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....

2011 Vol.24 Issue.6, Published 2011-12-25 Orignal Article        Orignal Article 725 Algebraic Representation and Model Solving Algorithm for CP-nets LIU Jing-Lei, LIU Zhao-Wei, SUN Xue-Jiao, WU Shuan-Hu Conditional preference networks (CP-nets)is a popular language which represents qualitative conditional preference relation. Aiming at the problem that graphical representation is not enough to fulfill operations on CP-nets, an algebraic representation of CP-nets is offered with the well-known adjacent list approach. In the approach, vertical nodes are organized by topological order, and horizontal nodes are organized by their parents. Particularly, conditional preference table of nodes are represented by main disjunctive normal form of proposition logic. A direct model solving algorithm is devised later, and an indirect model solving algorithm is gotten based on relation operation on direct model. In short, the representation approach reveals that CP-nets can be not only represented by simple and intuitive graphical approach, but also represented by compacted algebraic approach. 2011 Vol. 24 (6): 725-732 [Abstract] ( 365 ) [HTML 1KB] [ PDF 485KB] ( 667 ) 733 A Cascade Algorithm of Quantum Attribute Evolution Reduction and Classification Learning Based on Dynamic Crossover Cooperation DING Wei-Ping, WANG Jian-Dong, GUAN Zhi-Jin, SHI Quan Attribute reduction and rule classification learning are important contents for research and application of rough set theory. Taking advantage of quantum computing to accelerate the algorithm speed and co-searching of shuffled frog leaping algorithm, a cascade algorithm of attribute reduction and classification learning based on the dynamic quantum frog-leaping crossover cooperation is proposed. Individuals in the frog swarm are represented by multi-state gene qubits, and the dynamic adjustment strategy of quantum rotation angle is applied to accelerate its convergence. By the crossover coevolution mechanism, classification rules are extracted and reduced, and decision rule chains are introduced in the classification criterion of rough entropy thresholding. The double cascade model of attribute reduction and classification learning is constructed. Experimental simulations indicate the proposed algorithm has good performance for global optimization. Compared with other algorithms, it is more efficient on attribute reduction and rule classification learning. 2011 Vol. 24 (6): 733-742 [Abstract] ( 564 ) [HTML 1KB] [ PDF 823KB] ( 639 ) 743 Robust Speaker Recognition against Synthetic Speech CHEN Lian-Wu, GUO Wu, DAI Li-Rong With the development of the hidden markov model (HMM) based speech synthesis technology, it is easy for impostors to produce synthetic speech with the specific speakers characteristics, which becomes an enormous threat to the existing speaker recognition system. In this paper, the difference between natural speech and synthetic speech is investigated on the real part of cepstrum. And a speaker recognition system is proposed which is robust against synthetic speech. Experimental results demonstrate that the false accept rate (FAR) for synthetic speech is zero in the proposed system, while that of the existing speaker recognition system is 99.2% with the equal error rate (EER) for natural speech unchanged. 2011 Vol. 24 (6): 743-762 [Abstract] ( 727 ) [HTML 1KB] [ PDF 542KB] ( 1103 ) 763 Cost-Sensitive SVM Based on Loss Functions with Weighted Margin TAO Qing, LIANG Wan-Lu, KONG Kang, Wang Qun-Shan Almost all the available algorithms deal with the imbalanced problems by directly weighting the loss functions. In this paper, a loss by weighting the margin in hinge function is proposed and its Bayesian consistency is proved. Furthermore, a learning algorithm, called Weighting Margin SVM (WMSVM), is obtained and SMO can be modified to solve WMSVM. Experimental results on certain benchmark datasets demonstrate the effectiveness of WMSVM. Both of the theoretical and experimental analysis indicate that the proposed weighted margin loss function method enriches the cost-sensitive learning. 2011 Vol. 24 (6): 763-768 [Abstract] ( 659 ) [HTML 1KB] [ PDF 386KB] ( 633 ) 769 Multiple Data Streams Clustering Based on Grey Relational Analysis GUO Kun, ZHANG Qi-Shan As a hot research orientation of data stream mining, multiple data stream clustering tracks the evolution of multiple streams and partitions them according to their similarities. In this paper, a multiple data stream clustering approach is proposed, which is based on the combination of grey relational analysis and affinity propagation clustering. A grey relational degree is developed so that the raw data can be compressed into an incrementally updatable grey relational synopsis. The similarity between two data streams is measured by the grey relational degree calculated from the synopsis. Finally, the affinity propagation algorithm is used to cluster the streams. The experiments on the real data sets prove the effectiveness of the new method. 2011 Vol. 24 (6): 769-775 [Abstract] ( 614 ) [HTML 1KB] [ PDF 427KB] ( 702 ) 776 Low-Rank Approximation and Decomposition for Kernel Matrix Based on Column Correlation LIU Song-Hua, ZHANG Jun-Ying , DING Cai-Ying An effective method of low-rank approximation and decomposition for kernel matrix is proposed . Firstly, aiming at the assumption that column of the kernel matrix is independent from its class label, the correlation of columns is studied and a strategy for column selection is designed. Secondly, the kernel matrix is decomposed into two stages: low-rank matrix decomposition and extension. Then an expectation of low-rank approximation error bound is given. The proposed algorithm extracts discriminative sub-matrix without independent assumption. In this way, it avoids the decomposition of the entire kernel matrix and effectively reduces the computational complexity. Finally, the experimental results show that the proposed method is effective and reasonable. 2011 Vol. 24 (6): 776-782 [Abstract] ( 623 ) [HTML 1KB] [ PDF 470KB] ( 822 ) 783 Orthogonal Fuzzy k-Plane Clustering Algorithm YING Wen-Hao, WANG Shi-Tong A clustering algorithm named Orthogonal Fuzzy k-Plane Clustering (OFKPC) is presented by introducing orthogonal restriction into Fuzzy k-Plane Clustering (FKPC). Similar to KPC and FKPC, OFKPC still uses k group hyperplanes as the prototypes of cluster centers. According to the idea of KPC and FKPC, the hyperplanes are built to distinguish samples in different classes. So the matrices constructed by the normal vectors of these hyperplanes can be used to reduce dimensionality. Experimental results on both artificial and UCI datasets show that OFKPC not only has better clustering results than FKPC but also has the ability of reducing dimensionality. 2011 Vol. 24 (6): 783-791 [Abstract] ( 719 ) [HTML 1KB] [ PDF 721KB] ( 696 ) 792 A Multivariate Discretization Method for Continuous Attributes HOU Ju-Chi, LIANG Ying, REN Chang-Zhi Currently, most discretization methods only consider a single variable, which can not get optimal discretization scheme. Taking the relationship among multi-attributes into account, a data discretization method is proposed. A multivariate discretization measurement criterion is presented by means of probabilistic model selection and minimum description length principle (MDLP). An efficient heuristic algorithm is proposed to get the best discretization scheme based on the proposed criterion. Nine UCI datasets are classified and predicted. Experimental results show that the proposed method significantly enhances the learning accuracy of Nave Bayes classifier. 2011 Vol. 24 (6): 792-797 [Abstract] ( 834 ) [HTML 1KB] [ PDF 538KB] ( 1629 ) 798 An Algorithm of Adaptive Total Variation Image Denoising NIU He-Ming, DU Qian, ZHANG Jian-Xun Aiming at the algorithm of traditional total variation image denoising which needs to know the noise variance and staircase of the object, an algorithm of adaptive total variation image denoising is proposed. The approximation item of traditional algorithm is modified by replacing the original noisy image with a blurred image, thus it is not necessary to know the noise variation of the image, and the impact of image noise in fidelity item is also reduced. Lagrange multiplier is no longer a global variable, as well as its numerical size is determined by image local information. Simultaneously, Lagrange multiplier is weighted by approximate edge information. So the evolution of the image can be expressed in a unified evolution equation. The experimental results and data analysis show that the proposed algorithm is superior to the traditional total variation image denoising algorithm. 2011 Vol. 24 (6): 798-803 [Abstract] ( 659 ) [HTML 1KB] [ PDF 949KB] ( 1180 ) 804 Fast Complete Discriminant Locality Preserving Projections for Face Recognition LU Gui-Fu, WANG Yong , JIN Zhong Fast complete discriminant locality preserving projections (FCDLPP) is proposed. There is only one step of economic QR factorization for FCDLPP algorithm to obtain the optimal discriminant vectors in the null space of locality preserving within-class scatter. Then, one step of eigen-decomposition is used to obtain the optimal discriminant vectors in the principal space of the locality preserving within-class scatter. Besides, FCDLPP fuses the regular discriminant features in the principal space and irregular discriminant features in the null space. Theoretical analyses and experimental results show that the proposed FCDLPP outperforms complete discriminant locality preserving projections (CDLPP) on computational speed and recognition rates. 2011 Vol. 24 (6): 804-809 [Abstract] ( 465 ) [HTML 1KB] [ PDF 343KB] ( 619 ) 810 Image Recognition with Two-Dimensional Neighbourhood Preserving Embedding ZHANG Da-Ming , FU Mao-Sheng, LUO Bin Neighbourhood preserving embedding (NPE) is a subspace learning algorithm, which aims at preserving the local neighbourhood structure on the data manifold, and it is a linear approximation to Locally Linear Embedding (LLE). When image data are concerned, the dimensionality of vectorized image data is usually high. NPE can not be implemented due to singularity of matrix. NPE is extended to 2 dimensional senses, 2DNPE, which directly extracts image feature from 2D image matrices rather than from 1D vectors as NPE does. The proposed algorithm is evaluated on Yale face database and Binary Alpha digits database. 2011 Vol. 24 (6): 810-815 [Abstract] ( 567 ) [HTML 1KB] [ PDF 448KB] ( 820 ) 816 Semantic Edit Distance between Two Directed Labeled and Rooted Trees KANG Qi, MA Jun In graph theory, the tree edit distance (TED) between two directed labeled and rooted trees is a popular research issue. As a combination optimization problem, calculating TED is widely used in the detection of the structural similarity of semi-structural documents. In this paper, a concept named tree semantic edit distance (TSED) with the corresponding formula is proposed. Then a distance measure based on both TED and TSED is presented. The proposed distance is applied in clustering the document object model (DOM) trees of extensible markup language (XML) documents. Experimental results show the proposed measure is better than those used TED only in terms of clustering precision and recall. The time complexity of the proposed algorithm is the same as those of algorithms for TED based on dynamic programming. 2011 Vol. 24 (6): 816-824 [Abstract] ( 505 ) [HTML 1KB] [ PDF 507KB] ( 901 ) 825 Sparse Graph Based Transductive Multi-Label Learning for Video Concept Detection ZHAO Ying-Hai, CAI Jun-Jie, WU Xiu-Qing, SUN Fu-Ming A sparse graph based transductive multi-label learning method is proposed for video concept detection. Firstly, the sparse signal representation theory is exploited to mine the point-wise similarity relationships and the concept-wise distribution correlation relationships. Then, the multi-label sparse graph structure is constructed based on discrete hidden Markov random field to conduct transductive semi-supervised video concept detection. The sparse representation for correlative information can remove the negative effect of redundant information, reduce the complexity of graph-based classification problem and improve the model efficiency and discriminability. The proposed method is evaluated on the TRECVID 2005 dataset, and extensive comparative experiments are conducted with respect to multiple supervised and semi-supervised classification methods. The experimental results demonstrate the effectiveness of the proposed method. 2011 Vol. 24 (6): 825-832 [Abstract] ( 585 ) [HTML 1KB] [ PDF 721KB] ( 861 ) 833 Defogging Algorithm of Lossy Compression Video Image LI Long-Li, LIU Qing, GUO Jian-Ming, ZHOU Sheng-Hui The traditional defogging algorithm used in the conventional industrial images acquired by the lossy compression of video images can t meet the real time constraint. And it also will form a number of irregular regions. The irregular regions cause lots of regions of color non-uniformity after defogging and seriously affect defogging result. Wavelet transform is presented to divide image into high and low frequency sub-band to find out the irregular regions. Then the transmissions of these regions are treated. And the image is recovered by using dark channel prior. Meanwhile, aiming at the problem that much more complicated computation in the matting algorithm of traditional dark channel prior is required, the method of the combination of linear interpolation smoothing and threshold recovery is proposed to instead of the matting algorithm. Thus, storage capacity and computation complexity are reduced effectively. The proposed algorithm meets the real-time request. Simulation results show the effectiveness of the proposed algorithm. 2011 Vol. 24 (6): 833-838 [Abstract] ( 470 ) [HTML 1KB] [ PDF 578KB] ( 648 ) 839 Iris Feature Extraction Algorithm Using Vertically Expanded Blanket Dimension and Lacunarity LIU Kai, ZHOU Wei-Dong, WANG Yu Iris feature extraction is important in iris recognition. An iris feature extraction algorithm is proposed by using combination of blanket dimension and lacunarity. Because Human iris texture is characterized by fractal geometry due to its rich self-similarity and abundant variation, vertically expanded blanket dimension is employed to represent iris texture variation and radial pattern at different resolution levels. Lacunarity is introduced to extract iris features that have different texture and fractal patterns but have same fractal dimension value. The combination of blanket dimension and lacunarity in iris feature extraction can embody the minute change of texture information comprehensively, and improve the capacity of iris classification. The experimental results on the CASIA-IrisV3-Interval iris database show that the combination of blanket dimension and lacunarity can extract iris textural features accurately and effectively, and high performance for iris recognition is achieved by using those features. 2011 Vol. 24 (6): 839-845 [Abstract] ( 601 ) [HTML 1KB] [ PDF 1244KB] ( 613 ) 846 Refined Junction-Tree-Based Algorithm for Reasoning in Bayesian Network HU Chun-Ling, HU Xue-Gang, YAO Hong-Liang Two classical junction-tree-based algorithms for reasoning in Bayesian network, Shafer-Shenoy architecture and Hugin architecture,are analyzed and compared. For the limitation of the Hugin algorithm in the reasoning analysis, a refined Hugin algorithm, R-Hugin, is proposed, which introduces the zero-factor flag and zero-factor processing mechanism in the message propagation process of the Hugin algorithm. R-Hugin algorithm has good reasoning and analyzing performance. Meanwhile, the correctness and efficiency of the R-Hugin algorithm are validated by theory and experiments. 2011 Vol. 24 (6): 846-855 [Abstract] ( 554 ) [HTML 1KB] [ PDF 432KB] ( 931 ) 856 Adaptive Method for Video-Based Face Recognition under Variable Illumination WANG Hua-Feng, WANG Yun-Hong , MA Kai-Di , ZHANG Zhao-Xiang A method is proposed, which combines adaptive histogram equalization (AHE), Gabor wavelet and LTP, to improve the video-based facial recognition under left, right, up, down and front illumination. Firstly, the AHE is used to reduce illumination variations on the existed face images from YaleB and CMU PIE face databases. Then, the images are convolved with Gabor filters to extract their corresponding Gabor feature maps and the LTP is used on each Gabor feature map to extract the local neighbor pattern. Finally, the input face image is described by using the histogram sequence extracted from all these region patterns. The results compared with the published results on YaleB and CMU PIE face databases of changing illumination verified the validity of the proposed method. 2011 Vol. 24 (6): 856-861 [Abstract] ( 513 ) [HTML 1KB] [ PDF 826KB] ( 659 ) 862 Abnormal Event Detection Based on the Multi-Instance Learning CUI Yong-Yan, GAO Yang In trajectory-based abnormal event detection, abnormal trajectory usually has abnormality in some parts of the whole trajectory and the rest are normal. However, most of the previous approaches are not able to detect this kind of abnormality easily. Aiming at the problem, an approach is proposed for abnormal event detection based on trajectory segmentation within the framework of multi-instance learning. In the proposed method, every trajectory is segmented into independent sub-trajectories based on their curvature firstly. Then, the sub-trajectories are modeled by hierarchical Dirichlet process-hidden Markov model (HDP-HMM). Finally, within the multi-instance learning framework, the whole trajectory is considered as bags while normal ones are negative bags, abnormal ones are positive bags, and sub-trajectories are instances in the bags. Experimental results show the proposed method achieves higher precision and recall than traditional ones. 2011 Vol. 24 (6): 862-868 [Abstract] ( 646 ) [HTML 1KB] [ PDF 501KB] ( 948 ) 869 Object Detection Based on Multi-Configuration Feature Bag LI Qiu-Jie, MAO Yao-Bin, WANG Zhi-Quan To solve the problem of feature misalignment appearing in object detection, a multi-configuration feature bag is put forward and used to describe variant misalignments for the same feature. By boosting algorithm, the most discriminative feature bags, which consist of single features and their corresponding misalignment cases, are selected in the phase of classifier training. Based on those best feature bags, weak classifiers are generated and combined into the final ensemble classifier. Moreover, multiple instance learning is introduced to efficiently evaluate the discriminative ability of feature bags and train feature bag classifiers. The experimental results on public face dataset demonstrate that the proposed algorithm is more robust than the traditional method when the problem of feature misalignment is considered. Furthermore, compared to the feature bag with fixed size, the proposed multi-configuration feature bag models the feature misalignment better, gets smaller detector size while improving the detection accuracy. 2011 Vol. 24 (6): 869-874 [Abstract] ( 496 ) [HTML 1KB] [ PDF 506KB] ( 631 ) 875 Blurred Edge Detection Algorithm Based on Local Scale Control HUANG Qian, WU Wei-Feng, DONG Xiao It is hard to obtain edges in the natural images with ambiguity features and the fusion between the foreground and background. Aiming at this problem, the fault-tolerant based edge location establishes a unique, locally calculable minimum reliable scale for each pixel of the image to locate and extract the blur edges. The calculating of the second derivation for each pixel is simplified, which uses the convolution mask along its gradient direction. By combining the local scale control with LoG algorithm, it avoids abundant second derivation calculations in the gradient direction. An approach of approximately determining zero crossing position as well as the flow chart to localize edges is presented and analyzed. The results of different recognition methods for three kinds of typical blurred images are compared, which show that the proposed algorithm localizes and extracts edges quickly and correctly, and it is more practical. 2011 Vol. 24 (6): 875-881 [Abstract] ( 431 ) [HTML 1KB] [ PDF 875KB] ( 694 )

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