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.1, Published 2007-02-25 Papers and Reports    Researches and Applications    Surveys and Reviews        Papers and Reports 1 Discriminant Analysis on Nonlinear Manifold Based on Local Linear Discriminant Merging CHEN HuaJie, WEI Wei A local linear discriminant merging method is proposed and the discriminant analysis is carried out on nonlinear manifold. The local linear discriminants were constructed at the local regions obtained by using the Gabriel Graph, then merged to achieve the global nonlinear discriminant. Each local discriminant was assigned to the best weight coefficient in iterative manner by following the margin criterion. So the global nonlinear problem is decomposed into the local linear problem and the discriminant merging problem, which are relatively easy to conquer. The margin criterion based merging algorithm can solve the “sample size sample” problem and ensure that the performance of the global discriminant is independent of the distribution of sample data. The superiority of proposed algorithm is confirmed by experiments on synthesized data and face image set. 2007 Vol. 20 (1): 1-6 [Abstract] ( 242 ) [HTML 1KB] [ PDF 690KB] ( 533 ) 7 Fuzzy Support Vector Classification Based on Possibility Theory YANG ZhiMin , DENG NaiYang The fuzzy support vector classification is discussed, in which both the output of the training point and the value of the final fuzzy classification function are triangle fuzzy number. First, the fuzzy classification problem is formulated as a fuzzy chance constrained programming. Then, this programming is transformed into its equivalence quadratic programming. As a result, fuzzy support vector classification algorithm is proposed. An example is presented to show the rationality of the algorithm. 2007 Vol. 20 (1): 7-14 [Abstract] ( 305 ) [HTML 1KB] [ PDF 386KB] ( 404 ) 15 Binary Ant Colony Algorithm with Signal Transfer XIONG WeiQing, WEI Ping, ZHAO JieYu A kind of Binary Ant Colony Algorithm is designed. Each ant stands at the former place to form onedimension linear queue for transferring the signal from one to another. While the signal passes each ant, the ant randomly chooses the state (0 or 1) according to its own pheromone. Owing to the adoption of the binary coding, the requirment for the behavior of every single ant is lower. So the corresponding memory is relatively less, which greatly improves the efficiency of the algorithm. The test function and the multi 0/1 Knapsack problem show that the proposed algorithm has better convergence speed and stability. 2007 Vol. 20 (1): 15-20 [Abstract] ( 260 ) [HTML 1KB] [ PDF 387KB] ( 446 ) 21 Study of the Behavior of Game System Based on Neural Networks Lü BaiQuan, CAO Yuan In this paper, a Game system with two players is given. Each Player is an agent, and the input of one player is the output of another. The behavior of each player is described by the neural networks so as to make it like that of the humans as much as possible. It is adjusted with the change of neural network weights according to its cost function. Furthermore, 15 kinds of expert decision based on the psychology decision behavior of human being are used in the game system as a decision mechanism for the first time. Some simulations are carried out by using examples from reference [10]. The results indicate the validity of the proposed method which is based on neural networks and the psychology decision behavior of human being. 2007 Vol. 20 (1): 21-27 [Abstract] ( 230 ) [HTML 1KB] [ PDF 631KB] ( 428 ) 28 Evolutionary Algorithm for Magic Squares XIE Tao , ZHAO Bin , XIE DaoYu Magic square construction is a complex permutation problem with a long history. The complexity not only consists of the number of magic squares that increases rapidly with the order of magic square, but also of the percentage of magic squares in the possible permutation of the first n2 natural numbers that decreases with the order. Based on the twophase construction conjecture, an improved evolutionary algorithm for magic square construction is proposed. Mutation operators are specially designed so that the mutation domain can be located and the mutation probabilities can be adjusted adaptively which include the number transpositions, the row transpositions and column transpositions. In addition, some heuristicsbased local permutations, such as the local row/column rectification and the local diagonal rectification, are used to complement the stochastic mechanism. Computational results show that the twophase construction conjecture is computationally effective, and the improved evolutionary algorithm is highly efficient for magic square construction. 2007 Vol. 20 (1): 28-34 [Abstract] ( 411 ) [HTML 1KB] [ PDF 527KB] ( 594 ) 35 Background Subtraction Algorithm Based on Online Clustering XIAO Mei , HAN ChongZhao Based on the assumption that background appears with large frequency, a new online clustering background subtraction algorithm is proposed. The online clustering pixel intensity in a period of time is classified to select the pixel intensity classes whose appearance frequency is higher than a threshold as the background pixel intensity value. It represents the background model of the scene well. Once the background has been constructed, the background difference, the neighborhoodbased background difference and the frame difference are used to detect foregrounds. Simulation results show that the algorithm can handle complex situations with small motions, and the motion detection and the segmentation can be performed correctly. 2007 Vol. 20 (1): 35-41 [Abstract] ( 280 ) [HTML 1KB] [ PDF 1954KB] ( 464 ) 42 Research on the Dependency between Optimal Parameter and the Input Noise in Possibilistic Linear Model GE HongWei, WANG ShiTong Possibilistic linear model (PLM) based on possibility theory plays a pivotal role in fuzzy modeling. In order to enhance the generalization capability of the linear model, the regularized version is firstly extended, i.e. the regularized possibilistic linear model (RPLM). Then the RPLM is transformed into the corresponding equivalent MAP problem. Accordingly, with a series of mathematical derivation, the inversely proportional dependency between the parameter and the standard deviation of Gaussian noisy input is revealed. In the meanwhile, the simulation result has proved this conclusion. Obviously, the conclusion is helpful for the practical applications of both PLM and RPLM. 2007 Vol. 20 (1): 42-47 [Abstract] ( 231 ) [HTML 1KB] [ PDF 419KB] ( 376 ) 48 Semantic Web Oriented Description Logic JIANG YunCheng , TANG Yong , WANG Ju , ZHOU ShengMing The relationship between the description logic SHOIQ(D) and the semantic Web, especially the equality of SHOIQ(D) and semantic Web ontology language, is analyzed. Due to the complexity of SHOIQ(D), only the syntax and the semantics are given at present. Based on the reasoning method of the description logic SHOIQ, the tableaux of SHOIQ is presented at first. Then the reasoning method of SHOIQ(D) based on tableaux of SHOIQ is studied mainly, and a SHOIQ(D)tableaux reasoning algorithm is presented. Finally, the correctness of SHOIQ(D)tableaux algorithm is proved. 2007 Vol. 20 (1): 48-54 [Abstract] ( 278 ) [HTML 1KB] [ PDF 345KB] ( 675 ) 55 PGHMI: Mutual Information Based Feature Selection Method WANG Hao, SUN HongBin, ZHANG BoMing Conventional samplebased mutual information estimation methods can’t handle the mixed features directly that include both numeric attributes and nominal attributes. A Parzen window based general mutual information calculation method, PG method, is proposed in this paper, which could deal with the mixed attributes directly. A criterion named hybrid mutual information (HMI) is presented. Based on PG mutual information estimation method and HMI feature selection criterion, a feature selection algorithm (PGHMI) is proposed. Experimental results show the correctness of PG and the effectiveness of PGHMI. 2007 Vol. 20 (1): 55-63 [Abstract] ( 289 ) [HTML 1KB] [ PDF 560KB] ( 713 ) 64 TwoLevel Stereo Matching Algorithm Based on the Graph Cuts of Network WANG Zhe, CHANG FaLiang The stereo matching is a problem in the computer vision. In order to obtain the precise dense disparity map, a twolevel matching algorithm based on the graph cuts of network is proposed. The algorithm synthesizes the advantages of the areabased process algorithm and the graph cuts global algorithm. Firstly, the twolevel pyramid data structure for the original image pair is gotten and the global optimization matching in the lower resolution image pair is obtained by using the graph cuts method. Then under the constraint of the acquired disparity map, the areabased stereo matching algorithm is employed to get the dense disparity map of the original image pair. The algorithm not only reduces the search range of matching, but also ensures the validity of matching. The experimental results show the algorithm is efficient and feasible. 2007 Vol. 20 (1): 64-68 [Abstract] ( 277 ) [HTML 1KB] [ PDF 1142KB] ( 454 ) 69 A Method for Multiple Attribute Decision Making without Weight Information ZHANG FangWei , YAO BingXue The multiattribute decisionmaking problems are discussed, in which the information on attribute weights is completely unknown. Using the mathematical simulation methods, a model is given to solve this problem.This model can be realized on PC program and the valuation results are objective. Finally, one practical example is given. 2007 Vol. 20 (1): 69-71 [Abstract] ( 257 ) [HTML 1KB] [ PDF 258KB] ( 483 ) Surveys and Reviews 72 How to Add Transparency to Artificial Neural Networks HU BaoGang, WANG Yong, YANG ShuangHong, QU HanBing The main issue about “black box” inherent in artificial neural networks (ANN’s) is discussed. Adding transparency is well recognized to be an effective solution to dealing with this problem. Significant benefits are obtained through using this approach, such as providing a certain degree of comprehensive power, decreasing model size, speeding learning process and improving generalization capability. A hierarchical classification is applied to the existing approaches for better understanding of their intrinsic features and limitations. The first level of classification is made by two strategies: building prior knowledge into neural networks; extracting rules embedded within networks. Most of important approaches are introduced and compared in detail with further classifications within each strategy. Finally, the personal perspectives to the studies of machine learning are presented. Other objective functions are suggested for the extension of studies, such as performancetocost ratio and transparency. The study of increasing transparency to ANN’s is considered as the most fundamental and direct solution to the other existing issues. A new machine learning approach called Knowledge Increasing via Feedback is proposed. 2007 Vol. 20 (1): 72-84 [Abstract] ( 257 ) [HTML 1KB] [ PDF 782KB] ( 520 ) Researches and Applications 85 Adaptive Immune Algorithm and Its Track to Dynamic Function Optimization ZHANG ZhuHong, QIAN ShuQu ;An adaptive immune algorithm, based on the functions of the biological immune system such as adaptive learning, memory and surveillance, is proposed to solve high dimensional dynamical function optimization. In the algorithm, with structural simplicity, feasibility, and dynamical regulation of the execution time for different environments, the dynamical evolution and antibody rearrangement are involved. The dynamical memory pool consists of memory subsets related to the characteristics of immune memory and dynamical maintenance of the pool, in which each subset keeps some excellent memory cells obtained by the average linkage. In the meanwhile, dynamical surveillance and memory establish the environmental identification and generation rule of initial antibody populations. Experimental results and comparison illustrate the superiority and the effective tradeoff between performance effect and efficiency as well as the potential in the complex dynamical highdimensional optimization problems. 2007 Vol. 20 (1): 85-94 [Abstract] ( 248 ) [HTML 1KB] [ PDF 2660KB] ( 413 ) 95 Online Signature Verification Based on Segment Features and HMM MENG Ming , WU ZhongCheng , YU Yong , GE YunJian Writing forces of signature contain a great many individual characteristics, but in current online handwritten signature verification systems this information is often utilized deficiently because of the limitation of the existing input device. In this paper, an online signature verification algorithm using the shape features and the force features is presented. The dynamic trajectory and the writing force of signature are captured by a novel digital tablet, namely FTablet. Each signature is segmented according to its minimum velocity points. Then a 16dimensional vector of shape and force features is extracted for each segment. The resulting sequence is used for training a HMM to achieve signature verification. The experimental results on our database show that the force feature is more difficult to forge than the shape feature and the combination of the two features can effectively improve the performance. The proposed algorithm has achieved equal error rate (EER) of 3.9%. 2007 Vol. 20 (1): 95-100 [Abstract] ( 231 ) [HTML 1KB] [ PDF 722KB] ( 713 ) 101 Recognition of Radar Target Rangeprofiles Based on Nonlinear Canonical Subspace ZHOU DaiYing , YANG WanLin The novel subspace method is proposed. Firstly, the input rangeprofiles are transformed into high dimension feature space by nonlinear transformation. Then the canonical subspace is constructed in the high dimension feature space to extract feature for improving performance of classification. Finally, the minimum distance classifier is used to classify aircraft targets. The experimental results on the real data of three kinds of aircrafts show the efficiency of the proposed method. 2007 Vol. 20 (1): 101-104 [Abstract] ( 340 ) [HTML 1KB] [ PDF 286KB] ( 596 ) 105 Gait Recognition Method Based on Dynamic Energy Feature CHAI YanMei, ZHAO RongChun, TIAN GuangJian, JIA JingPing Recognizing people by their gait is a recent research hotspot. Compared with other biometrics, gait has the following three advantages: distant recognition, uninvasive and difficult to conceal. A new gait recognition method based on the dynamic energy feature is proposed in this paper. Firstly, the background is initialized automatically in the gait sequence. Then the binary silhouette of a walking person is detected by background subtraction technology. Next, the dynamic energy feature matrixes are extracted from binary silhouette sequences. Finally, the correlation coefficient measure and two different classification methods (NN and KNN) are used to recognize different subjects. Experimental results show that the new method is effective. Recognition rate of over 90% on both UCSD database and CMU database are achieved. 2007 Vol. 20 (1): 105-109 [Abstract] ( 320 ) [HTML 1KB] [ PDF 603KB] ( 594 ) 110 Distance Weighted 2D Kernel AutoAssociation Memory Model and Its Applications CHEN Lei , WANG ChuanDong , SUN ZhiXin , CHEN SongCan By using the kernel trick to modify Hopfield autoassociative memory model (HAM), a framework of kernel autoassociation memory model (KAM) is proposed. KAM makes exponential correlation associative memory (ECAM) and improved ECAM (IECAM) become two special cases. Then, the framework of distance weighted 2D kernel autoassociation memory model (DW2DKAM) is constructed by introducing distance factors to the kernels. DW2DKAM improves the storage capacity and errorcorrecting capability of KAM when recognizing binary visual images. Simulation results verify that DW2DKAM has higher storage capacity and better errorcorrecting capability than those of KAM, and outperforms the recently proposed modular HAM by Seow and Asari. 2007 Vol. 20 (1): 110-114 [Abstract] ( 327 ) [HTML 1KB] [ PDF 898KB] ( 493 ) 115 Study of SelfCalibration for Ion Sensor Based on LSSVM CHEN Feng, YANG DaFu, FANG Ke, WANG Bing The ion sensor(ion selectivity electrode) is one of the key technologies in the water quality monitoring, wastewater treatment and factory agriculture. The nonlinearity, drift and intercross sensitivity of the ion sensor impact on the accuracy and reliability obviously, therefore, it is difficult for majority of the ion sensors to continuously detect in situ. In order to detect dynamic environment on line, selfcalibration of the ion sensor is investigated and its response properties are analyzed in terms of the experimental data. The zero and time drift are taken into account, and a selfcalibration of ion sensor is proposed based on least squares support vector machines (LSSVM). The experimental results show response error of the ion sensor is decreased obviously and this approach is practical. 2007 Vol. 20 (1): 115-118 [Abstract] ( 288 ) [HTML 1KB] [ PDF 372KB] ( 471 ) 119 A Blind Audio Watermarking Algorithm LI XiaoWei, WANG ZengFu In this paper, a blind audio watermarking scheme based on spreadspectrum technology is proposed. The blind detection is implemented using linear prediction coding as a whitening procedure. By a series of techniques, including sophisticated subband selection and PN shaping, the robustness and the imperceptibility of proposed algorithm are comparable to those of the nonblind algorithms which are demonstrated in the experiments. Also, the watermark capacity of proposed algorithm is larger than that of other blind ones. 2007 Vol. 20 (1): 119-123 [Abstract] ( 251 ) [HTML 1KB] [ PDF 478KB] ( 779 ) 124 Recognition of Partially Occluded Polygon from Line Drawing ZHANG GuiMei , GAO ManTun , SHEN YunWen The existing methods for recognizing the partially occluded polygon always carry out under the similarity transformation. A new method for recognition of partially occluded polygon under the affine transformation is proposed. First, a new local invariant under the affine transformation is given based on the invariants in computer vision. Next, on the basis of the local invariant, a new similarity function is established. Then a transform function is designed to normalize the similarity measure value between 0 and 1, so it is convenient to select the similarity measure threshold. Finally, a lost feature judgement function is constructed to judge whether each local featuer is lost, and similarity measure is calculated only using the local features that are not lost. By comparing the similarity measure with the threshold, the polygon objects can be recognized from a partially occluded line drawing. Noise and occlusion are considered in the construction of the similarity function and the lost feature judgement function. The experiment results show that the proposed algorithms are insensitive for noise and occlusion. 2007 Vol. 20 (1): 124-130 [Abstract] ( 284 ) [HTML 1KB] [ PDF 610KB] ( 466 ) 131 AML: OrientedRequirment MultiAgent System Modeling Language SHAO Kun , LIU ZongTian , HU XueGang , LI XinKe An agent modeling language, named AML, is proposed. The language is based on the parliamentary architecture of multiagent system and involves eight models in requirement process, system analysis process and system design process. It defines the workflows in the construction of these models and the relationship between different models. All models of AML adopt UML standard notations to be consistent with UML. To meet the special requirements of AML, extensibility mechanism of UML and selfdefining notations are used to extend models. An AML tool, AMLTools, is introduced to support AML. Furthermore, a case of an autowarehouse system is given to demonstrate the application of AML. 2007 Vol. 20 (1): 131-137 [Abstract] ( 267 ) [HTML 1KB] [ PDF 470KB] ( 693 ) 138 Greedy Algorithms with Kernel Matrix Approximation DU JingYi,HOU YuanBin Kernel algorithm is a new nonlinear technique in the machine learning field with great practical value. In the Kernel algorithm Kernel Matrix serves as a bridge between datain and learning algorithms. The lowrank approximation of Kernel Matrix is an effective means of advancing the computational efficiency of Kernel algorithm and reducing the internal memory. Based on the correlation between data and minimal residual norm, the forward greedy algorithm, the backward greedy algorithm and the mixed greedy algorithm are proposed to look for the optimum lowrank approximation. A sparse regression algorithm (SRA) is presented which effectively cuts off “training samples” and keeps good generalization capacity. With comparison to SVM,KA and KPCR, the superiority of SRA is demonstrated on two datasets. 2007 Vol. 20 (1): 138-143 [Abstract] ( 416 ) [HTML 1KB] [ PDF 366KB] ( 729 )

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