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....

2012 Vol.25 Issue.6, Published 2012-12-25 Orignal Article        Orignal Article 885 A Delaunay Triangulation Based Diversity Metric for Solution Set of Multi-Objective Evolutionary Algorithms ZHENG Jin-Hua, WANG Kang, LI Mi-Qing, XIE Zhun-Zhi A Delaunay triangulation based metric (DTDM) is proposed for assessing the diversity metric in multi-objective evolutionary algorithms (MOEAs) by analyzing the characteristics and shortcomings of the current diversity metrics. The proposed metric is introduced by combining the neighborhood-based ideology and distance-based ideology. The metric independently searches the neighborhood by using the properties of the nearest and adjacent neighborhood of Delaunay triangulation net. The non-dominated relationship is eliminated according to a space mapping technique. The experimental results show that the proposed metric accurately evaluates the diversity of the solution set obtained by MOEAs. 2012 Vol. 25 (6): 885-893 [Abstract] ( 431 ) [HTML 1KB] [ PDF 903KB] ( 673 ) 894 Clustering Analysis of Gene Expression Data Based on Transitive Co-Expression WANG Wen-Jun Clustering analysis of gene expression data based on similar expression measures can not fully reveal the genetic function similarity between genes. Combined with gene transitive co-expression, a method for clustering analysis based on transitive co-expression is proposed to solve the problem. Firstly, the gene-related graph is built by using coefficient between gene expression profiles. Next, the transitive co-expression relationship between genes is obtained by the shortest path analysis. Then, clustering is performed by using k-means algorithm with transitive co-expression relationship as similarity measure. The experiments on Yeast gene expression data show that the transitive co-expression-based clustering method achieves better clustering performance compared with expression-based clustering method, and the clustering accuracy is significantly higher than that of the expression-based clustering method. The experimental results indicate that the proposed algorithm has better performance in revealing the nature of gene similarity compared with expression-based clustering method. 2012 Vol. 25 (6): 894-899 [Abstract] ( 667 ) [HTML 1KB] [ PDF 767KB] ( 623 ) 900 Lie Group Means Learning Algorithm GAO Cong, LI Fan-Zhang The method of mean computation on Lie group manifold is analyzed, and Lie group mean learning algorithm is proposed. The main idea of the algorithm is to find a one-parameter sub-group on the original Lie group which is decided by a Lie algebra element of intrinsic mean of all samples. The one-parameter sub-group is a geodesic on the original Lie group. Then, the projection of the sample to the geodesic is defined, and all samples to the geodesic are projected. In order to implement the discrimination in nonlinear Lie group space after projection, the ratio of between-class variance and within-class variance is maximized. The experimental results show that Lie group based algorithm is better than KNN, FLDA algorithms in classification performance. 2012 Vol. 25 (6): 900-908 [Abstract] ( 913 ) [HTML 1KB] [ PDF 621KB] ( 1289 ) 909 A Self-Adaptive Method for Optimizing the Parameters of Pulse Coupled Neural Network Based on QPSO Algorithm XU Xin-Zheng, DING Shi-Fei, SHI Zhong-Zhi, ZHAO Zuo-Peng, ZHU Hong Considering the parameters of pulse coupled neural network (PCNN) are mainly defined manually, a method based on quantum-behaved particle swarm optimization (QPSO) is presented to automatically determine the parameters in the neuron model of PCNN. Meanwhile, the time complexity of the proposed algorithm is analyzed. In proposed method, QPSO algorithm is used to automatically search the optimum values of parameters of the PCNN model or its simplified models in the solution space when the entropy of the image is defined as the fitness function of QPSO algorithm. The simulation results of image segmentation show that the proposed method obtains correct segmentation of Lena image. When mutual information (MI) is used as evaluation criteria, the performance of the proposed method is better than that of other methods, such as Otsu method, manual adjustment method of PCNN parameters, genetic algorithm and PSO. 2012 Vol. 25 (6): 909-915 [Abstract] ( 439 ) [HTML 1KB] [ PDF 1041KB] ( 804 ) 916 TLS-NAP Algorithm for Text-Independent Speaker Recognition HE Liang, YANG Yi, LIU Jia To improve the recognition accuracy rate of a text-independent speaker recognition system, a total least square-nuisance attribute projection (TLS-NAP) algorithm is proposed. The perturbation of the projection matrix is considered and its negative effect is minimized when hidden variables are estimated by the total least square algorithm. A better performance is obtained by the nuisance attribute projection space based on these variables. The effectiveness of the proposed method is demonstrated by the experimental results on NIST SRE 08 data corpus. 2012 Vol. 25 (6): 916-921 [Abstract] ( 682 ) [HTML 1KB] [ PDF 417KB] ( 686 ) 922 Multi-Level Attribute Reduction Methods Based on Concept Lattice YANG Kai, MA Yuan Attribute reduction is the kernel contents of rough set theory.Concept lattice is effective for knowledge representation and data analysis. Multi-level attribute reduction algorithm based on concept lattice is proposed by using concept lattice as reduction tool. The concepts including discriminable concepts, equivalent concepts and wane-n level are also introduced. The infuence of intent waned-value producing impact on the change of classification ability and the judge theorems of attribute reduction in concept lattice are mainly studied. The proposed algorithm discovers all the maximal reductions completely and an effective approach is presented to attribute reduction in concept lattice. Finally, a real example and experiment comparisons demonstrate both its feasibility and effectiveness. 2012 Vol. 25 (6): 922-927 [Abstract] ( 549 ) [HTML 1KB] [ PDF 392KB] ( 531 ) 928 Influence of Perturbations of Training Pattern Pairs on Stability of Polygonal Fuzzy Neural Network SUI Xiao-Lin, WANG Gui-Jun The concepts of the maximum perturbation error of polygonal fuzzy numbers and γ-perturbation of training pattern pairs are put forward, and the learning algorithm of connection weight is designed according to error-correction rules by introducing polygonal fuzzy numbers and their operations. Then the definition of the global stability of the polygonal fuzzy neural networks of the perturbation of training pattern pairs is introdued. Secondly, whenever the transfer function satisfies the Lipschitz condition and γ-perturbation occurs in the training pattern pairs,the stability of the connections of the three-layer polygonal fuzzy neural networks is proved by applying mathematical induction. Moreover, that the γ-perturbation of this network with respect to the training pattern pairs possesses the global stability is obtained. Finally, the influence of perturbations of training pattern pairs on the stability of polygonal fuzzy neural networks is explained by the simulative examples. 2012 Vol. 25 (6): 928-936 [Abstract] ( 427 ) [HTML 1KB] [ PDF 795KB] ( 589 ) 937 Gaussian Mixture Model Based on Variable Factor-Integration for Speaker Recognition LI Jie,LIU He-Ping To solve the problem that the recognition rate of traditional Gaussian mixture model decreases significantly in noisy conditions, a Gaussian mixture model based on α variable factor-integration is presented by adopting the α-integration mechanism of multiple stochastic models in the form of probability distributions. Through introducing the variable factor, the proportion of different compositions in the mixture model is adjusted again. By re-estimating the proposed model parameters, the experimental results show the performance of the proposed model is better than that of the traditional Gaussian mixture model on databases TIMIT/NTIMIT and different speaker numbers. Especially in noisy conditions with the optimal value of α,the recognition rate is increased by 8%. On NIST evaluation database the experimental results show that the recognition rate is increased as well compared with GMM-UBM system. 2012 Vol. 25 (6): 937-942 [Abstract] ( 329 ) [HTML 1KB] [ PDF 347KB] ( 549 ) 943 A Two-Stage Support Vector Machine Algorithm Based on Meta Learning and Stacking Generalization ZHU Min, LI Xue-Ling, LI Xiao-Lai, GE Yun-Jian A Two-Stage Support Vector Machine Algorithm (TSSVM) is proposed to improve the recognition accuracy of the surface electromyography (SEMG). The proposed algorithm is integrated with parallel method of meta-learning and the stacking idea of ensemble learning. In this algorithm, the basic classifiers are paralleled and distributed on the first stage and the outputs of the first-stage Support Vector Machine (SVM) are input into the second-stage SVM to integrate multi-source features and output the classification result. And then the proposed algorithm is used on test data set of the SEMG from human upper limb. The signals of SEMGs from individual muscles are respectively input into the first-stage SVMs. And the output of the first-stage SVMs is input into the second-stage SVM combiner to integrate and recognize the electromyographic signal features of individual muscle. Results show that TSSVM is superior to single SVM in classification accuracy. Moreover, TSSVM outperforms other state-of-art ensemble classifiers, such as random forest and rotation forest in classification accuracy, time cost and robustness. 2012 Vol. 25 (6): 943-949 [Abstract] ( 785 ) [HTML 1KB] [ PDF 528KB] ( 777 ) 950 Coordinate Descent Algorithms for Large-Scale SVDD TAO Qing, LUO Qiang, ZHU Ye-Lei, CHU De-Jun Support vector data description (SVDD) is an unsupervised learning method with significant application in image recognition and information security. Coordinate descent is an effective method for large-scale classification problems with simple operation and high convergence speed. In this paper, an efficient coordinate descent algorithm for solving large-scale SVDD is presented. The solution of concerned sub-problem at each iteration is derived in closed form and the computational cost is decreased through the accelerating strategy and cheap computation. Meanwhile, three methods for selecting sub-problem, analyzing and comparing their advantage and disadvantage are developed. The experiments on simulation and real large-scale database validate the performance of the proposed algorithm. Compared with LibSVDD, the proposed algorithm has great superiority which takes less than 1.4 seconds to solve a text database from ijcnn with 105 training examples. 2012 Vol. 25 (6): 950-957 [Abstract] ( 646 ) [HTML 1KB] [ PDF 539KB] ( 1185 ) 958 Review on Image Segmentation Based on Entropy CAO Jian-Nong The image segmentation based on entropy is analyzed and reviewed including one-dimensional maximum entropy, minimum cross entropy, maximum cross entropy and so on.The relations of Shannon entropy, Tsallis entropy and Renyi entropy are analyzed and commented, and the performance of two dimensional (high dimension) entropy and spatial entropy is also appraised. In conclusion, it points out the future research direction, such as the computational efficiency of the high-dimensional entropy model and one-dimensional entropy and other theories integrated. 2012 Vol. 25 (6): 958-971 [Abstract] ( 901 ) [HTML 1KB] [ PDF 1331KB] ( 3307 ) 972 Contour Curve Descriptor Based on Affine Invariance ZHANG Gui-Mei, XIONG Yi-Wen, MA Ke The traditional feature points only reflect some information of the contour curve. To describe the curve more accurately, a kind of feature points, nameed chord height point, is defined and extracted. Based on the feature points, a local curve descriptor is constructed to match the contour curves. Chord height point is sampled from each sub-curve and it can be employed to describe the curve more precisely than the common feature points such as corners, points of tangency and inflection points. It can solve the problem that the curve can not be described precisely, because the smooth curve has fewer feature points. The defined chord height points and the constructed recognition vector are local descriptors with invariance under affine transformation, so the matching method is robust under occlusion and affine transformation. The theory analysis and experimental results show that the proposed algorithm is effective. 2012 Vol. 25 (6): 972-978 [Abstract] ( 494 ) [HTML 1KB] [ PDF 445KB] ( 997 ) 979 A Method for Online Handwritten Uyghur Character Recognition Mayire IBRAYIM, ZHANG Heng, LIU Cheng-Lin, Askar HAMDULLA An approach for online handwritten Uyghur character recognition is proposed based on the analysis of the unique shapes and writing styles of Uyghur characters. The various techniques of normalization, feature extraction and classification are evaluated that have been successfully applied in handwritten Chinese character recognition. Specifically, eight normalization techniques and the normalization cooperated feature extraction (NCFE) method with different settings are used. Four classifiers are used for classification including the modified quadratic discriminant function (MQDF), the discriminative learning quadratic discriminant function (DLQDF), the learning vector quantization (LVQ) classifier, and the support vector classifier with RBF kernel (SVC-rbf). Furthermore, the geometric features which characterize the spatial context in handwritten documents are extracted for enhancing the recognition performance. In experiments on 38400 test samples of 128 classes, the proposed approach achieves an accuracy of 89.08%. 2012 Vol. 25 (6): 979-986 [Abstract] ( 662 ) [HTML 1KB] [ PDF 665KB] ( 858 ) 987 An Improved Fast FCM Image Segmentation Algorithm Based on Region Feature Analysis XU Shao-Ping, LIU Xiao-Ping, LI Chun-Quan, HU Ling-Yan, YANG Xiao-Hui A fast image segmentation algorithm based on region feature is proposed to estimate centroid number. In the preprocessing analysis stage, the feature vector based on the cooccurrence matrix statistics is used to describe the regional characteristics of sub-image, and the proposed algorithm combines with cluster validity function to estimate accurate centroid number and initialization of membership matrix. In the main clustering stage, the implicit feature of color and texture extracted by Gabor filter is used to accomplish clustering, which not only produces a more reasonable quality of region segmentation, but also has fine noise immunity. The experimental results show that the proposed algorithm effectively overcomes the deficiencies of pixel-level estimations, greatly accelerates the iterative speed of the FCM main clustering stage and achieves higher efficiency in the implementation. 2012 Vol. 25 (6): 987-995 [Abstract] ( 569 ) [HTML 1KB] [ PDF 1259KB] ( 1423 ) 996 A Text Clustering Method Based on Speech to Text and Improved Center Selection SHI Kan-Sheng, LIU Hai-Tao, SONG Wen-Tao The traditional k-means algorithm is sensitive to the initial point and easy to fall into local optimum. An improved speech to text and improved center selection (STICS) based text clustering method is proposed. Taking into account the speech to text, the optimal selection of centers and treatment of outliers concurrently, STICS has three aspects of improvement. The weighted vector space model (VSM) is used to represent text according to the speech to text. For the selection of the center, the sample average similarity is measured for each sample, and the sample with the largest sample average similarity is selected as the first center. In addition, STICS method eliminates the negative influences of isolated points or outliers. Both theoretical analysis and experimental results prove that the proposed algorithm has better clustering results. 2012 Vol. 25 (6): 996-1001 [Abstract] ( 407 ) [HTML 1KB] [ PDF 410KB] ( 928 ) 1002 Dynamic Self-Organizing Landmark Extraction Method Based on 2-Dimensional Growing Dynamic Self-Organizing Feature Map WANG Zuo-Wei, ZHANG Ru-Bo A dynamic self-organizing structural feature extraction method is presented based on distance sensor. The procedure consists of three parts: design of active exploration behavior, dimensionality reduction process of spatio-temporal information and self-organizing landmark extraction method. In this paper, active exploration behavior based on follow-wall is designed to obtain high correlative spatio-temporal sequence information. Activity neurons based on variety detection and activation intensity are used to reduce the dimensionality of spatio-temporal sequence. Finally, a method of 2-Dimensional growing dynamic self-organizing feature map (2-Dimensional GDSOM) is proposed to achieve self-organizing extraction and identification of environmental landmarks. The experimental results demonstrate the effectiveness of the method. 2012 Vol. 25 (6): 1002-1006 [Abstract] ( 382 ) [HTML 1KB] [ PDF 864KB] ( 604 ) 1007 Hand Motion Classification Based on Eye-Moving Assisted EEG MENG Ming, LUO Zhi-Zeng A classification method using wavelet packet transform (WPT) and support vector machine (SVM) is presented to classify the motor imagery electroencephalogram(EEG)of hand motion. Firstly, the relevant eye-moving assisted EEG at C3, C4, P3 and P4 during hand-motion imagery are recorded. Then, four feature rhythm waves are extracted using WPT, and the ratio of energy of each rhythm wave to the sum energy of all four rhythm waves is calculated respectively as the feature. Finally, the 16 dimension feature vector is input into SVM classifier to recognize the hand-motions. The average correct rate of four patterns of hand motions, namely wrist extension, wrist flexion, hand opening and hand grasping, is 82.3% in classification experiments and it shows that eye-moving assist improves the separability of motor imagery EEG. 2012 Vol. 25 (6): 1007-1012 [Abstract] ( 510 ) [HTML 1KB] [ PDF 841KB] ( 1172 ) 1013 Impact of Pattern Feature on Pattern Matching Problem with Wildcards and Length Constraints WANG Hai-Ping, HU Xue-Gang, XIE Fei, GUO Dan, WU Xin-Dong Pattern matching with wildcards and length constraints (PMWL) provides more convenience to users since its flexibility in definition which also leads to difficulties in solving problem. Currently, to our knowledge, no polynomial algorithms obtain the complete solution of this problem, and the analysis for completeness is far from sufficient. In this paper, the pattern feature is proved to be the key factor for the completeness of PMWL and a concept, denoted as rep, is provided which measures the repetitions in the pattern. The completeness of PMWL is proved under a certain condition when rep=0. And the reason of incompleteness under the condition of rep>0 is also explained clearly. In the experiments, approximation ratio is utilized as a measurement to demonstrate the impact of rep on the PMWL problem. 2012 Vol. 25 (6): 1013-1021 [Abstract] ( 320 ) [HTML 1KB] [ PDF 455KB] ( 690 ) 1022 A Multi-Objective Intelligent Optimization Based on the Principle of Light SHEN Ji-Hong, WANG Kan An intelligent multi-objective optimization is proposed with the foundation of light ray optimization. Relying on Fermat′s principle, the optimization successfully solves multi-objective problems by using the refraction and reflection principle. In this paper, the occurrence mechanism of the reflection is also proved strictly. The gray system is introduced into external archives, and a principle of the external archive maintenance is proposed. It shows that the gray system increases the uniformity of non-inferior solution. In the numerical experiment, the optimization is compared with MOPSO and NSGA-II by convergence index and diversity index. The results show that the proposed optimization has better effects and provides an idea for solving the high dimension multi-objective problem. 2012 Vol. 25 (6): 1022-1030 [Abstract] ( 505 ) [HTML 1KB] [ PDF 636KB] ( 865 )

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