moreover, the performance comparisons show that the proposed algorithm outperforms the conventional ones. 2007 Vol. 20 (2): 180-184 [Abstract] ( 239 ) [HTML 1KB] [ PDF 735KB] ( 529 ) 185 A Model Selection Method of Influence Diagrams Based on PSEM Algorithm and BP Neural Network YAO HongLiang, ZHANG YouSheng , WANG Hao, Wang RongGui In the model selection of influence diagrams(IDs), the problems of the data dependency, the computation complexity and nonprobability relation are discussed. Based on the structure decomposition of IDs, a PSEM algorithm is presented. A BP Neural Network is introduced by learning local utility function of each utility node, and the overfitting is avoided by inducing the threshold of weights. To reduce the data dependency, a new MDL scoring is presented which includes the prior knowledge of network structures. Based on SEM algorithm, PSEM algorithm induces the new MDL scoring, and separates parameters learning from structures scoring to improve the computation efficiency. Compared with SEM algorithm, the performances of both the computation complexity and the data dependency of PSEM algorithm are improved, and the model selection of the utility part is easy to achieve. 2007 Vol. 20 (2): 185-190 [Abstract] ( 273 ) [HTML 1KB] [ PDF 454KB] ( 603 ) 191 An ImmuneBased Algorithm for MultiRobot Complete Exploration in an Unknown Environment GAO YunYuan, WEI Wei A complete exploration algorithm using multiple robots in an unknown environment is proposed. Based on the principle of the immune system, the system is fully decentralized. It has no requirement for the starting points and the formation of the robots. Each local environment condition sensed by the robots is considered as antigen while robot is regarded as Bcell and the next exploration target as antibody respectively. Each robot works independently according to the antigen information. The coordination of the robots can be realized by utilizing the interactions among antibodies. The system carries out the exploration of unknown environment with the combination of robots’ independence and coordination. Instead of recording information about environment map, the complete exploration is realized using the frontier in the memory library. Simulation results validate the algorithm is efficient, complete and robust to the fail of robot and the lost of the communication. 2007 Vol. 20 (2): 191-197 [Abstract] ( 299 ) [HTML 1KB] [ PDF 820KB] ( 482 ) 198 Temporal Rule to Formalism and Measure Estimation PAN Ding , SHEN JunYi A formalism framework for the temporal rule is proposed to define the main concepts used in the rule induction. The concept of the linear state structure allows each state to associate with a symbol of the restricted firstorder language and measure the truth range of a formula. Thus the measure sequence with coherent properties is generated. The novelty of the temporal rules is testified by using dynamic time warping distance. The diffusion estimation algorithm applicable to the small sample is proposed to calculate the parameters of measure sequence. Experimental results show effectiveness, robustness and simplicity of the proposed method. 2007 Vol. 20 (2): 198-204 [Abstract] ( 241 ) [HTML 1KB] [ PDF 397KB] ( 367 ) 205 Comparison of Constructions for Hierarchical Structure Based on Confusion Matrix XIONG YunBo, LI RongLu, HU YunFa The hierarchical clustering and confusion classification are used to construct a documenttype hierarchical structure based on confusion matrix. The experimental results using hierarchical classification show that the performance of confusion classification excels that of hierarchical clustering, and the confusion classification improves the precision and recall of flat document classifier. 2007 Vol. 20 (2): 205-210 [Abstract] ( 300 ) [HTML 1KB] [ PDF 400KB] ( 544 ) Surveys and Reviews 211 A Survey for Study of Feature Selection Algorithms MAO Yong , ZHOU XiaoBo , XIA Zheng , Yin Zheng , SUN YouXian Feature selection is a hot topic in current information science, especially in the field of pattern recognition. In this paper, feature selection algorithms are classified from different points of view. Several embranchments of feature selection and the development situation are introduced. Some difficulties in the theoretic analysis and application are involved. From a practicality angle, using support vector machine to select features is considered as the research direction in machine learning. 2007 Vol. 20 (2): 211-218 [Abstract] ( 339 ) [HTML 1KB] [ PDF 440KB] ( 2318 ) Researches and Applications 219 A Moment and Dominant PointsBased Method for Polygonal Approximation XIE MingHong , ZHANG YaFei , FU Kun , WU YiRong A method based on geometric moments and dominant points is proposed. Algorithm for detecting dominant points could keep the original contour feature as much as possible, but the number of remain vertexes is not controllable. On the contrary, the method based on geometric moments can reduce the amount of vertexes to any number, but it makes the fitting result get into local optimum. Thus, a new method is introduced which integrates the two algorithms. In this way, most closed curves can be fitted to polygons with specified number of vertexes in a global optimal way. 2007 Vol. 20 (2): 219-224 [Abstract] ( 258 ) [HTML 1KB] [ PDF 530KB] ( 817 ) 225 Chaos Immune Evolutionary Algorithm and Its Applications to Function Optimization ZHANG HaiYing , HAN GuiJin , PAN YongXiang Based on the clonal selection principle in the immune system and utilizing the ergodic property of the chaotic sequence, a chaos immune evolutionary algorithm is proposed. Firstly, the chaotic sequence is introduced into the generation of the initial population and expansion process of the antibody. Secondly, the affinity of antibody ready for expansion in the population is varied to modulate the choose probability. Finally, the algorithm is proved to be convergent by utilizing the method of probability analysis. In order to test the validity of the algorithm, it is applied to solving the problem of function optimization. Simulation experiments are made using several different functions and the results show many virtues of the algorithm, such as avoiding local optima, high precision solution and quick convergence. 2007 Vol. 20 (2): 225-229 [Abstract] ( 226 ) [HTML 1KB] [ PDF 340KB] ( 593 ) 230 An Improved GGAPRBF Algorithm and Its Application to Function Approximation LI Bin , LAI XiaoPing An improved learning algorithm for RBF neural network based on the GGAPRBF algorithm is proposed. The adaptive adjusting algorithm for the width of hidden neuron radial basis function and the dynamical regulation of overlap threshold are introduced into the GGAPRBF algorithm. The presented algorithm is compared with RAN, RANEKF, MRAN,IRAN and GGAPRBF(GAPRBF) algorithms by simulation on three benchmark problems in the function approximation area. The results indicate that the proposed algorithm provides good generalization performance with considerably reduction in network size and training time. 2007 Vol. 20 (2): 230-235 [Abstract] ( 375 ) [HTML 1KB] [ PDF 709KB] ( 437 ) 236 Discovery Algorithm for Option Based on Exploration Density MENG JiangHua, ZHU JiHong, SUN ZengQi A new method, named exploration density(ED) inspection, is presented. Useful options were discovered by the method through inspecting the influence of the state on agent’s explore ability in state space. The simulation results show that the proposed algorithm has better performance in reinforcement learning. The method has characteristics of taskindependence, no need of prior knowledge, etc. The created options can be directly shared among different tasks in the same environment. 2007 Vol. 20 (2): 236-240 [Abstract] ( 232 ) [HTML 1KB] [ PDF 666KB] ( 435 ) 241 An Image Segmentation Algorithm Based on Color and Texture Analysis SHEN XiangJun, WANG ZengFu A fast image segmentation algorithm based on seed region growing is proposed. Firstly, the algorithm represents an input color image by color quantization. Then, the seed regions are selected from the quantized image with windows of different sizes according to the distribution of the quantized color labels. Finally, both of the color and the texture features are used to fulfill the image segmentation quickly by seed region growing based operation. The experimental results show that the proposed algorithm has good performance in not only timeconsuming but also segmentation. 2007 Vol. 20 (2): 241-247 [Abstract] ( 239 ) [HTML 1KB] [ PDF 1265KB] ( 480 ) 248 A Remote Sensing Image Fusion Algorithm Based on à trousContourlet Transformation LUO Li , YUAN Zhen , WANG Ke A new algorithm is developed which merges a highresolution images and multispectral images based on the à trousContourlet transform. First, à trousContourlet transform is presents based on the advantages of à trous wavelet transform and Contourlet transform. Then the images to be merged are decomposed by multiresolution based on the IHS transform and à trousContourlet transform. According to the high frequency information, a new algorithm of weighted average is proposed according to its deviation and collection coefficient. Finally, the image is fused by high frequency addition. The experimental results show that better fusion result by the proposed method is acquired compared with the traditional transform methods. 2007 Vol. 20 (2): 248-253 [Abstract] ( 257 ) [HTML 1KB] [ PDF 1097KB] ( 386 ) 254 An InternalInference Based Multiagent Learning Method HAN Wei , CHEN YouGuang , JIANG ChangHua In multiagent environment, the optimal policy of an agent depends on the policies of the others, which makes the learning more problematic. Previous algorithms based on the observed behavior of opponents can not fully present individual rationality. An efficient online learning algorithm based on the internal inference is proposed, which integrates the observed objective behavior and the subjective inferential intention of the opponents. By the internal inference, agents can obtain more information about opponents, and thus learn more efficiently. Simulations results prove that the proposed algorithm performs well in classical coordination game. 2007 Vol. 20 (2): 254-260 [Abstract] ( 232 ) [HTML 1KB] [ PDF 422KB] ( 533 ) 261 Seal Identification Based on Delaunay and Polygon Triangulation YUAN ZhanTing, ZHANG QiuYu, JIN YanFeng ;The gesture adjustment and the match of the seal image are studied. An approach for the seal image matching based on DT grid is proposed. After the topological structures of two minutiae which are based on line and polygon are triangulated and the minutiae taken from the template and the query seal images are joined, the DT grid is gained. Firstly, the Delaunay point set triangulation cutting algorithm is used to triangulate the minutiae based on line, while the polygon triangulation cutting algorithm is employed to triangulate the minutiae based on polygon. Then the reference junction pairs are obtained by searching two trigonal nets. Finally, the query seal image is adjusted according to the templated seal image with parameters computed from the reference junction pairs and the match score is caculated using a simply match algorithm. Simulation results show that the proposed method gets more reference points and the accuracy of rotating and translating parameters of the seal imprint is also ensured. 2007 Vol. 20 (2): 261-265 [Abstract] ( 265 ) [HTML 1KB] [ PDF 416KB] ( 620 ) 266 Complex System Behavior Forecasting Method Based on BMACRLS Model and Its Application YANG XiaoYu , FU ZhongQian , WANG WeiPing ;Complex system behavior forecasting is quite important in complex system management and decision. It is the key of improving forecasting stability and extension without the loss of precision. A new method based on PCA (principle component analysis) and CMACRLS (recursive least squares) is proposed. PCA is used to reduce the input space dimensions. CMACRLS algorithm combined with the Bspline is introduced to ensure the weight convergence and provide the differential information of function adapted to the online modeling. Then the load forecasting is performed by PCABMACRLS and RBF neural network on the data of FuYang Power Land in 2004. The result comparison between two algorithm illustrates the validity of the proposed method. 2007 Vol. 20 (2): 266-270 [Abstract] ( 251 ) [HTML 1KB] [ PDF 456KB] ( 399 ) 271 Similarity Measuring Method in Time Series Based on Slope ZHANG JianYe , PAN Quan , ZHANG Peng , LIANG JianHai The similarity search in time series attracts much attention recently, and the similarity measuring method is one of the most important problems in this area. Aiming at the time series data with high dimension, multiproperty, noise and default, a similarity measuring method based on slope is proposed in this paper. Based on PLR, the method distributes the power among their slope differences so that its physical concept is more clear. The slope distance completely satisfies the basic rules of similarity measuring, and the example shows its validity. 2007 Vol. 20 (2): 271-274 [Abstract] ( 370 ) [HTML 1KB] [ PDF 316KB] ( 1211 ) 275 An ACOBased Fair Energy Usage Routing Algorithm for Wireless Sensor Networks LIANG HuaWei , CHEN WanMing , LI Shuai , MEI Tao , MENG Max How to make good use of the limited energy to maximize the network life span is an important problem in the study of the wireless sensor networks (WSN). The life of a WSN depends on the minimum of the residual energy of its nodes. A fair energy usage routing algorithm is proposed which uses the Ant Colony Optimization Algorithm (ACO) to balance the network energy distribution and extend the network life. The proposed algorithm utilizes the dynamic adaptability and optimization capabilities of the Ant Colony to get a treadoff between the shortest path and the fair energy usage. Simulation results show that the proposed algorithm is good at balancing the energy usage, and it effectively extends the span of the network life. The network life span using the ACObased fair energy usage routing algorithm is extended over 33% compared with the one using the shortest path optimization algorithm. 2007 Vol. 20 (2): 275-280 [Abstract] ( 242 ) [HTML 1KB] [ PDF 809KB] ( 537 ) 281 A Method for Human Gait Recognition Using SpatialTemporal Analysis SU Han , HUANG FengGang A new method for modelfree recognition of gait based on silhouette in computer vision sequences is proposed. The silhouette shape is represented by a novel approach which includes not only the spatial body contour but also the temporal information. Using this shape representation, the temporal information is extracted with low cost of computation. First, a background subtraction is used to separate objects from background, and gait cycle is obtained by analyzing the variety of the silhouette width and height. Then, the spatial shape of walker and their motion by the temporal matrix are presented, and Discrete Fourier analysis is used to analyze the gait feature. The nearest neighbor classifier is used to distinguish the different gaits of human. The performance of the proposed approach is tested on different gait databases. Recognition results show it is efficient. 2007 Vol. 20 (2): 281-286 [Abstract] ( 284 ) [HTML 1KB] [ PDF 526KB] ( 581 ) 287 A Method for Regional Segmentation and Semantic Categorization Based on Rough Set Theory XIE Zhao, GAO Jun A rough set theory based method for image understanding is proposed. The images are regarded as the information system and each pixel in them as an object in the system. The reduction process and extend models with lowerupper approximations and core attribute concepts in rough sets are considered. Then a segmentation algorithm and a rule reduction and inference method are proposed. The experimental results demonstrate the feasibility and the accuracy of proposed method by comparing it with Ncuts segmentation algorithms and statistical learning ways. 2007 Vol. 20 (2): 287-294 [Abstract] ( 263 ) [HTML 1KB] [ PDF 1288KB] ( 499 )

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