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....

2015 Vol.28 Issue.5, Published 2015-05-30 Papers and Reports    Researches and Applications        Papers and Reports 385 Orthogonality Criterion Based Wavelet Filtering and Anisotropic Diffusion for Astronomical Texture Extraction SHAN Hao Astronomical images have complex morphological and hierarchical structures and irregular shaped textures, and they can be represented at different scales and directions. The purpose of this paper is to represent astronomical textures, and its mechanism is assumed from the perspective of orthogonality to extract the texture information. Based on orthogonality optimization criterion(OOC), wavelet filters, and anisotropic diffusion (AD), a method is presented to extract texture features for astronomical images. The theory assumes that the oscillation/texture component and the smooth piecewise/cartoon component are orthogonal to each other. The core technology is a parameter estimation method based on the orthogonality and AD. Firstly, the orthogonality measurement based wavelet thresholding scheme is adopted, and the multi-scale framework is used to extract and analyze the astronomical textures at different scales and directions. Then, the filtered smooth piecewise component is used to initialize AD. The parameter estimation is mainly applied to estimate the thresholds for multiscale wavelet filtering and AD iteration number. The images of galaxies and gravitational lensing are adopted for numerical experiments, and comparisons are implemented with 6 types of the currently used methods of image decomposition. The experimental results show that the proposed method can gain satisfying results in extracting astronomical textures, and it has advantages and advancement compared to other methods. 2015 Vol. 28 (5): 385-393 [Abstract] ( 471 ) [HTML 1KB] [ PDF 1664KB] ( 457 ) 394 Shuffled Frog Leaping Algorithm Based on Central Point Double Thresholds and Fuzzy Subgroups LIU Li-Qun, HUO Jiu-Yuan, WANG Lian-Guo, HAN Jun-Ying To overcome the demerits of basic shuffled frog leaping algorithm(SFLA), such as low optimization precision and falling into local optimum easily, a shuffled frog leaping algorithm based on central point double thresholds and fuzzy subgroups(CDTFSFLA) is proposed. The distance between frogs and central point in one subgroup is computed to measure compactness degree by selecting the central point randomly in each subgroup. The absolute threshold and the relative threshold of each subgroup are computed by the optimization method, and a strategy of fuzzy grouping with central point double thresholds and fuzzy subgroups is proposed to partition frogs into different fuzzy subgroups. In every local search, the update method of the worst individual in subgroups is improved according to the relation among central point membership, absolute threshold and relative threshold. The simulation results show that the proposed strategy and the update method are effective and feasible. CDTFSFLA can effectively improve convergence speed and precision in the optimization of unimodal and multimodal functions with fixed parameters, and it can maintain optimal performance under the condition of high dimensions, especially under the fitting condition that the number of neighborhood frogs is between 30 and 40 with dynamic parameters. The proposed algorithm improves the optimization performance of basic shuffled frog leaping algorithm effectively. 2015 Vol. 28 (5): 394-403 [Abstract] ( 431 ) [HTML 1KB] [ PDF 994KB] ( 469 ) 404 Research on Context-Awareness Mobile SNS Recommendation Algorithm ZHANG Zhi-Jun, LIU Hong Although patterns of human activity show a large degree of freedom, they exhibit structural patterns subjected by geographic and social constraints. Aiming at various problems of personalized recommendation in mobile networks, a social network recommendation algorithm is proposed with a variety of context-aware information and combined with a series of social network analysis methods.Based on geographical location and temporal information, potential social relations among users are mined deeply to find the most similar set of users for the target user, then recommendations are carried out incorporating with social relations of the mobile users to effectively solve the problem of recommendation precision. The above study can not only help LBSN designers and developers to better understand their users and grasp their want, but also help to refine the design of their system to provide users with more appropriate applications and services.The experimental results on the real-world dataset verify the feasibility and effectiveness of the proposed algorithm, and it has higher prediction accuracy compared with existing recommendation algorithms. 2015 Vol. 28 (5): 404-410 [Abstract] ( 512 ) [HTML 1KB] [ PDF 516KB] ( 925 ) 411 A Mixture Crossover Dynamic Constrained Multi-objective Evolutionary Algorithm Based on Self-Adaptive Start-Up Strategy GENG Huan-Tong, SUN Jia-Qing, JIA Ting-Ting Aiming at the slow convergence speed by using the cold start only, the poor adaptiveness of a single crossover operator and the poor diversity of the normal mutation, a mixture crossover dynamic constrained multi-objective evolutionary algorithm based on self-adaptive start-up strategy is proposed. Firstly, the hybrid cold-and-hot start-up mode is designed to identify the change degree of dynamic environment and the Cauchy mutation is used to enhance the diversity of evolutionary population. Then, to enhance the adaptiveness of crossover operation to the dynamic environment, three classical crossover operators, BLX_α,SBX and DE, are used simultaneously, and the respective competitiveness are adjusted adaptively according to their contributions. Finally, the cooperation of the elitist population and the evolutionary population balance the global searching ability and the local searching ability. The simulation results on 6 standard testing functions show that the proposed algorithm not only can dynamically identify the change degree in different environments and improve dynamic tracking effect by enhancing the diversity of initial population, but also can choose crossover operators automatically to accelerate the convergence. 2015 Vol. 28 (5): 411-421 [Abstract] ( 719 ) [HTML 1KB] [ PDF 641KB] ( 780 ) 422 Multi-granularity Search Algorithm Based on Probability Statistics ZHANG Qing-Hua, GUO Yong-Long, XUE Yu-Bin The basic idea of granular computing is to solve complicated problems in different granularity levels. To a great extent, the method indicates human intelligence in the solving process. Based on the human multi-granularity mechanism for solving complicated problems and the principle of probability statistics, an efficient multi-granularity search model based on statistical expectation is proposed from the viewpoint of granular computing. Then, the variety rule of the expectation in quotient spaces with different granularities is analyzed in detail. The experimental results demonstrate that with the granules divided into many sub-granules, the efficiency of the proposed method is gradually reduced and tends to be stable for searching a certain target. In addition, the complexity for solving the problem can be greatly reduced in different probability models. 2015 Vol. 28 (5): 422-428 [Abstract] ( 439 ) [HTML 1KB] [ PDF 426KB] ( 647 ) 429 Multi-manifold Connection Learning Algorithm Based on Frame Bundle ZHANG Qi-Ming, LI Fan-Zhang Traditional manifold learning methods need a large number of training samples. All samples are regarded as a manifold and then discriminative features are extracted for practical application such as classification. But in many situations, only one sample is existed during the training phasesince there are not enough training samples. Therefore, frame bundle connection learning method is presented and a multi-manifold structure is constructed. Besides,intermanifold and intramanifold features are extracted to get more discriminative information to solve the problem. When dealing with the multi-manifold structure, learning models of two subspaces based on frame bundle are used to project the data in high-dimensional space to horizontal space for maximizing margins of different manifolds. Simultaneously, the data structure is maintained with the same manifold in the vertical space. Finally, a simulation experiment is presented to prove the validity of the proposed algorithm. 2015 Vol. 28 (5): 429-436 [Abstract] ( 395 ) [HTML 1KB] [ PDF 647KB] ( 683 ) 437 A Term Specific Thresholding Method Based on Improved Score Distribution LU Li-Hua, ZHANG Lian-Hai To improve the precision of the spoken term detection system, a term specific thresholding method based on improved score distribution is presented. At the decision stage of the system, different thresholds are set for every query according to the posterior scores. The distribution of all posterior scores retrieved for a query term is modeled by exponential mixture model. The parameters are estimated by the expectation maximization (EM) algorithm in an unsupervised manner. The threshold value is calculated by Bayes minimum risk rule. Since EM algorithm is sensitive to initial values, K-means clustering is used in the initialization instead of randomization. Posterior scores are firstly divided into two classes, the prior distributions are calculated and the intial values of the model parameters are estimated by maximum likelihood method. The experimental results show that the performance of the proposed thresholding method is better than that of others. 2015 Vol. 28 (5): 437-442 [Abstract] ( 352 ) [HTML 1KB] [ PDF 542KB] ( 510 ) Researches and Applications 443 A Similarity Measurement Method of Facial Expression Based on Geometric Features HUANG Zhong, HU Min, WANG Xiao-Hua In facial animations such as performance-driven and expression cloning, it needs to find the most similar expression to enhance the reality and fidelity of animations. A feature-weighted expression similarity measurement method is proposed based on facial geometric features. Firstly, chain code is used to characterize shape features for local expression regions, meanwhile deformation features are built based on topological relations among regional feature points to reflect holistic expression information. Then, feature-weighted method is adopted to measure the similarities of fused geometric features, and the solving process of feature weights is transformed to minimizing process of the weighted objective function. Finally, the solved weights as well as feature weighting functions are performed to measure similarities between two expressions and seek the most similar image with a input expression image. The experimental results on BU-3DFE database and FEEDTUM database show that the proposed method has significantly higher accuracy in seeking similar expressions than existing measurement methods, and it keeps better robustness for the expressions with different categories and different intensities, especially in local details such as the shape of mouth, the contraction of cheek, and the open-close amplitude of mouth. 2015 Vol. 28 (5): 443-451 [Abstract] ( 455 ) [HTML 1KB] [ PDF 1210KB] ( 1017 ) 452 Cluster Validity Indexes for FCM Clustering Algorithm PIAO Shang-Zhe, Chaomurilige, YU Jian The clustering quality of fuzzy C-means (FCM) clustering algorithm is affected by several factors, such as initial setting of cluster centroid, the number of clusters and fuzzy index. In this paper, a comparative study on recently published five clustering validity measurement in different application fields is presented, e.g., different dimension of data, different cluster number and different fuzzy index. The experimental results show that the validity index based on ratio of within-class compactness and between-class separation is robust to data dimension and noise, and the validity index based on degree of membership can be applied to dataset with low dimension. The research results provide researchers with an option of selecting a suitable fuzzy clustering validity index for different application environments. 2015 Vol. 28 (5): 452-461 [Abstract] ( 527 ) [HTML 1KB] [ PDF 672KB] ( 1143 ) 462 Finding and Applying Typical User Group in Recommender Systems TAN Chang, LIU Qi, WU Le, MA Hai-Ping, LONG Bo Recommender system (RS) provides an effective way to solve the personalized information needs of users. However, with the expansion of the user scale, it is necessary to find some subsets of vast amounts of RS users, and the continuous and in-depth analysis for these user subsets can be used to improve the RS. Therefore, the typical user group (TUG) is defined as a representative subset of the entire users in RS to correctly reflect the preferences of all the users. Then, a weighted typical user group finding algorithm (WTFA) is designed to compare the contributions of the candidate typical users and choose the typical users with higher contribution, so that a TUG is built with high item coverage rate and rating accuracy. A modified TUG-based collaborative filtering(TUG-CF)algorithm is developed to discover the nearest neighbors in TUG. The experimental results on real world dataset show that TUG is better than most rating user group and maximizes diversified user group on item coverage rate and rating accuracy, and TUG-CF has better recommendation results than traditional collaborative filtering methods. 2015 Vol. 28 (5): 462-471 [Abstract] ( 538 ) [HTML 1KB] [ PDF 944KB] ( 926 ) 472 BMGSJoin: A MapReduce Based Graph Similarity Join Algorithm CHEN Yi-Fan, ZHAO Xiang, HE Pei-Jun, ZHANG Wei-Ming, TANG Jiu-Yang Graph similarity join has extensive use in the field of data mining, especially in data pre-processing, it could be applied to data cleaning, near duplicate detection, etc. Thus, it is of great importance to study graph similarity join. Graph similarit join based on edit distance constraints is studied, that is, all the edit distances in the return pair of graphs are no larger than a given threshold. Based on MapReduce programming model, an algorithm named MGSJoin is proposed with the ″filtering-verification″ framework, and it relies on graph signatures of path-based q-grams for filtering out non-promising candidates, i.e. count filtering.With the potential issue of too many key-value pairs, Bloom Filter is introduced to improve the algorithm and BMGSJoin is designed. The improvement of efficiency and scalability by the proposed algorithm is demonstrated by extensive experimental results, and it may meet the current challenges of big data mining and analysis. 2015 Vol. 28 (5): 472-480 [Abstract] ( 466 ) [HTML 1KB] [ PDF 551KB] ( 616 )

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