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.6, Published 2008-12-01 Papers and Reports    Researches and Applications        Papers and Reports 713 A Computational Model of Social Inference in Multi-Agent Interactive Environment MAO Wen-Ji Social causal reasoning is a core aspect of social intelligence. Modeling social inference can enhance the cognitive and social functions of intelligent systems and promote the design of multi-agent systems. In this paper, a computational model of inferring social causality and social judgment is proposed based on an agent's causal knowledge and communicative interactions with other agents. An example is provided to demonstrate how the model works, and the effectiveness of the computational model is empirically validated by comparing with the related work. 2008 Vol. 21 (6): 713-720 [Abstract] ( 278 ) [HTML 1KB] [ PDF 778KB] ( 459 ) 721 SVDD Based Learning Algorithm with Progressive Transductive Support Vector Machines XUE Zhen-Xia, LIU San-Yang, LIU Wan-Li In semi-supervised learning, progressive transductive support vector machine (PTSVM) has some drawbacks, such as few sample labeled in each iteration, low training speed, many backtrack learning steps, and unstable learning performance. Aiming at these problems, a fast progressive transductive support vector machines learning algorithm is proposed. It selects new unlabeled samples based on support vector domain description (SVDD) by using the information of support vectors. Using region labeling rule instead of pairwise labeling rule of PTSVM, the algorithm inherits progressive labeling and dynamic adjusting of the PTSVM. And meanwhile it increases the computational speed and keeps even improves the accuracy. Experimental results on synthetic and real datasets show the validity of the proposed algorithm. 2008 Vol. 21 (6): 721-727 [Abstract] ( 342 ) [HTML 1KB] [ PDF 436KB] ( 582 ) 728 Axiomatic Definition and Measure Method of Knowledge Granularity in Incomplete Information System ZHAO Ming-Qing, YANG Qiang An axiomatic definition of knowledge granularity in incomplete information system is presented. A series of methods of measuring knowledge granularity are proposed, including concrete measurements lacking parameters and general measurements with parameters. The sizes of three knowledge granularities are compared. Furthermore, four combinatorial forms of different granularity formulas are described. The presented methods of measuring knowledge granularity have high theoretic and practical values to granularity computation building in information system. 2008 Vol. 21 (6): 728-729 [Abstract] ( 290 ) [HTML 1KB] [ PDF 279KB] ( 417 ) 730 Efficient Symbolic and Numerical Attribute Reduction with Neighborhood Rough Sets HU Qing-Hua, ZHAO Hui, YU Da-Ren Rough set theory is widely used in attribute reduction. Computational complexity is one of the factors to limit applicability in reduction techniques, especially in the neighborhood rough set based reduction. In this paper, some mathematical properties of neighborhood rough set model are analyzed. An efficient method is proposed for forward attribute selection strategy based on dependency by using the property that positive region monotonously increases with the amount of attributes. By this algorithm, the comparison times of the samples in computing positive region and neighborhood are reduced, and thus the computational efficiency is improved. The experimental results show that the proposed method is effective. 2008 Vol. 21 (6): 730-738 [Abstract] ( 494 ) [HTML 1KB] [ PDF 0KB] ( 121 ) 739 Cross-Media Retrieval Method Based on Feature Subspace Learning ZHANG Hong, WU Fei, ZHUANG Yue-Ting The latent correlation between low-level features of different modalities is studied, and an optimizing algorithm is proposed to improve cluster quality of both image and audio datasets in the feature subspace. To speed up the convergence of query process, three active learning strategies in relevance feedback are incorporated. Thus, the condition probability of unlabeled samples around labeled examples is calculated. Experimental results show that overall cross-media retrieval results are encouraging, and the mutual retrieval between image and audio data can be effectively realized by the proposed algorithm. 2008 Vol. 21 (6): 739-745 [Abstract] ( 289 ) [HTML 1KB] [ PDF 684KB] ( 692 ) 746 Fast Iterative Algorithm for Two-Dimensional Otsu Thresholding Method WU Cheng-Mao, TIAN Xiao-Ping, TAN Tie-Niu A fast iterative algorithm for two-dimensional Otsu thresholding method is proposed. Considering the disadvantages of the classical two-dimensional Otsu thresholding method and its recursive algorithm, it is supposed that the two-dimensional histogram which is composed of original segmented image and its local neighborhood average image is a two-variable continuous probability distribution function. The method for seeking extreme value of multivariate function is employed and the fast iterated algorithm of two-dimensional Otsu thresholding method is obtained. The experimental results show that the proposed fast iterative algorithm is feasible and has better segmentation performance. 2008 Vol. 21 (6): 746-757 [Abstract] ( 329 ) [HTML 1KB] [ PDF 1625KB] ( 1070 ) 758 A Subtractive Clustering Method Based on Genetic Algorithms GU Lei, WU Hui-Zhong The performance of the traditional subtractive method greatly depends on the choice of the parameters of mountain function. And only with proper parameters, the subtractive method can produce good results. Therefore, a subtractive clustering method based on genetic algorithms is proposed. Firstly, the traditional subtractive method is modified, and then the genetic algorithms are employed to optimize the relevant parameters of the improved subtractive method. Finally, experimental results on three synthetic datasets and two real datasets show that the proposed algorithm is valid and has encouraging clustering performance. 2008 Vol. 21 (6): 758-762 [Abstract] ( 401 ) [HTML 1KB] [ PDF 317KB] ( 418 ) 763 Included Angle Distance of Time Series and Similarity Search ZHANG Peng , LI Xue-Ren, ZHANG Jian-Ye, ZHANG Zong-Lin A method for time series approximation representation and similar measurement is proposed. Based on the adaptive piecewise linear representation, the time series are represented approximately with a sequence of the included angles between a pair of neighboring line segments. The basic concepts and properties of the included angle distance are proposed and proved. The included angle distance overcomes the problem when the point distance is used as the similar measurement, such as the poor robustness and ambiguous concepts. The proposed method is also invariant to translation and rotation. Experimental results on synthetic data and stock data show that the proposed method is effective. 2008 Vol. 21 (6): 763-767 [Abstract] ( 346 ) [HTML 1KB] [ PDF 385KB] ( 428 ) Researches and Applications 768 Robustness Analysis of Local Learning Algorithm Based on Nearest Neighbor BI Hua, WANG Jue Robustness in statistical inference means that the departure of real data from an assumed sample distribution has little influence on the results of the remarkable prediction performance of the algorithm. The research methods of statistical robustness are introduced into machine learning in this paper. The nearest neighbor estimation algorithm, a kind of local learning, can converge to Bayes optimal estimation in the case of large number of samples, and meanwhile the nearest neighbor estimation algorithm is a kind of robust algorithm under the convergent condition. Finally, experimental results on synthetic and real datasets demonstrate that the generalization performance of the nearest neighbor estimation algorithm can be guaranteed when the model is affected by some outliers. 2008 Vol. 21 (6): 768-774 [Abstract] ( 273 ) [HTML 1KB] [ PDF 398KB] ( 590 ) 775 Intelligent Recommendation Based on Multiple Data Sources and Co-Clustering WANG Rui-Qin, KONG Fan-Sheng With the development of internet and e-commerce, intelligent recommendation system emerges as the time requires. Collaborative filtering (CF) is regarded as the most effective recommender technique, but it has some limitations such as sparsity, scalability and cold start problems. In this paper, a hybrid recommendation method is proposed to overcome the limitations of CF. Firstly, a smooth filling technique is used on rating matrix with multiple data sources to solve the sparsity problem. Next, co-clustering technique from both user and item aspects is adopted to improve the scalability and precision of the system. The experimental results demonstrate the proposed approach has higher recommend accuracy than traditional collaborative filtering, meanwhile its online recommendation speed is fast. 2008 Vol. 21 (6): 775-781 [Abstract] ( 317 ) [HTML 1KB] [ PDF 416KB] ( 561 ) 782 An Approach to User-Orientated Highlights Extraction from a Sport Video BU Qing-Kai,HU Ai-Qun An approach to user-orientated sports highlights extraction is proposed based on video excitement time curve. Firstly, some low-level features are extracted from visual and audio information and mapped into users excitement. Then, the excitement time curve of the whole video is computed. Next, the position and the length of each highlight are determined, and all highlights are ranked on the proposed concept of importance-level of highlights. Finally, all highlights are exhibited in a friendly user-interface. Experimental results show the proposed approach can effectively detect most sport video from users' perspective. 2008 Vol. 21 (6): 782-786 [Abstract] ( 322 ) [HTML 1KB] [ PDF 806KB] ( 424 ) 787 Method for Rule Extraction from Neural Networks Based on Functional Point of View CHEN Guo A method for rule extraction from neural networks based on the functional point of view is studied. The key algorithms are introduced, including the sort and selection of features, the discretization of continuous attributes, the generation of training samples of neural network (NN), the training of NN, the generation of the instance samples from the trained NN, and the rule extraction. The UCI data and the population classifying data are used to verify the rule extraction method. The results show the correction and effectivity of the proposed method. 2008 Vol. 21 (6): 787-793 [Abstract] ( 301 ) [HTML 1KB] [ PDF 385KB] ( 447 ) 794 Speaker Verification Based on Gaussian Probability Distribution and SVM GUO Wu, DAI Li-Rong, WANG Ren-Hua In the text-independent speaker verification research, the probability distribution against the universal background model (PD-UBM) is calculated. And the score of each UBM Gaussian mixture is adopted as the input feature of the support vector machine (SVM) during the training and testing process. The proposed PD-UBM algorithm with linear kernel function can obtain the same or better performance as the generalized linear discriminant sequence (GLDS) kernel system. Furthermore, if the scores of the Gaussian mixture models (GMM-UBM), the GLDS and the PD-UBM are combined, the significant improvement of the system can be achieved. In 2006, on NIST 1conv4w-1conv4w speaker recognition evaluation (SRE) corpus, the fusion system obtained 25% relative improvement equal error rate (ERR) of over the GMM-UBM system. 2008 Vol. 21 (6): 794-798 [Abstract] ( 267 ) [HTML 1KB] [ PDF 348KB] ( 533 ) 799 Adaptive Gradient Vector Flow Algorithm for Boundary Extraction ZHANG Rong-Guo, LIU Xiao-Jun, WANG Rong, LIU Kun Adaptive gradient vector flow algorithm is proposed for boundary extraction as an improved method of gradient vector flow. Firstly, adjust factors are introduced to improve characters of diffusion field near the boundary. Then, an additional force is added in the normal direction of the active contour edge. According to the current location, the evolution directions of the curve can be determined in gradient vector flow field. Combined with gradient vector flow, the proposed algorithm can speed up the convergence with its large capture range maintained. It can solve deep concave problem as well as bottleneck problem. The experimental results demonstrate that the proposed method is efficient. 2008 Vol. 21 (6): 799-805 [Abstract] ( 239 ) [HTML 1KB] [ PDF 551KB] ( 528 ) 806 Discriminative Learning of TAN Classifier Based on KL Divergence FENG Qi, TIAN Feng-Zhan, HUANG Hou-Kuan Tree-augmented Nave bayes (TAN) classifier is a compromise between model complexity and classification rate. It is a hot research topic currently. To improve the classification accuracy of TAN classifier, a discriminative method based on Kullback-Leibler (KL) divergence is proposed. Explaining away residual (EAR) method is used to learn the structure of TAN, and then the TAN parameters are obtained by an objective function based on KL divergence. The experimental results on benchmark datasets show that the proposed method can get relatively high classification rates. 2008 Vol. 21 (6): 806-811 [Abstract] ( 284 ) [HTML 1KB] [ PDF 327KB] ( 829 ) 812 Semantic Annotation of Image Based on Information Bottleneck Method XIA Li-Min, Tan Li-Qiu, ZHONG Hong Firstly, a fully unsupervised segmentation algorithm with improved k-means is employed to divide images into regions. Then, a method of information bottleneck is proposed to cluster the segmented region and the relationship between image semantic concept and clustering regions is established. Image segmentation is used in the unannotated image so that the conditional probability of each semantic concept can be calculated under the condition of segmenting region. The image semantics is automatically annotated by keywords with maximal conditional probability. The system is implemented and tested on a 500-image database, and the experimental results show that the effectiveness of the proposed method outperforms others. 2008 Vol. 21 (6): 812-818 [Abstract] ( 263 ) [HTML 1KB] [ PDF 875KB] ( 398 ) 819 A Feature Extraction Method Based on ICA and Fuzzy LDA WANG Jian-Guo, YANG Wan-Kou, ZHENG Yu-Jie, YANG Jing-Yu Independent component analysis (ICA) and linear discriminant analysis (LDA) are two classical feature extraction methods. To extract optimal features, fuzzy technology is introduced into the fusion method of ICA and LDA. The proposed method can extract discriminative features from overlapping (outlier) samples effectively. Firstly, ICA is employed to extract initial features. Then, fuzzy k-nearest neighbor (FKNN) is implemented to achieve the distribution information of original samples. Finally, fuzzy LDA (FLDA) is performed on the basis of the above computation, and the effective feature vectors are extracted. Experimental results on the AR, ORL and NUST603 face databases demonstrate the effectiveness of the proposed method. 2008 Vol. 21 (6): 819-823 [Abstract] ( 273 ) [HTML 1KB] [ PDF 321KB] ( 599 ) 824 Rotate Invariant Vector Extraction of Pressed Protuberant Characters on Metal Label and PCA Based Recognition SONG Huai-Bo, LU Chang-Hou, LI Jian-Mei, LU Guo-Liang A sample vector formation method is presented to decrease the effect of the rotated characters on recognition rate. Firstly, the invariant features are extracted, such as centroid and principal axis. Then, the coordinate transformation is carried out and the image is converted to the polar coordinate system. Finally, the pixels are re-arranged according to the rules. The recognition is carried out directly on the gray-level images by adopting the improved PCA subspace method. Experimental results show that the proposed method can decrease the number of sieving samples with a higher recognition rate, compared with the typical method. 2008 Vol. 21 (6): 824-830 [Abstract] ( 254 ) [HTML 1KB] [ PDF 947KB] ( 487 ) 831 Predictive Control Based on Lexicographic MultiObjective Genetic Algorithm ZHANG Qian, ZHENG Tao To solve lexicographic optimization problem, the fitness function with dynamic coefficients is introduced into multi-objective genetic algorithm. Then a multi-objective controller is built based on this genetic algorithm. Simulation results of Shell standard control problem illustrate the proposed algorithm is efficient and its performance is also investigated under uncertainty of gains or big disturbance. 2008 Vol. 21 (6): 831-835 [Abstract] ( 379 ) [HTML 1KB] [ PDF 658KB] ( 615 ) 836 Facial Expression Recognition Using an Improved Embedded HMM ZHENG Fang-Ying, ZHAO Jie-Yu An embedded hidden markov model (e-HMM) based approach for facial expression recognition is proposed. It makes use of an optimized set of observation vectors obtained from the 2D-DCT coefficients of the facial region of interest. The e-HMM is trained with segmental K-means algorithm and used for the facial expression recognition. The experimental results show the remarkable improvement of the performance and robustness of the facial expression recognition system. 2008 Vol. 21 (6): 836-842 [Abstract] ( 334 ) [HTML 1KB] [ PDF 1057KB] ( 531 ) 843 An Approach to Mobile Robot Simultaneous Localization and Mapping Based on Improved Particle Filter PAN Wei, CAI Zi-Xing, CHEN Bai-Fan An approach to mobile robot simultaneous localization and mapping (SLAM) based on an improved particle filter is presented. The standard particle filter and particle swarm optimization algorithm are incorporated into the filtering framework of the proposed approach. The newest observations are introduced to adjust proposal distribution of particles, and thus the necessary number of particles for localization and mapping is greatly reduced and the particle degeneracy problem is effectively abated. In addition, considering that the typical resampling process always leads to the loss of diversity in particles,a probabilistic operator is introduced to keep the diversity of particle swarm. Experimental results show the feasibility and effectiveness of the proposed approach. 2008 Vol. 21 (6): 843-848 [Abstract] ( 315 ) [HTML 1KB] [ PDF 524KB] ( 598 ) 849 Video Foreground and Shadow Automatic Segmentation Based on Discriminative Model CHU Yi-Ping, CHEN Qin, HUANG Ye-Jue, ZHANG San-Yuan Moving cast shadows are factors affecting segmentation quality. Efficient shadow detection and removal is a difficult problem in video segmentation. A method based on discriminative model for video foreground and shadow segmentation is proposed. It has capability of shadow detection and removal under different scenes. The proposed algorithm models background, shadows and foreground at pixel levels. These models are constrained by using 2-dimensional conditional random fields. Inference algorithm of probabilistic graphical models is adopted to obtain globally optimal segmentation results. The experimental results demonstrate the validity of the proposed algorithm, and the results are compared with other algorithms by using outdoor and indoor video data. 2008 Vol. 21 (6): 849-855 [Abstract] ( 307 ) [HTML 1KB] [ PDF 605KB] ( 431 ) 856 A Speaker Recognition Algorithm Based on Speech Component Unit HUANG Chang-Cun, WANG Zeng-Fu The linear prediction coefficients (LPC) are selected as features to construct the corresponding feature space. Firstly, the samples for training of each individual in system are clustered in the feature space to form initial clusters by using k-mean clustering algorithm. Then, the initial clustering results are optimized based on Gaussian mixture model (GMM) iterative algorithm to obtain the speech component unit representation of the individual. On the basis of the obtained speech component units, a text-independent speaker verification method, called averaging method, and a text-independent speaker identification method are presented. Experimental results show that the proposed algorithm can produce a satisfying result even in short utterances. 2008 Vol. 21 (6): 856-866 [Abstract] ( 271 ) [HTML 1KB] [ PDF 315KB] ( 557 )

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