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....

2009 Vol.22 Issue.2, Published 2009-04-25 Papers and Reports    Researches and Applications        Papers and Reports 169 A Combination Rule of Evidence Theory Based on Similarity of Focal Elements YANG Shan-Lin, LUO He, HU Xiao-Jian Aiming at the problems of irrational assignment of conflict evidence, one-vote negation and poor robustness in evidence combination, a combination rule is proposed based on the notion of the similarity of focal elements and the distances between focal elements. The process of combination is divided into two parts, combination with and without conflicted mass. The mathematical proof of the proposed combination rule is presented, and the comparative experiments are conducted. The results show that the proposed rule can assign the conflicted mass rationally and avoid one-vote negation with good robustness. 2009 Vol. 22 (2): 169-175 [Abstract] ( 274 ) [HTML 1KB] [ PDF 378KB] ( 716 ) 176 A Generalized Form of Fisher Linear Discriminant Function CHENG Zheng-Dong, ZHANG Yu-Jin, FAN Xiang A generalized form of Fisher discriminant function is presented. It overcomes the limitations of two common discriminant functions. The presented form uniforms the discriminant functions in two subspaces of the dual subspace discriminant analysis (DSDA). A new orthogonal discriminant vector set is obtained by QR decomposition, and its discriminant property is approximate to that of the Foley-Sammon orthogonal discriminant vector set with smaller computational complexity. The experiments on ORL and JAFFE database show that theory analysis is consistent to the experimental results. 2009 Vol. 22 (2): 176-181 [Abstract] ( 254 ) [HTML 1KB] [ PDF 560KB] ( 461 ) 182 Path Search Strategies for Handwritten Character String Recognition YU Jin-Lun, ZHOU Xiang-Dong, LIU Cheng-Lin The optimal path search is usually used to obtain the results of character segmentation and character recognition in character string recognition . In this paper, two search fashions are applied to handwritten character string recognition: character-synchronous and time-synchronous. Their performance is compared by combining different path evaluation criteria and search strategies. Moreover, a modified path evaluation criterion is proposed. The dynamic programming (DP) algorithm can find the optimal path by the proposed criterion. Experimental results of online handwritten Japanese character string recognition show that time-synchronous search is more efficient than character-synchronous search for lexicon-free character string recognition. Under the proposed path evaluation criterion, the equivalent accuracies of the segmentation and recognition to the normalized path evaluation criterion are obtained with greatly reduced search time. 2009 Vol. 22 (2): 182-187 [Abstract] ( 354 ) [HTML 1KB] [ PDF 329KB] ( 518 ) 188 Clustering Based Pseudo-Parallel Genetic Algorithms LI Jun-Hua, LI Ming, YUAN Li-Hua The traditional genetic algorithm (GA) for multi-modal function optimization is studied and the characteristics of Niche GA and multi-population GA are analyzed. A clustering based pseudo-parallel genetic algorithm is proposed. Cluster analysis is carried out on all the individuals. Local search algorithm is used to search the optimum in all clusters. A new subpopulation is created by the unclassified individuals and the representations of all clusters. To get better global search capacity, niche technology is applied in the subpopulation. The convergence of the algorithm is proved theoretically. Moreover, a new method is designed for automatically calculating clustering threshold. Finally, the presented algorithm is compared with EGA、DCGA and MPGA. Results show that the new algorithm is well in searching global optimum and maintaining population diversity. 2009 Vol. 22 (2): 188-194 [Abstract] ( 278 ) [HTML 1KB] [ PDF 434KB] ( 799 ) 195 A Data Reduction Algorithm with Unification of Attribute and Attribute Value DENG Shao-Bo, GUAN Su-Jie , Li Min, LIU Qing A data reduction algorithm is proposed. It takes account of the equivalence class family of condition attributes in the decision table. Through analyzing the equivalence class family of decision value, both the attribute reduction and the attribute value reduction can be performed simultaneously in the process of reduction in a decision table. Compared with the traditional method based on the analytical method or discernibility matrix method, the proposed algorithms can omit the complicated comparison process in the reduction of the attribute value. The comparison times can be decreased and the reduction efficiency can be improved by the proposed algorithm. 2009 Vol. 22 (2): 195-201 [Abstract] ( 273 ) [HTML 1KB] [ PDF 344KB] ( 518 ) 202 Adaptive Clonal Selection Algorithm and Its Simulation WEI Yuan-Yuan, TANG Chao-Li , HUANG You-Rui Based on the basic principle of clonal selection algorithm, an adaptive clonal selection algorithm (ACSA) for function optimization is proposed. The clone number of antibody, the high frequency mutation ratios and the renewal number of each generation can regulate automatically in ACSA. Meanwhile, mutation antibodies have the ability of immune memory. The results indicate that the ACSA has stronger convergence and adaptability through the convergence analysis and simulation compared with standard clonal selection algorithm. 2009 Vol. 22 (2): 202-207 [Abstract] ( 272 ) [HTML 1KB] [ PDF 761KB] ( 581 ) 208 Image Thresholding Based on Two-Dimensional Arimoto Entropy ZHUO Wen, CAO Zhi-Guo, XIAO Yang A thresholding technique is proposed based on two-dimensional Arimoto entropy. Firstly, a two-dimensional histogram is determined by the gray value and the local average gray value of the pixels. Then, the two-dimensional Arimoto entropy is obtained from the two-dimensional histogram. The pair of gray values which makes the two-dimensional Arimoto entropy largest is the thresholding. By introducing in a two-dimensional joint power-probability distribution, a fast algorithm is proposed. The fast algorithm speeds up the implementation and makes the method suitable to real-time systems. Experiments indicate that the thresholding method based on two-dimensional Arimoto entropy gives a steady performance and it is better than the methods based on Renyi entropy and Shannon entropy. 2009 Vol. 22 (2): 208-213 [Abstract] ( 376 ) [HTML 1KB] [ PDF 698KB] ( 843 ) 214 Classifier Design Method Based on Piecewise Linearization WANG Qi, WANG Zeng-Fu The minimax risk criterion based decision is an important method for making decisions when priori probabilities are unknown. However, the performance of a minimax risk criterion based classifier is poor in most cases. To improve the performance of the designed classifier, a piecewise linearization based design method is presented. Firstly, the proposed method makes a rough estimation of the prior probability. Then, it decides the right interval where the estimated prior lies. Finally, the corresponding classifier is employed to make a decision. The theoretical deduction and experimental results show that the presented method is efficient and the performance of the corresponding classifier designed by the method approaches to Bayesian classifier. 2009 Vol. 22 (2): 214-222 [Abstract] ( 239 ) [HTML 1KB] [ PDF 624KB] ( 619 ) 223 A Latin Hypercube Sampling Based Multi-Objective Evolutionary Algorithm ZHENG Jin-Hua, LUO Biao Two evolutionary models, individual based evolutionary model (IND) and population based evolutionary model (POP) are proposed. Based on these two models, two kinds of multi-objective evolutionary algorithms (LHS) are designed based on Latin hypercube sampling, namely LHS-MOEAs. In LHS-MOEAs, the LHS local search is designed for exploiting promising areas and the evolutionary operator is designed for exploring new searching areas in feasible space. The combination of LHS local search and evolutionary operator in LHS-MOEA can prevent degeneration effectively. Experimental results demonstrate that the proposed LHS-MOEAs performs better and it is more preponderant than the classical NSGA-II in solving CPS_MOPs. 2009 Vol. 22 (2): 223-232 [Abstract] ( 305 ) [HTML 1KB] [ PDF 2332KB] ( 1295 ) 234 A Fast Scalable Attribute Reduction Algorithm WU Zi-Te, YE Dong-Yi The existing rough set based attribute reduction algorithms are mainly designed for the problem of the underlying data residing in the main memory. Therefore, the limitation of their application to attribute reduction computation of huge data results in a relatively poor scalability. Inspired by supervised learning in quest (SLIQ) algorithm, a specific data pre-processing strategy is introduced and a fast attribute reduction algorithm is proposed with time complexity O(｜U｜｜C｜). The experimental results show that the proposed algorithm is of good scalability. 2009 Vol. 22 (2): 234-239 [Abstract] ( 256 ) [HTML 1KB] [ PDF 316KB] ( 431 ) 240 Analysis of Robust Background Modeling Techniques for Different Information Levels WANG Zhi-Ling, ZHOU Lu-Ping, CHEN Zong-Hai Background modeling is a critical element of detecting and tracking moving objects, and the robustness problem of the background model attracts more and more attention. Firstly, the requirements for robustness are analyzed by taking two aspects into account: different applications/environments and different modalities of sampling sets. Then, a review of current background modeling algorithms and systems is presented according to their information levels and performance. After analyzing these algorithms and comparing typical background modeling systems, several promising directions of background modeling are pointed out for future research. 2009 Vol. 22 (2): 240-245 [Abstract] ( 364 ) [HTML 1KB] [ PDF 372KB] ( 551 ) Researches and Applications 246 Data Stream Clustering Based on Immune Principle WANG Shu-Yun, ZHANG Cheng-Hong, HAO Xiu-Lan, HU Yun-Fa The learning based on immune principle adapts well to the dynamic environment, and thus it can be applied to the data stream processing which is dynamic and requires high-speed processing. Therefore, an algorithm of clustering data streams based on immune principle is proposed, namely AIN-STREAM. The proposed algorithm can track the evolving clusters on noisy data sets. AIN-STREAM is capable of adjusting the recognition zone of B-cells automatically according to the requirement of users by creating and maintaining the B-Cell feature vectors. Thus, the stability of the clustering result is ensured. Theoretical analysis and comprehensive experimental results demonstrate that AIN-STREAM is superior over other immune principle based clustering algorithms under the circumstance of similar clustering results. Moreover, the results show that AIN-STREAM has a high clustering quality. 2009 Vol. 22 (2): 246-255 [Abstract] ( 248 ) [HTML 1KB] [ PDF 989KB] ( 410 ) 256 Feature Selection Algorithm Based on Association Rules WU Jian-hua, SONG Qin-Bao, SHEN Jun-Yi, XIE Jian-Wen A feature selection algorithm based on association rules is presented, and the impact of support and confidence on the presented method are studied. The experimental results show that the feature subset size and classification accuracy of the presented method are better than those of other methods. Furthermore, the results indicate high support and confidence levels do not guarantee high classification accuracy and small feature subset, and the sufficient number of rules is the precondition for high efficiency of feature selection based on association rules. 2009 Vol. 22 (2): 256-262 [Abstract] ( 399 ) [HTML 1KB] [ PDF 469KB] ( 953 ) 263 An Improved SLAM Algorithm with Sparse Extended Information Filters GUO Jian-Hui, ZHAO Chun-Xia How to achieve a sparse information matrix exactly is a key issue in sparse extended information filter (SEIF) simultaneous localization and map building (SLAM). A sparsification rule is put forward based on the deep analysis of correlation. The rule can utilize observation information of sparsification time, observe the correlation globally and reserve the features with the strongest correlation. The precision and consistency of the algorithm are improved without an increase of computational burden. Results of Monte-Carlo simulation experiments indicate the validity of the improved algorithm. 2009 Vol. 22 (2): 263-269 [Abstract] ( 393 ) [HTML 1KB] [ PDF 609KB] ( 483 ) 270 A PCA Method Based on Speaker Session Variability LONG Yan-Hua, GUO Wu, DAI Li-Rong In the text-independent speaker verification systems, the mismatch and variability of the channel and environment between training and testing is a session variability problem. It can greatly degrade the speaker recognition performance. To deal with the problem more efficiently, a modified PCA method is proposed called session variation principal component analysis (SVPCA) which can integrate with within class covariance normalization (WCCN). In the NIST 2006 verification task, the proposed method is compared with our previous baseline general linear discriminative sequence-support vector machine (GLDS-SVM) system. The experimental results show a relative reduction of up to 24.2% in error equal ratio (EER). Moreover, the proposed method has advantages in computational and memory costs, compared with the state-of-art systems. 2009 Vol. 22 (2): 270-274 [Abstract] ( 257 ) [HTML 1KB] [ PDF 395KB] ( 605 ) 275 Graph Cuts and Shape Statistics Based Cardiac MR Image Segmentation Using Active Contours Model LIU Fu-Chang, ZHU Jin, YANG Ya-Fang, HENG Pheng Ann, XIA De-Shen To analyze heart function effectively, it is necessary to segment the left and right ventricles precisely. In cardiac MR images, the weak edges, broken boundaries, region inhomogeneity and noises cause difficulties in segmenting the contours of left and right ventricle precisely. In this paper, the training samples are aligned and analyzed, and the allowable shape space of the left and right ventricles is constructed. An active contours model based on graph cuts and shape statistics is proposed for segmentation of cardiac MR images. It uses graph cuts based active contours (GCBAC) to convert the image segmentation into the globally optimal partition after transforming the image into a graph. Next, GCBAC uses graph cuts to iteratively deform the contour. Consequently, it has a large capture range. Then, the shape statistics is introduced into GCBAC. The introduction of shape statistics prevents the deformation curve form leaking out of actual boundaries. Experimental results demonstrate the proposed method achieves a higher segmentation precision and a better stability than other approaches and it provides a feasible way for clinical applications. 2009 Vol. 22 (2): 275-281 [Abstract] ( 493 ) [HTML 1KB] [ PDF 1971KB] ( 734 ) 282 Clustering Analysis of Experts’ Opinion and Its Visualization in Hall for Workshop of Meta-Synthetic Engineering XIONG Cai-Quan, LI De-Hua, ZHANG Yu Consensus building in the hall for workshop of meta-synthetic engineering relies on the experts’ mutual-question and mutual-elicitation in discussion. Therefore, it is necessary to visualize the individual opinion and the coherent state of the group in real time, thus it facilitates the experts to converge their thinking quickly. In this paper, a heuristic clustering algorithm is proposed. It partitions the group into a set of different subgroups, and a series of indexes for the consistency analysis of the group and subgroups are defined. Then a method, called parallel coordinate, is used to display the coherent state of the group so as to facilitate the discussion. Finally, an example is given to illustrate the validity of the proposed method. 2009 Vol. 22 (2): 282-287 [Abstract] ( 374 ) [HTML 1KB] [ PDF 391KB] ( 587 ) 288 Small Sample Optimal Discriminant Transform Based on PSO under Fisher Criterion RUI Ting, ZHOU You, QI Tian, FANG Hu-Sheng, RONG Xiao-Li The within-class scatter matrix Fisher criterion is singular under small samples. Therefore, it can not be solved directly. A method based on PSO is proposed to get optimal discriminant transform under small samples without calculating inverse of the within-class scatter matrix. The methods and steps are discussed to get optimal discriminant projection vector by velocity-position search model of particle swarm optimization. The eigenvectors method and the proposed method are compared, when within-class scatter matrix is non-singular. Experimental results on both small and large samples demonstrate the accuracy of the proposed method. 2009 Vol. 22 (2): 288-292 [Abstract] ( 246 ) [HTML 1KB] [ PDF 443KB] ( 494 ) 293 Feature Selection Method Based on Fractal Dimension and Ant Colony Optimization Algorithm NI Li-Ping, NI Zhi-Wei, WU Hao, YE Hong-Yun Feature selection plays an important role in machine learning and data mining as a primary preprocessing step. A feature selection algorithm is presented based on fractal dimension and ant colony optimization algorithm. In this algorithm, fractal dimension is used as an evaluation mechanism and ant colony optimization algorithm is employed to accelerate the selection process. To evaluate the efficiency of the proposed algorithm, the SVM algorithm and K-fold cross validation are utilized to evaluate the classification accuracy on four datasets. The experimental results show the proposed algorithm can identify the better feature space with a great decrease of dataset dimension in a short time. 2009 Vol. 22 (2): 293-298 [Abstract] ( 266 ) [HTML 1KB] [ PDF 351KB] ( 434 ) 299 A K-means Algorithm Based on Optimized Initial Center Points WANG Zhong, LIU Gui-Quan, CHEN En-Hong Aiming at the problems of K-means algorithm, a method is proposed to optimize the initial center points through computing the density of objects. Thus, the initial center of the samples can be built in a heuristic way. Then, a new evaluation function is proposed, namely equalization function, and consequently the cluster number is generated automatically. Compared with the traditional algorithms, the proposed algorithm can get initial centers with higher quality and steadier cluster results. Experimental results show the effectiveness and feasibility of the proposed algorithm. 2009 Vol. 22 (2): 299-304 [Abstract] ( 396 ) [HTML 1KB] [ PDF 369KB] ( 1460 ) 305 Generalized Fuzzy Entropy Thresholding Based on Quantum Genetic Parameter Optimization YU Hai-Yan, FAN Jiu-Lun Taking advantage of quantum genetic algorithm, a nested optimization method is proposed aiming at the generalized fuzzy entropy parameters. Quantum genetic algorithm is used to automatically determine the optimal parameter m in (0,1) based on an image segmentation quality evaluation criterion and the parameters of the fuzzy membership function corresponding to each m based on the maximum fuzzy entropy criterion. Thus, the automatic selection of threshold is realized in generalized fuzzy entropy-based image segmentation method. Experimental results show that the proposed method can obtain good segmentation results for images with poor illumination. 2009 Vol. 22 (2): 305-311 [Abstract] ( 280 ) [HTML 1KB] [ PDF 641KB] ( 366 ) 312 An Adaptive Image Watermarking Scheme Based on Support Vector Machine and Genetic Algorithm MENG Fan-Man, PENG Hong, PEI Zheng, WANG Jun An adaptive blind image watermarking scheme in DCT domain is proposed based on support vector machine (SVM) and genetic algorithm (GA). The original image is divided into small image blocks, and then SVM is used to classify image blocks into several classes based on their local characteristics. The embedding strength of each block is adaptively determined according to the types of the image block, and GA is used to seek optimal embedding positions. Experimental results demonstrate that the proposed scheme has good invisibility and strong robustness against several attacks. 2009 Vol. 22 (2): 312-317 [Abstract] ( 292 ) [HTML 1KB] [ PDF 1062KB] ( 679 ) 318 Mining Algorithm for Minimal Rule Based on Concept Lattice QIU Wei-Gen The concept lattice is an important mathematic tool for knowledge treatment and data analysis, its efficient construction algorithm is significant in rules acquisition of decision table. In this paper, the formal context and concept lattice model of decision table are constructed, the relationships among the extended indistinguishable matrix, concept lattice and minimal rule are analyzed. All the concept node intension comes from property element of extended indistinguishable matrix and all the condition properties of optimal decision rule are from the intension reduction of a concept lattice node. Two algorithms are developed for constructing the corresponding concept lattice incrementally and acquisition of minimal decision rule based on the concept lattice, and their simplicity and efficiency are proved by an enterprise example. 2009 Vol. 22 (2): 318-324 [Abstract] ( 287 ) [HTML 1KB] [ PDF 455KB] ( 475 ) 325 Multiple Subclassifier Integration Method of Decision Forest Based on General Information Theory WANG Li-Min, XU Pei-Juan, LI Xiong-Fei To improve the scalability and adaptability of traditional decision tree learning algorithm, a novel multiple subclassifier integration method of decision forest is proposed based on general information theory. It adopts down-top learning strategy and combines discretization with logical representation of decision tree naturally. The learning procedure does not require any human intervention. The number and structures of subtrees can be set automatically. Experimental results and instance analysis on UCI machine learning data sets prove the feasibility and effectiveness of the proposed method. 2009 Vol. 22 (2): 325-329 [Abstract] ( 292 ) [HTML 1KB] [ PDF 359KB] ( 586 ) 330 An Improved KNN Algorithm for Boolean Sequence WANG Zhen-Hua, HOU Zhong-Sheng, GAO Ying As a special classification problem, classification of Boolean sequences is seldom studied. Definitions of the ordering and piecewise mapping are given. And then a dimension-reduction method called ordering and piecewise mapping (OPM ) is put forward. Thus an improved KNN algorithm (OPM-KNN) is presented by integrating OPM with KNN. Analytical and experimental results show the speed of OPM method is improved compared with that of traditional PCA algorithm in dimension reduction. As for classification, the accurate rate of OPM-KNN is almost equivalent to the traditional KNN algorithm or appreciably higher than it and the speed is also faster. 2009 Vol. 22 (2): 330-335 [Abstract] ( 288 ) [HTML 1KB] [ PDF 438KB] ( 702 )

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