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.6, Published 2006-12-25 Papers and Reports    Researches and Applications    Surveys and Reviews        Papers and Reports 689 Estimation of Posterior Probability and Applications: An Approach Based on Kernel Logistic Regression LI Tao, WANG JunPu, WU XiuQing, TANG JinHui A method based on feature vector set is proposed to render a sparse solution for kernel Logistic Regression (LR) and decrease computation complexity of posterior probability estimation. The proposed method is combined with Markov Random Field (MRF) in terms of Bayes rule, in which the conditional probability is replaced with posterior probability estimated by kernel LR. The combination is applied to image segmentation. Experiments on texture image segmentation show the performance of the proposed method is suporior to that of Gaussian MRF method. 2006 Vol. 19 (6): 689-695 [Abstract] ( 280 ) [HTML 1KB] [ PDF 1124KB] ( 310 ) 696 Semantic SimilarityBased Information Retrieval Method WANG Jin, CHEN EnHong, SHI DeMing, ZHANG ZhenYa With exponential growth of web information and emergence of semantic web, semantic information retrieval becomes a hotspot of current research. In this paper, a novel information retrieval method based on semantic similarity is studied. By exploiting the advantage of ontology in describing semantic as well as in computing the ConceptSimilarity and PropertySimilarity for the semantic retrieval algorithm, the proposed method greatly improves the retrieval precision. The statistic analysis of the semantic retrieval experiment results illustrates that the retrieval accuracy is improved significantly with the increase of the feature vector concepts and the properties of the retrieved documents. 2006 Vol. 19 (6): 696-701 [Abstract] ( 246 ) [HTML 1KB] [ PDF 414KB] ( 317 ) 702 Approach to Segment MultiSize Machine Printed Characters by Removing Serifs GUO JianXiong, YANG LiHua To segment machineprinted English characters of small font is a challenging problems. It is difficult to segment small font characters because of their small energies and the overlap by letters. In this paper, many kinds of overlaps caused by the serifs of letters are analyzed, and an algorithm to remove the serifs is proposed. Consequently, a method to segment machineprinted English characters of small fonts is designed. Experiments give encouraging segmentation results for not only small font characters but also large font characters. 2006 Vol. 19 (6): 702-707 [Abstract] ( 242 ) [HTML 1KB] [ PDF 1075KB] ( 592 ) 708 An Evolutionary Algorithm Using Utility Function of MultiPlayer Gameas Evolution Directing Function XU Min, ZHANG SiHai, WANG XuFa An evolutionary algorithm based on multiplayer noncooperation game is proposed to solve the combinatorial optimization problems, and the algorithm is modelled as a Markov chain. In this algorithm, combinatorial optimization problem is considered as an nperson game, and the solution of problem is optimized through agents’ rational behavior. A welldefined and expandable frame of the algorithm is constructed, and three constraints that the algorithm must satisfy are presented: finite constraint, weakly consistent constraint and convergent constraint. The algorithm is used to solve some typical NPComplete combinatorial optimization problems. The theoretical analysis and experimental results compared with other optimization algorithms show the proposed algorithm has a good ability of problem solving. 2006 Vol. 19 (6): 708-714 [Abstract] ( 236 ) [HTML 1KB] [ PDF 398KB] ( 532 ) 715 Framework and Algorithm Model of Schema Matching Problem ZHANG Zhi, CHE HaoYang, SHI PengFei In this paper, an algebraic framework for Schema Matching Problem (SMP) is built. By using universal algebra, the mathematical foundations of SMP is studied. A SMP instance can be viewed as a pair of structures (algebras), then the solutions to the problem are the structure preserving mappings between these two finite structures, i.e. schema matching can be formulized as finding the homomorphism between two structures. It is proved that the schema homomorphism (SHOM) is equivalent to the SMP, and the SMP can be reduced to the homomorphism between two finite structures. Based on this framework, the algorithmic model of SMP is presented. 2006 Vol. 19 (6): 715-721 [Abstract] ( 220 ) [HTML 1KB] [ PDF 349KB] ( 312 ) 722 OnRoad Vehicle Detection Approach Based on HypothesisValidation Structure SUN Chong, CHENG Hong, ZHENG NanNing A vehicle detection approach based on Gabor feature and multiresolution hypothesisverification structure is proposed. The proposed approach includes two basic phases. Firstly, the Regions of Interest (ROI) in an image are determined according to the lane vanishing points. Then a hypothesis list in each ROI is created according to the vertical edges and horizontal edges. Finally, a hypothesis list for the whole image is obtained by combining these three lists. In the hypothesis validation phase, a vehicle validation approach using Support Vector Machine (SVM) is proposed. The proposed algorithm decreases the computational cost by eliminating uninteresting area, and in the hypothesis verification phase, the positive false is low. The experimental results show that the average right detection rate reaches 90% and the execution speed is 20fps using a Pentium(R)4 CPU 2.4GHz. 2006 Vol. 19 (6): 722-726 [Abstract] ( 282 ) [HTML 1KB] [ PDF 1084KB] ( 580 ) 727 Cluster Analysis Based on Mathematical Morphology LUO HuiLan, KONG FanSheng, YANG XiaoBing, LIU BiHong Mathematical morphology has been widely used in image analysis. Classical clustering methods often fail to deliver satisfactory results, especially when clusters have arbitrary shapes. Through some techniques for selecting discretization parameters and structural elements, a new approach to cluster analysis is proposed, which is based on the mathematical morphology operations. Clusters are well separated by means of hierarchical mathematical morphology procedures. Experimental results demonstrate that the proposed clustering algorithm clusters data better than the classical clustering algorithms, and an optimal number of clusters could be found. 2006 Vol. 19 (6): 727-733 [Abstract] ( 278 ) [HTML 1KB] [ PDF 1108KB] ( 489 ) 734 Modeling Based on Hybrid Radial Basis Function Neural Networks and Its Backward Model Control CHEN ZongHai, YUAN MingZhe, XIANG Wei, ZHANG YanWu Traditional control methods are not satisfactory in more and more complex process control, and the generalization ability of neural networks in control is weak. In this paper, a novel structure, the combinations of the process fundamentals and RBFNN is presented to direct the neural network convergence and exert the excellent capability on nonlinear approach of neural networks. Simulation results show that the compute velocity of the backward model controller using the hybrid RBFNN, while the control precision index is ensured, is much higher than the backward model controllers using common RBFNN. The hybrid RBFNN backward model controller also has excellent control quality and shows good adaptation to disturbance, time delay, nonlinear and the drift of plant parameters. 2006 Vol. 19 (6): 734-738 [Abstract] ( 230 ) [HTML 1KB] [ PDF 504KB] ( 292 ) 739 Feature Selection Based on NeuroFuzzy Networks SANG Nong, XIE YanTao, GAO RuXin, ZHANG TianXu To some extent, the feature selection algorithms based on artificial neural networks can be regarded as the special cases of the architecture pruning algorithms. However, they usually require preprocessing of data normalization, which may change the distribution of the original data which is important to the classification. Neurofuzzy networks are fuzzy inference systems with selfstudy ability. In this paper it is combined with the architecture pruning algorithm based on membership space and a new feature selection algorithm is proposed. The membership functions of the algorithm are learned adaptively, and the learning process is finished before the feature selection. Experiments on natural and synthesized data are given and compared with some traditional techniques. The results show that the proposed method is superior to the traditional ones. 2006 Vol. 19 (6): 739-745 [Abstract] ( 261 ) [HTML 1KB] [ PDF 544KB] ( 475 ) Surveys and Reviews 746 Pattern Discovery in Complex System: Review of Epsilon Machine XIANG Kui, JIANG JingPing System complexity means hybridity and emergence. Pattern discovery is to find the hidden pattern of complex system which is a new way to analyze and understand the complex system. Epsilon machine is an achievement of theoretical physics, which could discover the hidden pattern of the process by formal language. In this paper, the basic theory of epsilon machine is firstly presented and its properties and merits are summarized. Epsilon machine has two reconstruction algorithms: subtree merging and causal state splitting reconstruction which are compared in this paper. Then, a simple example about reconstruction of even process is given. As a measurement of nature structure, statistical complexity and its computation are introduced based on epsilon machine reconstruction. Finally, all the progress and application of epsilon machine in the past are reviewed, and the recommendation of the future research is given. 2006 Vol. 19 (6): 746-752 [Abstract] ( 358 ) [HTML 1KB] [ PDF 439KB] ( 427 ) Researches and Applications 753 Study of Dynamic OD Matrix Estimation Algorithm and Applications LI Jie, CHEN Feng, WANG JiaJie OriginDestination(OD) matrix is one of the most important research fields in the intelligent traffic. It plays an important role in the microscopic traffic simulation and urban traffic plan, management and control. By combining the microscopic traffic simulation with the traffic control system, a dynamic OD matrix estimation algorithm based on the maximum entropy method is proposed. The dynamic variation of the traffic flow in the roadway segment is analyzed, and a recursion formula of the historical OD matrix is given to enhance the estimation accuracy of OD matrix. The simulation experiment shows the proposed approach is effective and practical. 2006 Vol. 19 (6): 753-757 [Abstract] ( 297 ) [HTML 1KB] [ PDF 373KB] ( 711 ) 758 A Method for Shape Recognition CHEN XiaoChun, YE MaoDong, NI ChenMin Any kind of the shapes corresponds to one set of curves which forms the curve equivalence classes based on an equivalence relation. In this paper, the curve equivalence classes are used to describe the shapes. And different kinds of shapes are mapped to the curve equivalence classes. The shape description is invariant to the translation, scale, and rotation. Based on the description, the distance between the curve equivalence classes is defined which reflects the similarity between the shapes. This method for the shape recognition has immunity to noise and small perturbations. 2006 Vol. 19 (6): 758-763 [Abstract] ( 240 ) [HTML 1KB] [ PDF 459KB] ( 435 ) 764 Airport Runway Marking Detection and Identification of Unmanned Landing Vehicle Based on Vision WANG HongQun, PENG JiaXiong, LI LingLing In this paper, the proplem how to detect and recognize the airport runway marking in images is discussed. These images are obtained by a landing unmanned air vehicle (UAV) based on vision. Some spots of high light are extracted and grouped into several small clusters using a special clustering algorithm. An identification model is constructed using the perspective model, the rectangular flatplate constraint and the prior views of the scene constraint conditions. Then the runway spots in these clusters using this identification model can be datected. The computational time is greatly cut down by this classification process. The experimental results show that the proposed algorithm is very suited to the landing of UAV. 2006 Vol. 19 (6): 764-770 [Abstract] ( 229 ) [HTML 1KB] [ PDF 1098KB] ( 407 ) 771 An Algorithm for Text Extraction in Complex Color Image LIU XinXing, WANG ZengFu In this paper, a text extraction method for the complex color image with image enhancement by Symmetric Neighborhood Filters(SNF) is proposed. Firstly, the original image is enhanced by SNF and the generate edge feature map on the enhanced image is obtained. The candidate text regions are then generated by merging the bounding blocks which are extracted by utilizing the edgebased connectedcomponent searching method and taking the edgedivided color information into consideration. Finally, the texture and stroke feature of the text are used to eliminate the false candidates. Experimental results show this method can extract the text including Chinese and English characters accurately. It is the key point to ensure the success of extraction method that the original image is enhanced by SNF because SNF not only smoothes the interior of the image but keeps the true edges of the image. 2006 Vol. 19 (6): 771-775 [Abstract] ( 251 ) [HTML 1KB] [ PDF 492KB] ( 716 ) 776 Improvement of Speaker Identification Performance Using Nonlinear Features HOU LiMin, DENG DeChun, WANG ShuoZhong Chaotic characteristics in speech by calculating the maximum Lyapunov exponents of 38 Mandarin phonemes are presented. The physical significance of three nonlinear features of human speech, i.e. the largest Lyapunov exponent, the secondorder dynamical entropy, and the fractal dimension, is studied. A speaker recognition system based on the Gaussian mixture model is established. On the decision layer, the recognition results obtained from MFCC and nonlinear dynamics are combined in a serial manner to give an improved performance. The experimental result shows nonlinear dynamics coefficients can distinguish different speaker and aid speaker identification only by MFCC features. 2006 Vol. 19 (6): 776-781 [Abstract] ( 221 ) [HTML 1KB] [ PDF 793KB] ( 375 ) 782 Left Ventricle MRI Images Segmentation by Unifying Level Set Method and Snake Model with Shape Constraints ZHOU ZeMing, YOU JianJie, FAN ChunHui, Pheng Ann Heng, XIA DeShen A segmentation algorithm for the left ventricle MRI images is proposed by unifying the level set method and the snake model with shape constraints. Due to the weak edges and the low contrast regions, the deformation curve leaks from the outer boundary of the left ventricle when the snake model is used to segment the MRI image of the left ventricle. After the training samples have been aligned and the variation modes have been analyzed, the allowable shape space of the left ventricle is constructed. According to the properties of the cardiac MRI images, the shape constraint energy field around the average shape is created by the level set method. After the shape constraint energy term is added, the snake model can effectively prevent the deformation curve from leaking out of the low contrast regions. The evolving curve is subject to the shape constraints by mapping it to the allowable shape space. The segmentation experiments demonstrate the effectiveness of the proposed model. 2006 Vol. 19 (6): 782-786 [Abstract] ( 269 ) [HTML 1KB] [ PDF 1183KB] ( 410 ) 787 Target Tracking Based on Mean Shift Algorithm and Particle Filtering Algorithm MA Li, CHANG FaLiang, QIAO YiZheng Having been improved, the mean shift algorithm and the particle filtering algorithm are combined effectively. Under the nonocclusion and the occlusion that is not serious the improved mean shift algorithm is adopted while under the serious occlusion the improved particle filtering algorithm is employed. Whether the real tracking is resumed is checked after occlusion. Effective occlusion detection method based on subblock is proposed and the color template is not updated under occlusion. Experimental results indicate the proposed algorithm is realtime and robust and has good tracking performance under complex background. 2006 Vol. 19 (6): 787-793 [Abstract] ( 281 ) [HTML 1KB] [ PDF 2906KB] ( 352 ) 794 A Mixed Strategies Pareto Evolutionary Programming DONG HongBin, HUANG HouKuan, HE Jun, HOU Wei , MU ChengPo A evolutionary approach to solve the multiobjective optimization problems, Mixed Strategies Pareto Evolutionary Programming (MSPEP), is presented. Based on the performance of mutation strategies, the mixed strategy distribution is dynamically adjusted. By combining the Pareto strength ranking procedure with the mixed mutation strategies, a new evolutionary algorithm is proposed. The proposed approach is compared with other evolutionary optimization techniques in several benchmark functions. Experimental results demonstrate that the proposed method could rapidly converge to the Pareto optimal front and spread widely along the front. 2006 Vol. 19 (6): 794-800 [Abstract] ( 189 ) [HTML 1KB] [ PDF 533KB] ( 365 ) 801 Design and Implementation of OffLine Handwritten Document Recognition System of Manchu Manuscript ZHAO Ji, LI JingJiao, ZHANG GuangYuan, WAN Jie Based on an offline handwritten Manchu manuscript recognition system, a corresponding system model is established. Firstly, the digital image processing method is used to preprocess and extract the words from the identified targets. Next, the processed words are decomposed into the stroke units. The statistics pattern recognition method is employed to identify them and obtain the stroke sequence. Then the stroke sequence is converted into the root sequence. The fuzzy identification method is used to achieve the output of ManchuRoman characters. Hidden Markov Model method is also involved to postprocess the recognition results of every single word and enhance the recognition rate. The experimental results show that the recognition rate and the selfadaptability of the system are increased substantially on the basis of the single font stroke learning and probability statistics of great corpus of twoword simultaneity. 2006 Vol. 19 (6): 801-805 [Abstract] ( 249 ) [HTML 1KB] [ PDF 503KB] ( 491 ) 806 Fingerprint Image Enhancement by PseudoLinear Directional Diffusion PDE ZHU LiXin, OUYANG XiaoLi, XIA DeShen A pseudolinear directional diffusion filter is proposed to enhance the fingerprint images according to their directional infomation. In this paper, the defects of coherent diffusion expressed in a divergence form are discussed which produce a artificial structure and a new nonlinear directional diffusion PDE is proposed to overcome its drawbacks. The new algorithm outperforms the coherent diffusion based image enhancement in denoising ability and structure information preservation. Finally, a pseudolinear expression of the method is proposed and discussed to decrease the computational burden. The pseudolinear method is more suitable to the realtime fingerprint recognition system because of its efficiency. 2006 Vol. 19 (6): 806-811 [Abstract] ( 221 ) [HTML 1KB] [ PDF 1639KB] ( 486 ) 812 EOM Face Detection Method Based on Real AdaBoost Algorithm CHEN HuaJie, WEI Wei A Real AdaBoost algorithm based EOM (edgeorientation matching) method is proposed for face detection. The edge orientation feature is extracted from the original face images to eliminate the influence of some disturbances such as variable lighting to a certain extent. The global face pattern (global feature point set) is obtained by using the Real AdaBoost algorithm through multiple iterative learning procedures, and the local pattern (local feature point set) is acquired by utilizing the areaselecting strategy during each iterative procedure. A precise face pattern is found by the proposed method rather than by the original EOM method, which is confirmed by the experiment of frontal face detection. 2006 Vol. 19 (6): 812-817 [Abstract] ( 309 ) [HTML 1KB] [ PDF 1370KB] ( 486 ) 818 Algorithm for Car License Plate Location Based on Fractal Dimension of Waveforms ZHUGE Bin, ZHOU HeQin According to the distribution of the characters on a car license plate (CLP), a new location method based on distribution of fractal dimension of waveforms (FDW) is proposed in this paper. Using the method of sliding windows, the FDW for each line gray curve of every photo is estimated and two dimension distribution of FDW are gotten. Through processing the distribution of FDW, the CLP zone can be segmented easily. The experimental results show that FDW is effective for the attribute of the characters alignment on CLP and it is not affected by the background, the color and the type of CLP. Besides, the proposed algorithm is not sensitive to the sloping license plate and the accuracy rate of location for 1000 complicated background photos reaches 98. 9%. 2006 Vol. 19 (6): 818-824 [Abstract] ( 203 ) [HTML 1KB] [ PDF 822KB] ( 507 ) 825 Object Tracking Using Particle Filter Based on Mean Shift LIU ZhiMing, WEI Wei A novel particle filter based on mean shift is presented. The standard particle filter is supposed to track a target in a whole state space. The proposed algorithm decomposes the target state space into two independent subspaces: a displacement subspace and a deformation subspace. Mean shift algorithm embedded in the particle filter is utilized to track the state transfer in displacement subspace while the particle filter is used to track the transfer in deformation space. In this way, the particle filter tracks the target in a lower order subspace and thus its realtime performance is improved. On the other hand, the mean shift algorithm is granted a new capability to track the target deformation. The validity and efficiency of the new algorithm are demonstrated by a series of real time tracking experiments. 2006 Vol. 19 (6): 825-830 [Abstract] ( 222 ) [HTML 1KB] [ PDF 1170KB] ( 373 ) 831 Genetic Gradient Algorithm Based RBF Neural Network and Its Applications to Traffic Information Prediction GUO Lin, FANG TingJian, YE JiaSheng The realtime and accurate predicted traffic information is critical to the intelligent traffic inducement and the traffic management. A twostep learning algorithm GGA (Genetic Gradient Algorithm) for radial basis function (RBF) neural network is proposed in this paper. A genetic algorithm (GA) initially determines the parameters of the RBF network including the number and locations of the selected centers and the widths of Gaussian kernel functions in the hidden layer. Then a gradient descent algorithm is adopted to further adjust these parameters of the RBF network. A smaller network with better generalization capability can thus be obtained. The experimental results of the realtime traffic information prediction in Ningbo city show good performance of the proposed method. 2006 Vol. 19 (6): 831-835 [Abstract] ( 212 ) [HTML 1KB] [ PDF 418KB] ( 394 )

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