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....

2013 Vol.26 Issue.3, Published 2013-03-30 Orignal Article        Orignal Article 225 Representation of Topological Relations between a Concave Region and a Simple Region with a Hole LI Jian,OUYANG Ji-Hong,FU Qian,CHEN Gang Most of the spatial topological relation models are dealing with the same kind of spatial object,which are restricted in the practical applications at some degree. 4-intersection matrix model is extended to represent the topological relations between a concave region and a simple region with a hole. Thus,totally 161 topological relations are obtained,in which the illustrations of the first 10 relations are given. The algorithm is also presented to prove that the 161 topological relations are all feasible in the real world. The topological relations are proved to be exclusive and complete. Compared with other relative research work,the representation model is more expressive than other similar models. 2013 Vol. 26 (3): 225-230 [Abstract] ( 674 ) [HTML 0KB] [ PDF 392KB] ( 483 ) 231 Random Truth Theory of Proposition Logic and Its Application LIU Xiao-Ling,ZHANG Jia-Lu In this paper,the concept of random truth degree of proposition formulas based on a random variable sequence is introduced,which is a common generalization of various concepts of truth degree existing in references,and the set of random truth degree of all logic formulas is proved to have no isolated point in [0,1]. The random similarity degree and random pseudo-metric space between two logic formulas are defined by means of random truth degrees,and the random logic pseudo-metric space is proved to have no isolated point. The random truth degree of proposition logic is a generation of various truth degree of proposition logic. Using convergence theorem of integration in probability,a limit theorem of truth degrees is given,which shows the connection of various truth degrees. Various logic operations are continuous in random logic pseudo-metric space,and the fundamental theorem of probability logic is extended to the multi-valued proposition logic. Two diverse approximate reasoning ways are proposed in random logic pseudo-metric space. 2013 Vol. 26 (3): 231-241 [Abstract] ( 501 ) [HTML 0KB] [ PDF 433KB] ( 451 ) 242 Sparse Regularized Non-Negative Matrix Factorization through Online Learning XUE Mo-Gen,XU Guo-Ming,WANG Feng In order to overcome the inefficiency of non-negative matrix factorization,a fast approach based on online learning for sparse regularized non-negative matrix factorization is proposed. Firstly,the objective function is defined by imposing the regularization term to control the sparsity of the coefficient matrix,and the problem is transformed into the dictionary learning problem of sparse representation. Therefore,the object function can be solved by the online dictionary learning algorithm. Then,the block-coordinate descent algorithm is used to update the matrix in every iterative process,consequently,the convergence rate is improved. The experimental results show that the proposed method effectively preserves the structure information of images and simultaneously enhances the running efficiency evidently. 2013 Vol. 26 (3): 242-246 [Abstract] ( 670 ) [HTML 0KB] [ PDF 1054KB] ( 1276 ) 247 Sequence-Input Based Quantum Neural Networks Model and Its Algorithm LI Pan-Chi,SHI Guang-Yao To enhance the approximation capability of neural networks,a quantum neural networks model is proposed whose input of each dimension is in discrete sequence. This model includes three layers,in which the hidden layer consists of quantum neurons,and the output layer consists of common neurons. The quantum neuron consists of the quantum rotation gates and the multi-qubits controlled-not gates. By using the information feedback of target qubit from output to input in multi-qubits controlled-not gate,the overall memory of input sequences is realized. The output of quantum neuron is obtained from the entanglements of multi-qubits in controlled-not gates. The learning algorithm is designed in detail according to the basis principles of quantum computation. The characteristics of input sequence can be effectively obtained from the width and the depth. The simulation results show that,when the input nodes and the length of the sequence satisfy a certain relations,the proposed model is superior to the common artificial neural networks. 2013 Vol. 26 (3): 247-253 [Abstract] ( 536 ) [HTML 0KB] [ PDF 643KB] ( 588 ) 254 Voice Conversion Based on Speaker Independent Model CHEN Ling-Hui,LING Zhen-Hua,DAI Li-Rong A voice conversion method based on speaker independent (SI) model is proposed. Considering the phoneme information that commonly exists in every speakers speech,an SI space described only by the phoneme information is assumed to exist. Gaussian mixture model (GMM) is adopted to model the distribution of the SI space,and the mapping relations from speaker dependent (SD) space to SI space are described by linear transformations. The SI model is trained by using speaker adaptive training (SAT) algorithm on a multi-speaker database. In the conversion phase,the conversion functionfromsource space to target space is quickly and flexibly built by joining the transformations from source space to SI space and SI space to target space. The advantage of the proposed method is proved by the results of some listening tests compared with two representative conventional methods. 2013 Vol. 26 (3): 254-259 [Abstract] ( 660 ) [HTML 0KB] [ PDF 386KB] ( 779 ) 260 A Parallel Algorithm Generating Fuzzy Formal Concepts ZHANG Zhuo,CHAI Yu-Mei,WANG Li-Ming,FAN Ming Formal concept analysis (FCA) is extensively applied in various fields of computer. Currently,constructing fuzzy concepts directly is still one of most important issues of the FCA field. However,the construction process is always with exponential time complexity. In order to improve the efficiency of building fuzzy concepts,a parallel algorithm called Parallel Fuzzy Next Closure (ParaFuNeC) is presented. It is parallel developed from the serial construction algorithm of fuzzy concepts. The proposed method maps the combination search space of fuzzy set into the natural number interval,so that search space is simply expressed,divided and traversed through natural number. Moreover,the algorithm produces balanced and independent sub-search spaces according to the number of CPU in present computing environment. It also avoids the time costs of synchronization and communication among parallel tasks. By the time complexity analysis and experimental evaluation of the proposed algorithm,it is proved that the speedup ratio of the proposed algorithm increases proportionally to the number of CPU in the case of large-scale computing tasks. Besides,the criterion of serial fraction is used to analyze the scalability of the proposed algorithm in experiments. The results show that the algorithm ParaFuNeC also has better scalability in the case of large-scale computing tasks. 2013 Vol. 26 (3): 260-269 [Abstract] ( 465 ) [HTML 0KB] [ PDF 569KB] ( 535 ) 270 Document Clustering Based on Constrained Principal Component Analysis WANG Ming-Wen,YE Hao,ZUO Jia-Li Principal component analysis is an effective method to improve the performance of clustering in high dimension. On the other hand,principal component analysis is easy to lose the components which benefits for clustering. In order to preserve these beneficial components,an iteration algorithm of dimensionality reduction and clustering,named constrained principal component clustering,is proposed. Each iteration step can be represented as a constrained optimization problem which has a analytical solution. This iterative clustering algorithm is called document clustering based on constrained principal component analysis. The experimental results on Reuter21578 and WebKB show that the proposed algorithm outperforms to k-means,Non-Negative Matrix Decomposition and Spectral Clustering. 2013 Vol. 26 (3): 270-275 [Abstract] ( 531 ) [HTML 0KB] [ PDF 369KB] ( 679 ) 276 A Robust Affine Invariant Local Image Features Extraction Method ZHOU Tao,ZHANG Mao-Jun,XIONG Zhi-Hui,XU Wei A robust affine invariant local image features extraction method is proposed. Firstly,the image is transformed with M band wavelet. The feature points of the image are detected according to the energy of the M band wavelet transform coefficients. Then,every detected feature point is set to be a center. According to the local image information around the center,affine invariant feature is constructed with a moment structure. The experiments prove the invariant efficiency of the proposed method in image rotation,shift,scaling and viewpoint change etc. 2013 Vol. 26 (3): 276-281 [Abstract] ( 496 ) [HTML 0KB] [ PDF 1083KB] ( 701 ) 282 A Fast Q(λ) Algorithm Based on Second-Order TD Error FU Qi-Ming,LIU Quan,SUN Hong-Kun,GAO Long,LI Jing,WANG Hui Q(λ) algorithm is a classic model-free-based off policy reinforcement learning with multiple steps which combines the value iteration and stochastic approximation. Aiming at the low efficiency and slow convergence for traditional Q(λ) algorithm,the n-order TD Error is defined from the aspect of the TD Error which is used to the traditional Q(λ) algorithm,and a fast Q(λ) algorithm based on the second-order TD Error (SOE-FQ(λ)) is presented. The algorithm adjusts the Q value with the second-order TD Error and broadcasts the TD Error to the whole state-action space,which speeds up the convergence of the algorithm. In addition,the convergence rate is analyzed,and the number of iteration mainly depends on 11-γ、1ε under the condition of one-step update. Finally,the SOE-FQ(λ) algorithm is used to the random walk and mountain car,and the experimental results show that the algorithm has the faster convergence rate and better convergence performance. 2013 Vol. 26 (3): 282-292 [Abstract] ( 433 ) [HTML 0KB] [ PDF 626KB] ( 786 ) 293 Human Action Dynamic Modeling Recognition Based on Spatial Distribution Feature LIN Guang-Feng,ZHU Hong,FAN Cai-Xia,ZHANG Er-Hu The appearance feature and dynamic feature of human action have not an integrate description,which leads to distinguish human action inaccurately. In this paper,human action dynamic modeling recognition based on the spatial distribution feature (DMRSD) is proposed. Firstly,the spatial region of the feature is divided into a number of local regions by the relative polar coordinates,the statistic number of the nonzero information points is obtained in these local regions,and these numbers form a spatial distribution feature which describes the action appearance feature. Then,these spatial distribution feature sequences are modeled by autoregressive moving average model,then the feature of model parameter is obtained,which represents the dynamic time structure. Finally,the linear relation of the affinity matrix of these parameter features is hypothesized,the appearance feature structure and the dynamic motion feature structure are fused,and an integrate description is generated. Human action recognition is directly performed on the fusion structure of an integrate description by the nearest neighbor classification. Compared to the recognition results of current methods,DMRSD obtains better recognition rate on Weizmann and KTH databases. 2013 Vol. 26 (3): 293-299 [Abstract] ( 533 ) [HTML 0KB] [ PDF 580KB] ( 825 ) 300 Graph-Regularized Constrained Non-Negative Matrix Factorization Algorithm and Its Application to Image Representation SHU Zhen-Qiu,ZHAO Chun-Xia Non-negative matrix factorization (NMF) is an effective image representation method and has considerable attention in pattern recognition. The NMF is an unsupervised learning algorithm which can not take into account the label information and the intrinsic geometry structure simultaneously. In this paper,a matrix decomposition method called graph-regularized constrained non-negative matrix factorization (GRCNMF) is proposed,which preserves the label information with resorting to hard constraints,and hence the discriminating ability is improved. Meanwhile,a neighbors graph preserves the intrinsic geometrical structure of the data. The clustering experiments on the COIL20 and ORL image database demonstrate the effectiveness of the GRCNMF compared to other approaches. 2013 Vol. 26 (3): 300-306 [Abstract] ( 383 ) [HTML 0KB] [ PDF 436KB] ( 838 ) 307 An Improved Artificial Bee Colony Algorithm with Guided Normative Knowledge LIN Xiao-Jun,YE Dong-Yi An improved artificial bee colony (ABC) algorithm is proposed to solve numerical function optimization problems. Inspired by the double evolutionary space of cultural algorithm,the proposed algorithm takes advantage of the normative knowledge of reliability space to guide the search region and control the radius of the local search space self-adaptively. Thus,the convergence speed and the exploitation ability are enhanced. In order to maintain diversity,a dispersal strategy is designed to balance global exploration and local exploitation of population capacity.Moreover,different approaches are used to explore new positions in various evolutionary stages. The experimental results demonstrate that the proposed algorithm outperforms existing artificial bee colony algorithms on a number of standard test functions both in convergence speed and solution quality. 2013 Vol. 26 (3): 307-314 [Abstract] ( 531 ) [HTML 0KB] [ PDF 578KB] ( 481 ) 315 An Ensemble Classifier Based on Structural Support Vector Machine for Imbalanced Data YUAN Xing-Mei,YANG Ming,YANG Yang To improve the performance of Support Vector Machine(SVM) classifier for imbalanced data,an ensemble classifier model based on structural SVM is introduced by incorporating cost-sensitive strategy. In the proposed classifier model,the training data is partitioned into several group by Ward hierarchical clustering algorithm,the structure information hidden in data is obtained,and the weight of every sample is initialized by using the prior knowledge hidden in clusters. Furthermore,employing AdaBoost strategy,the weight of each sample is dynamically adjusted effectively,and the weights of minority class samples are relatively increased. Hence,the cost of the misclassified positive samples is also increased for improving the classification accuracy of positive samples(minority class samples). The experimental results show that the proposed model effectively improves the classification performance of the imbalanced data. 2013 Vol. 26 (3): 315-320 [Abstract] ( 610 ) [HTML 0KB] [ PDF 459KB] ( 1482 )

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