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.4, Published 2007-08-25 Papers and Reports    Researches and Applications        Papers and Reports 439 A Line Description and Detection Method Based on Combination of 3Pixel Elementary Units WEI Hui, LIU Bin One of the most important clues to image understanding is the edges and profiles of the objects, and lineedges take a high proportion. By applying edge detectors to image, only some discrete points are obtained, and line detection can integrate them. A method of line description and line detection is proposed, which is based on combinations of multiple threepixel micro patterns. This method includes defining primary units that can be combined into dentate line segment in grid manner, establishing the combination rules of different type units in different positions, and proving the feasibility of the combination rules. Then a unitbased clustering algorithm is given. Finally, this method is tested on many real pictures with background, and compared with other classical line detection algorithms. The experimental results show the proposed method has made great improvements over computation time, memory requirement and detection accuracy. The output of the proposed method can be taken as the input of subsequent object recognition procedures directly, and searching line end points, filtering false lines, segmenting collinear line segments are needless. 2007 Vol. 20 (4): 439-449 [Abstract] ( 245 ) [HTML 1KB] [ PDF 1685KB] ( 414 ) 450 Rough Set Model Based on Characteristic Relations in Incomplete Information System YANG XiBei , YANG JingYu , WU Chen , YU DongJun In this paper, the generalized incomplete information system is taken into consideration, in which unknown values are considered as not only absent but also missing. In this kind of generalized incomplete information system, the characteristic relation is deeply investigated. The classification analysis based on the characteristic relation is discussed due to the two possible unreasonable situations. Therefore, two kinds of new characteristic relations, named as type Ⅰand type Ⅱ, are proposed respectively. Then, characteristic relation of type Ⅲ is constructed by combining the advantages of the previous two new characteristic relations. Finally, an illustrative example is analyzed to indicate the validity of the three new characteristic relations when dealing with the generalized incomplete information systems. 2007 Vol. 20 (4): 450-457 [Abstract] ( 289 ) [HTML 1KB] [ PDF 349KB] ( 365 ) 458 Adaptive Parallel Ant Colony Optimization Algorithm YAO BaoZhen Ant colony optimization algorithm is a new simulated evolutionary algorithm, which has the faculty of global optimization. An adaptive parallel ant colony optimization algorithm (APACO) is presented. It dynamically adjusts the parameters according to the searching phases, thus the convergence is accelerated to a certain extent. The adaptive migration rule could not only enrich the diversity of the colonies but also reduce the communication between colonies. Finally, the CHN144 problem of China in Nonnumerical Parallel Algorithm: the Simulated Annealing Algorithm by Lishan Kang is applied to calibrate the algorithm. Results show that the proposed algorithm improves the searching speed with good global convergence. 2007 Vol. 20 (4): 458-462 [Abstract] ( 324 ) [HTML 1KB] [ PDF 405KB] ( 391 ) 463 Detection of Outlier Samples in Multivariate Time Series WENG XiaoQing , SHEN JunYi Multivariate time series (MTS) datasets are commonly used in the fields of finance, multimedia and medicine. MTS samples, namely outlier samples, are significantly different from the other MTS samples. In this paper, a method for detecting outlier samples in the MTS dataset based on local sparsity coefficient is proposed. An extended Frobenius norm is used to compare the similarity between two MTS samples, and knearest neighbor (kNN) searches are performed by using twophase sequential scan. MTS samples that are not possible outlier candidates are pruned, which reduces the number of computations and comparisons. Experiments are carried out on two realworld datasets, stock market dataset and BCI (Brain Computer Interface) dataset. The experimental results show the effectiveness of the proposed method. 2007 Vol. 20 (4): 463-468 [Abstract] ( 311 ) [HTML 1KB] [ PDF 603KB] ( 504 ) 469 Automatical Extraction of Isotherm from PhaseChange Thermography Sequence WANG XiaoNian, FENG ZuRen, ZHANG Miao A new segmentation method of image sequence is proposed to get the isotherm from phasechange thermography sequence. The original phasechange thermography sequence is transformed into a series of relative synthesized images by compression and conversion. The final isotherms are extracted from the segmentation of synthesized images. To remove the specular reflection and get the results which meet the need of similarity, the illumination model is formulated and the parameters are obtained by coordinationoptimization. Then the dynamic programming algorithm is employed to get the optimal segmentation. The proposed method could not only remove the specular reflection of the characteristic images, but also reduce the computational complexity. Moreover, it augments the robustness of the algorithm. The eventual results are rational and they could also demonstrate the efficiency of the proposed method. 2007 Vol. 20 (4): 469-477 [Abstract] ( 274 ) [HTML 1KB] [ PDF 1886KB] ( 394 ) 478 A TwoStage Tabu Search Algorithm Based on Rough Set Theory LI Fan, LIU QiHe, YANG GuoWei A twostage tabu search algorithm based on rough set theory is proposed for combinatorial optimization problems which are represented by TSP. For most of the adaptive tabu search algorithms, the balance between intensification and diversification is achieved by tuning tabu search parameters dynamically. Unlike them, a twostage search strategy is used in the proposed approach. The aim of the first stage is diversification. In this stage, the search area is stimulated to move away from the initial solution, and the whole solution space is explored to a certain degree. Then, based on the solutions obtained in diversification, a promising area decision table is constructed and the corresponding promising area is found. The goal of the second stage is intensification. In this stage, the search procedure begins with the best solution which contains the promising area. In the search procedure, the selection of the new current solution is limited so as to utilize the useful information obtained in the first stage. The proposed algorithm is tested by TSP benchmark problems. The results show that it is feasible and effective. 2007 Vol. 20 (4): 478-484 [Abstract] ( 320 ) [HTML 1KB] [ PDF 552KB] ( 451 ) 485 An Automatic Approach to Lip Localization, Contour Extraction and Tracking WANG XiaoPing , HAO YuFeng , FU DeGang , YUAN ChunWei ;An automatic approach to lip localization, contour extraction and tracking is presented, which combines CbCr color space, Fisher transform and deformable templates. Firstly, a skincolor model is constructed for skin detection in CbCr color space so that the approximate lip region can be obtained based on the geometry features of human face. Then, the color difference between lip and skin is enhanced by Fisher transform. Preprocessing of brightness is carried out before segmentation, and the threshold is obtained by Otsu method. Next, the lip color model is used to validate the segmentation result of accurate localization and the deformable templates are used for lip contour extraction. Based on the segmentation result, a method of curves fitting for edges of inner mouth is presented to extract the inner contour robustly. Finally, locating result of previous frame is predicted as the next lip region, in which lip localization and contour extraction are executed for lip tracking. 2007 Vol. 20 (4): 485-491 [Abstract] ( 325 ) [HTML 1KB] [ PDF 1460KB] ( 950 ) 492 Human Detection Based on Support Vector Machine of Adaptive Gaussian Kernel for Indoor Application HU ChunHua, MA XuDong, DAI XianZhong Human detection is fundamental for human localization, recognition, and tracking. And it is still a difficult problem because of environment complexity and vision system movement. A new human detection algorithm based on mobile robot vision system is proposed to solve the problem effectively. The approach consists of the following two steps: (1)Waveletbased multiscale edge detection combined with edgelinked operator method is introduced to extract the edges of images. In this image a new morphology method is employed to get the object contourclosed for improvement of the correct recognition rate. And invariant Hu moments are calculated as pattern features vectors. (2)The adaptive Gaussian kernel soft margin support vector machine (CSVM) classifier is designed to distinguish human images from nonhuman ones. Experimental comparisons have been conducted,including adaptive Gaussian CSVM classifiers based on different features and the classifiers with different classification methods. The results validate effectiveness and robustness of the algorithm. 2007 Vol. 20 (4): 492-498 [Abstract] ( 303 ) [HTML 1KB] [ PDF 397KB] ( 377 ) 499 A Fast Computation Method for λ in RealTime Fuzzy Integral Fusion HE ShengPing , QIN Zheng The efficiency of computing λ plays a very important role in realtime fuzzy integral fusion. A numeric solution to obtain λ with Newton iterative method is discussed which could satisfy the realtime fuzzy integral fusion. By analyzing the range of λ and the relation between λ and fuzzy set ｛gi｝ under general case, some useful conclusions are derived. A concise numeric solution is proposed for computing f(λk) and f′(λk). Some numerical examples are carried out which show the efficiency and the excellence performance of the proposed algorithm. 2007 Vol. 20 (4): 499-504 [Abstract] ( 299 ) [HTML 1KB] [ PDF 301KB] ( 408 ) 505 Online Series Pattern Detection Based on Advanced Segmental SemiMarkov Model LING GuangJie, QIAN YunTao, JIA Sen Efficient online detection of similar patterns under arbitrary time scaling is a challenging problem in time series data mining. A modelmatching based segmental semiMarkov model is improved by introducing offset distribution, amplitude difference distribution and prepattern state. It overcomes the parameter estimation difficulty and the lack of robustness. The experimental results demonstrate that the advanced segmental semiMarkov model could rapidly and precisely detect scaling similar patterns under arbitrary time. 2007 Vol. 20 (4): 505-511 [Abstract] ( 239 ) [HTML 1KB] [ PDF 432KB] ( 427 ) Researches and Applications 512 Mining Algorithms of NMost Frequent Itemsets CHEN XiaoYun , HU YunFa The computing complexity of the frequent itemsets mining algorithm and the number of frequent itemsets are increased exponentially with the number of items in a transaction set. The minimum support threshold becomes a key to control such an increase. However, in practical application it will be difficult to control frequent itemsets scale, if only support threshold is used. The problem of Nmost frequent itemsets is introduced, and the breadthfirstsearch algorithm NApriori and the depthfirstsearch algorithm IntvMatrix based on the dynamic minimum support threshold are presented to solve the problem. Experimental result shows the proposed algorithms are faster than nave method, and the improvement of the speed is remarkable when N is low. 2007 Vol. 20 (4): 512-518 [Abstract] ( 261 ) [HTML 1KB] [ PDF 457KB] ( 432 ) 519 Distributed WEB Information Retrieval Based on Link Partition ZHANG Gang , WANG Bin , WU LiHui Distributed information retrieval is an effective way for large scale WEB information retrieval. A link based clustering algorithm ( LIBCA) is proposed for document partition. The BloomFilter Algorithm is selected to improve the efficiency of LIBCA. CORI collection selection algorithm and OKAPI BM25 are used in the process of distributed information retrieval. Based on TREC WEB dataset for the recent three years, a performance comparison is performed among the methods of link based distributed information retrieval, centralized retrieval, and random based distributed information retrieval. The experiment indicates that at P@10 the results of link partition based distributed WEB information retrieval are equal or even better than that of centralized retrieval. The efficiency experimental results indicate that the LIBCA plus BloomFiltern achieves a high system performance and it can deal with large dataset. 2007 Vol. 20 (4): 519-524 [Abstract] ( 285 ) [HTML 1KB] [ PDF 350KB] ( 432 ) 525 MultiAgent Dynamic Influence Diagrams and Its Approximation of Probability Distribution YAO HongLiang, WANG Hao, ZHANG YouSheng, YU Kui MultiAgent dynamic influence diagrams (MADIDs) are presented by extending MultiAgent Influence Diagrams (MAIDs) over time. Thus the structural relationships of coordination can be represented in dynamic environment. With the guidance of the strategic relevance, a decomposition approximation method of probability distribution and the approximation of probability distribution in inference are discussed to compute the probability distribution of MADIDs efficiently. The complexity, inducing error and error propagation over time are analyzed. Furthermore, based on the KLdivergence, a function is introduced to establish equilibrium between the precision and the complexity of approximate distribution. Finally, the experimental results on a dynamic coordination model show the validity of the probability distribution approximation method. 2007 Vol. 20 (4): 525-532 [Abstract] ( 225 ) [HTML 1KB] [ PDF 527KB] ( 551 ) 533 An Affine Transform Invariant Watermarking Algorithm Based on Secret Sharing and High Order Spectra ZHANG Li, XIAO WeiWei, JI Zhen An affine transform invariant blind watermarking technique is proposed based on secret sharing and high order spectra in this paper. Watermark is divided into n shadows to enhance security, according to secret sharing scheme. only t or more than t shadows can reconstruct the secret watermark. The watermark is embedded in the rotation, scaling, and translation invariant vector of bispectrum. Bispectrum detection is utilized to detect whether the test image contains the watermark or not. Even with small signal noise ratio, the detector can get a high detection probability. The independent component analysis is adopted, and thus the shadow could be detected as well as extracted. Before watermark reconstruction, one way hash function is employed to withstand cheating attacks. The experimental results show that the proposed algorithm is robust against a variety of attacks by watermark test software-Stirmark. 2007 Vol. 20 (4): 533-537 [Abstract] ( 284 ) [HTML 1KB] [ PDF 300KB] ( 357 ) 538 Mining Method of Frequent Patterns Based on Temporal Constraints DU Yi , LU DeTang , LI DaoLun , WEI WuZhou Temporal data is a kind of useful information. Temporal attributes in data can be used to find some potential change rules of data and predict the possible tendency. An algorithm, Temporal Frequent Pattern mining algorithm (TemFP) is presented. According to the existing function of temporal query, a Double B+tree for storing time attributes of frequent patterns is described. Using a temporal frequent pattern tree including the Double B+tree, the queries of temporal rules, which defined by users, could be realized rapidly. Experimental results demonstrate that the proposed algorithm is efficient and scalable. 2007 Vol. 20 (4): 538-544 [Abstract] ( 249 ) [HTML 1KB] [ PDF 453KB] ( 560 ) 545 Feature Extraction and Retrieval of Personal ID Photo ZHU QiuYu, HUANG SuJuan, WANG ShuoZhong Feature extraction from ID photos is discussed to retrieve subjects in ID photo database. By an adaptive threshold, a general skin color detection algorithm is modified for face region segmentation. The extracted face region is then segmented into individual organs by using the proposed projection DCT coefficients. Thus organ feature points are obtained. To improve the face description ability, the parameters of cheek profile are added to feature set by polynomial curve fitting. A photo retrieval system is implemented using the described methods and satisfactory results are acquired. 2007 Vol. 20 (4): 545-551 [Abstract] ( 242 ) [HTML 1KB] [ PDF 880KB] ( 733 ) 552 Proximal Support Vector Machines for Samples with Unbalanced Classification TAO XiaoYan , JI HongBing , Dong ShuFu Aiming at the problem that unbalanced data classification is disregarded in the standard Proximal Support Vector Machines (PSVM), a modified PSVM algorithm is presented, namely MPSVM. The different penalty factors are assigned to the positive and negative training sets according to the unbalanced population. The penalty values are transformed into a diagonal matrix. Then the decision functions for the linear and nonlinear MPSVM are achieved. Finally, the comparisons of algorithmic principle and performance are drawn. The experimental results show that MPSVM has a better generalization performance than PSVM and higher efficiency than the unbalanced SVM. 2007 Vol. 20 (4): 552-557 [Abstract] ( 223 ) [HTML 1KB] [ PDF 353KB] ( 526 ) 558 Hierarchical Fuzzy MinMax Clustering Algorithm YANG Jing , GAO Jun , XU XiaoHong , LIU Xu Clustering is considered as the most important problem of unsupervised learning. A Hierarchical Fuzzy MinMax Clustering Algorithm (HFMM) is presented based on the original Fuzzy MinMax Clustering Neural Network (FMMCN) and hierarchical clustering. Compared with the existing methods for clustering, the proposed algorithm dynamically determines the number of clusters to meet the demands of the problem. Moreover it overcomes the shortcomings of FMMCNorder dependent. Experimental results on three databases demonstrate that HFMM has high clustering performance. 2007 Vol. 20 (4): 558-564 [Abstract] ( 292 ) [HTML 1KB] [ PDF 411KB] ( 320 ) 565 Algorithm for Projective Reconstruction with Occlusions LIU ShiGang , WU ChengKe , LI LiangFu , PENG YaLi An algorithm for projective reconstruction with occlusions is presented. In this algorithm the reprojective points replace all the occlusion points, thus projective reconstruction is obtained. After several iterations, the positions of the occlusion are found and the accurate projective reconstruction could be finished. The innovation of the algorithm is that images and image points are treated uniformly. The experimental results of both simulated data and realworld data show that the algorithm is efficient and robust and it has a good property of convergence with small reprojection errors. 2007 Vol. 20 (4): 565-570 [Abstract] ( 244 ) [HTML 1KB] [ PDF 1162KB] ( 375 ) 571 Research and Implementation of Parallel Algorithm for GraphBased Image Segmentation YING WeiQin , LI YuanXiang , XU Xing , WANG LingLing A parallel solution of the graphbased method is proposed to improve the segmentation speed. In this solution, the similarity computation is parallelized by means of grid partition. And a parallel Lanczos algorithm is designed to compute the eigenvalues in view of the sparseness of the similarity matrix and the inner parallelism of matrixvector multiplication. The experimental results under MPI environment show that the parallel solution effectively improves the realtime performance of the graphbased segmentation method. 2007 Vol. 20 (4): 571-576 [Abstract] ( 285 ) [HTML 1KB] [ PDF 568KB] ( 572 ) 577 Adaptive DistanceWeighted Median Filtering Algorithm JIN LiangHai , LI DeHua , YAO XingZhong According to the relationship between spatial distances and human psychological perception, a new weighting function is presented based on the psychological distance. Combined with the standard median filter (SMF), a kind of adaptive median filter is proposed. The experimental results show that the proposed filter, compared with the SMF and several other representative improved median filters, achieves better filtering effect on noise attenuation, and edge or detail preservation. 2007 Vol. 20 (4): 577-581 [Abstract] ( 253 ) [HTML 1KB] [ PDF 1029KB] ( 444 )

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