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....

2008 Vol.21 Issue.2, Published 2008-04-01 Papers and Reports    Researches and Applications    Surveys and Reviews        Papers and Reports 129 Fuzzy Systems Based on POFI Method and Their Response Ability PENG JiaYin Using the pointwise optimization fuzzy inference method (POFI), fuzzy systems and their response functions are studied based on 6 commonly used fuzzy implication operators. How to implement FMP algorithm and FMT algorithm with these fuzzy implication operators is presented on the basis of POFI method respectively. Then, it is pointed out that fuzzy systems based on POFI method with 6 fuzzy implication operators have no interpolation property, and their fuzzy controllers only have step output property. Furthermore, they are not universal approximators. 2008 Vol. 21 (2): 129-135 [Abstract] ( 214 ) [HTML 1KB] [ PDF 306KB] ( 333 ) 136 Balance Method for Imbalanced Support Vector Machines LIU WanLi, LIU SanYang, XUE ZhenXia An adjustment method is proposed for the separation hyperplane of binaryclassification imbalanced data. Firstly, the original samples are preliminarily trained by the standard support vector machines, and a normal vector of the separation hyperplane is obtained. Secondly, onedimensional data are generated by projecting the high dimensional data onto the normal vector. Then, the ratio of the twoclass penalty factors is determined based on the information derived from the standard deviation of the projective data and the twoclass sample sizes. Finally, a new separation hyperplane is presented by the second training. Experimental results show the efficiency, i.e., the two error ratios can be balanced and even be decreased generally. 2008 Vol. 21 (2): 136-141 [Abstract] ( 245 ) [HTML 1KB] [ PDF 354KB] ( 337 ) 142 A Heuristic Algorithm for Global Partial Order Mining WANG JinLong, XU CongFu Sequential pattern mining is an important data mining research topic. In this paper the global partial order algorithm is firstly analyzed. Then, a heuristic algorithm is proposed for improving the process of global partial order construction. By using the local sequence pattern information, the problem of constructing the partial order model with high mathematical complexity can be avoided, and accurate results can be obtained. Finally, the efficiency and accuracy of the proposed method are validated by the experimental results on synthetic and real dataset. 2008 Vol. 21 (2): 142-147 [Abstract] ( 263 ) [HTML 1KB] [ PDF 535KB] ( 485 ) 148 A Hybrid Learning Algorithm for Elliptical Basis Function Neural Networks XING HongJie, WANG Yong, HU BaoGang A hybrid learning method for the elliptical basis function neural network (EBFNN) is presented. Firstly, the parameters of elliptical basis function (EBF) units in the hidden layer of the EBFNN are initialized by the expectationmaximization (EM) algorithm, while the connection weights plus bias term is initialized by the linear leastsquared method. Then, the gradient descent based optimization procedure adjusts all the parameters simultaneously. The comparison results show that the gradient descent elliptical basis function neural network (GDEBFNN) trained by the proposed hybrid learning method upon the test datasets has higher accuracy than the other three related models. Compared with support vector machine (SVM), the GDEBFNN can achieve comparable generalization ability. Moreover, the GDEBFNN obtains better generalization performance than the decision tree constructed by the Adaboost method. 2008 Vol. 21 (2): 148-154 [Abstract] ( 324 ) [HTML 1KB] [ PDF 859KB] ( 534 ) 155 A Fast Infrared Image Segmentation Method Based on TwoDimensional Entropy and Particle Swarm Optimization Algorithm LIU YiTong, FU MengYin The computation consuming of 2D maximum entropy method is often an obstacle in the image segmentation. In this paper logarithm is replaced by subtraction and the threshold vector is obtained by using a new optimization algorithm. The new algorithm is proposed to realize the 2D maximum entropy method instead of exhaustive search method, thus it is faster than the traditional method. The proposed method has been proved to be efficient through the example for segmenting the infrared image. 2008 Vol. 21 (2): 155-159 [Abstract] ( 272 ) [HTML 1KB] [ PDF 445KB] ( 357 ) 160 Thermal Infrared Face Image Recognition Based on PCA and LDA HUA ShunGang, ZHOU Yu, LIU Ting A method for infrared face recognition is proposed based on principal component analysis (PCA) and linear discriminant analysis (LDA). According to the characteristics of infrared face images, a set of normalized infrared face images is gotten by preprocessing. The dimensionality of the image vector is reduced and the global features are extracted. The global features are used to generate a classifier which can minimize the withinclass scatter and maximize the betweenclass scatter. Finally, an infrared face recognition experiment based on the combination of PCA and LDA is performed and the results show the high performance of the proposed method. 2008 Vol. 21 (2): 160-164 [Abstract] ( 351 ) [HTML 1KB] [ PDF 629KB] ( 956 ) 165 An Improved Segmentation Method for Cytoskeleton Images ZHOU HongQiong, WANG ZengFu, LIN WanHong, DING Bai It is a key step to quantitatively describe the images of the cytoskeleton in the analysis of cell morphology. However, to describe the cytoskeleton accurately, the areas which only include microfilaments or microtubules must be separated from the cytoskeleton image. Based on the ChanVese method and the Otsu method, an improved method for the segmentation of cytoskeleton images is introduced. The method realizes the segmentation of cytoskeleton images by combining the ChanVese method and the Otsu method which has stable and fast performance. The experimental results show that compared with the ChanVese method the improved method reduces the time significantly. 2008 Vol. 21 (2): 165-170 [Abstract] ( 237 ) [HTML 1KB] [ PDF 1050KB] ( 383 ) 171 A Feature Detection Method Based on Harris Corner and Difference of Gaussian GAO Jian, HUANG XinHan, PENG Gang, WANG Min, WU ZuYu Aiming at the heavy computation burden and poor realtime performance of the existing scaleinvariant feature detection methods, a incomplete pyramid frame in scale space is presented. Its influence on the performance of the method is analyzed in theory. Then, a quick feature detecting method is presented based on Harris corner and difference of Gaussian. It computes the Harris corners at each level in incomplete pyramid scale space of the image and the difference of Gaussian is used to select the feature points. The method not only can ensure high performance but also decrease the computation time. Its validity has been proved by the experiment. 2008 Vol. 21 (2): 171-176 [Abstract] ( 231 ) [HTML 1KB] [ PDF 420KB] ( 456 ) Surveys and Reviews 177 Swarm Robotics: A Survey XUE SongDong, ZENG JianChao The field of swarm robotics is surveyed from a computerscience perspective. Both the historical basis and the current research of the field are summarized. Swarm robotics is a novel discipline and approach which is inspired by social insects. By this approach, large numbers of relatively simple robots can be controlled to deal with prescribed complex tasks in an economical, robust, flexible and scalable fashion. In this paper, the key definitions of this newly emerging method are introduced and a set of criteria is presented which can be used to distinguish swarm robotics research from other multirobot studies. Then, the desirable properties of swarm robotics in the systemlevel function are described. Finally, central issues of swarm robotics are discussed including interaction between individual robots, communication, cooperation and control, selforganizing and selfassembling etc. 2008 Vol. 21 (2): 177-185 [Abstract] ( 338 ) [HTML 1KB] [ PDF 499KB] ( 1536 ) Researches and Applications 186 A Ranking SVM Based Algorithm for Automatic Extraction of Acronym GAO YongMei, HUANG YaLou, NI WeiJian, XU Jun A ranking SVM based method is proposed for automatically extracting acronyms and their corresponding expansions in free format text. A twostep ranking model is established, including local extraction and global ranking. For each extracted acronym, the model returns to an ordered list of expansion candidates, where the candidates are ranked according to their correctness and the degree of popularity. The ranking model can effectively eliminate noise and help user find the true expansions. Experimental results on real data validate its higher performance and better general adaptation in different domains than other methods. 2008 Vol. 21 (2): 186-192 [Abstract] ( 245 ) [HTML 1KB] [ PDF 572KB] ( 369 ) 193 Image Segmentation with Wavelet Transform Based on Spatial MultiResolution Analysis WANG ZhenHua, CHEN Jie, DOU LiHua A method based on spatial multiscale analysis is put forward. Firstly, an image is divided into subdomains with different sizes by analyzing the energy projection distribution on rows and columns of the image, and the sizes are adjusted adaptively according to both holistic and local features. Then, wavelet transform to each subimage is adopted and the statistics which reflect the local features are extracted. Finally, an improved fuzzy cmeans (FCM) method is adopted to cluster the eigenvectors, thus the image segmentation is realized. Experimental results have proved that the proposed method can speed up operation as well as improve the segmentation results. 2008 Vol. 21 (2): 193-198 [Abstract] ( 266 ) [HTML 1KB] [ PDF 666KB] ( 414 ) 199 An AntiGeometric Diffusion Classification Based Image Binarization Method HUANG Qian, WU Yuan, YIN JunXun It is difficult to extract objects from background due to the uneven and complex background information. In this paper, a binarization method is developed, which is based on the antigeometric diffusion, a special form of the anisotropic diffusion. The antigeometric diffusion method is used to blur and diffuse the edge of images as much as possible, and thus many threshold surfaces are formed. Each pixel is classified during the diffusion process according to the developed classification criterions. Finally, a postprocessing approach is proposed to extract the object from background. The numerical experimental results show that the presented method is robust to the noise restriction. Furthermore, the results for handling Xray images of casting products with uneven background by the presented method are given. 2008 Vol. 21 (2): 199-205 [Abstract] ( 207 ) [HTML 1KB] [ PDF 2143KB] ( 261 ) 206 Plant Leaf Identification Based on Radial Basis Probabilistic Neural Network DU JiXiang, WANG ZengFu In this paper, a plant species identification approach is proposed based on Gabor features and radial basis probabilistic neural network (RBPNN). The multiscale Gabor wavelet transform is applied to extract the texture features. The experimental results show that the RBPNN achieves high recognition rates and classification efficiency by using radial basis function neural network (RBFNN) for the plant species recognition and identification task. 2008 Vol. 21 (2): 206-213 [Abstract] ( 230 ) [HTML 1KB] [ PDF 635KB] ( 420 ) 214 Motion Segmentation Based on Region Shrinking and DIRECT Algorithm LI ZhiHui, HUANG FengGang Segmentation of motion image sequences is an important problem in computer vision. In this paper, maximizer of the posterior marginalsmaximum a posteriori (MPMMAP), is adopted based on Bayesian frame for motion segmentation. Firstly, the smoothness term of likelihood function in Bayesian frame is redefined. The region shrinking algorithm is used to estimate the supporting regions of moving objects during the iteration. Then a model is proposed which represents the affine motion with 6 parameters by the center and main axes of a region. Motion parameters are estimated merely by pixels on main axes and derived more quickly than before. The estimation is transformed into a kind of optimal problem with parameters in limited ranges, and DIRECT algorithm is used to compute the motion parameters. Compared with the traditional algorithms, the proposed method improves the accuracy and stability in motion parameter estimation. The results of simulated experiments show the effectiveness of the proposed method. 2008 Vol. 21 (2): 214-220 [Abstract] ( 225 ) [HTML 1KB] [ PDF 1051KB] ( 506 ) 221 A Fast FCM Cluster MultiThreshold Image Segmentation AlgorithmBased on Entropy Constraint TIAN JunWei, HUANG YongXuan, YU YaLin Aiming at the time consuming problem of traditional fuzzy cmean (FCM) cluster segmentation, a fast entropy constraint based fast FCM segmentation algorithm is proposed. The minimum resample ratio is studied, in which the sampled image can keep the most information of the initial image, and the limitation function of resample ratio is deduced. During calculation of the resample ratio, the relative entropy loss constraint of histogram is introduced to keep the sampled image out of serious distortion. Experiments are performed with entropy loss , and the results show that the multithreshold segmentation results of the proposed algorithm are almost as good as that of traditional FCM cluster segmentation algorithm. Moreover, the average consuming time of the proposed method is reduced greatly compared with traditional FCM algorithm, the calculational efficiency is increased obviously. The experiment results accord with the theory and the validity of the proposed algorithm is verified. 2008 Vol. 21 (2): 221-226 [Abstract] ( 207 ) [HTML 1KB] [ PDF 1251KB] ( 446 ) 227 Pattern Recognition of Hand Motions Based on HHT and ARModel LUO ZhiZeng, MA WenJie, MENG Ming In order to recognize the hand motions based on the surface electromyogram (SEMG), a feature extraction algorithm is presented which is built by the combination of HilbertHuang transform (HHT) and ARmodel. According to the frequencycredit of each intrinsic mode function (IMF) after HHT, six intrinsic mode functions (IMFs) are selected. In the meantime, the rectangle window is built to cut motion signals of the six IMFs based on the motionstart and the motionend points. The motionstart and motionend points are decided by the instantaneous amplitude of the IMF with the largest frequencycredit. ARmodel of each IMF is built to extract the handmotion features. Finally, the motionfeature vector processed by principal component analysis (PCA) is input into the SVM classifier to recognize the hand motions. The experimental results indicate that the proposed method can discriminate the four handmotion patterns (namely, palmar dorsiflexion and flexion, hand opening and closing) with the correct rate up to 91%. 2008 Vol. 21 (2): 227-222 [Abstract] ( 254 ) [HTML 1KB] [ PDF 688KB] ( 517 ) 233 A Supervised Locality Preserving Projection Algorithm for Dimensionality Reduction SHEN ZhongHua, PAN YongHui, WANG ShiTong Aiming at the unsupervised property of locality preserving projection (LPP), a linear dimensionality reduction method called supervised locality preserving projections (SLPP) is proposed, which integrates the locality preserving property in LPP and the class separability. Experimental results show SLPP is superior to some classical and recently presented methods. The linear SLPP method can also be extended to nonlinear dimensionality reduction scenarios by using the kernel method. 2008 Vol. 21 (2): 233-232 [Abstract] ( 401 ) [HTML 1KB] [ PDF 1084KB] ( 930 ) 240 Image Retrieval Method Based on ROI and MCS HAO HongWei, HUANG FangYi, ZHOU Jing Since there is no effective algorithm for semantic level region extraction, it is difficult to obtain the region that reveals users' retrieval purposes exactly. Besides, different retrieval purposes demand various features. In order to solve these problems, an image retrieval method is proposed based on region of interest (ROI) and multiple classifier systems (MCS). Firstly, ROI is selected by users. Then, different features of ROI are extracted for constructing corresponding classifiers. Finally, the retrieval results are obtained by combining the outputs of individual classifiers. Experimental results show that the proposed method can exactly grasp the retrieval users' purpose and greatly improve the precision of retrieval systems. 2008 Vol. 21 (2): 240-245 [Abstract] ( 277 ) [HTML 1KB] [ PDF 608KB] ( 375 ) 246 Mining Frequent Itemsets with Positive and Negative Items Based on FPTree ZHANG YuFang, XIONG ZhongYang, PENG Yan, ZHAO Ying Using the concept of frequent pattern tree of FP_growth, a new frequent pattern tree containing positive and negative items is constructed. The frequent itemsets with positive and negative items are mined through extending frequent patterns on the tree. Compared with the algorithms of directly using FP_growth, the proposed algorithm has no requirement for growing negative item to original database as well as the construction or destruction of additional data structures. Only some modifications to the original frequent pattern tree are needed. Therefore it has certain advantages in time and space costs. Experiments show that the algorithm has better efficiency than the existing mining algorithms and algorithms of directly using FP_growth. 2008 Vol. 21 (2): 246-253 [Abstract] ( 258 ) [HTML 1KB] [ PDF 480KB] ( 493 ) 254 A Method to Design Reinforcement Function Based on Fuzzy Rules in QLearning ZHAO XiaoHua, LI ZhenLong, CHEN YangZhou, RONG Jian Qlearning is a reinforcement learning method to solve Markovian decision problems with incomplete information. The design of reward function is an important factor that affects the learning results of Qlearning. A method to design the reward function of Qlearning based on fuzzy rules is introduced to improve the performance of reinforcement learning, and the method is applied to traffic signal optimal control. According to different traffic condition, the switching time and switching sequence of phase can be adapted. The performance of the system is evaluated by Paramics microcosmic traffic simulation software. And the results show that the learning effect of Qlearning based on fuzzy rules is better than that of conventional Qlearning for traffic signal control. 2008 Vol. 21 (2): 254-259 [Abstract] ( 286 ) [HTML 1KB] [ PDF 498KB] ( 421 ) 260 Application of RunLength Texture Features to SPOT Remote Sensing Image Classification CAO ZhiGuo, XIAO Yang, ZOU LaMei Combined with neural network, a method for remote sensing image classification based on runlength features is proposed. According to the criterion of variances between and intra classes, the efficient features are selected and the redundant ones are excluded successfully by the method of rough set. Runlength features, cooccurrence features, gray levelgradient cooccurrence features and gray levelsmoothed cooccurrence features are respectively used as inputs of three types of classifiers: BP net, RBF net and a nearest neighbor classifier-KNN method, when applying remote sensing classification for large scale panchromatic SPOT images with high spatial resolution. The result demonstrates the efficiency of the proposed algorithm. 2008 Vol. 21 (2): 260-265 [Abstract] ( 267 ) [HTML 1KB] [ PDF 839KB] ( 288 ) 266 Distributed Intelligent Management Model Based on Active Network SHEN MingYu, ZHANG YouSheng Based on the analysis of current network management and active network, an active network based distributed intelligent management model is presented. Multiagent system is introduced to characterize and describe the model. A management control message (MCM) has been designed for multiagent in the active network which is specially used for commission, management and control. In order to enhance the reliability of fault diagnosis, evidential theory is adopted to combine different fault symptoms for network fault diagnosis. The model has advantages of domain autonomy and cooperation across multiple domains, and it is suitable for largescale and heterogeneous network management. 2008 Vol. 21 (2): 266-230 [Abstract] ( 249 ) [HTML 1KB] [ PDF 576KB] ( 318 )

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