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....

2009 Vol.22 Issue.4, Published 2009-08-25 Papers and Reports    Researches and Applications    Surveys and Reviews        Papers and Reports 505 Harris Correlation and Feature Matching WANG Xu-Guang, WANG Zhi-Heng, WU Fu-Chao An image feature based on image gradient is proposed, called Harris correlation. It is used to construct feature descriptors including point descriptor, line descriptor as well as curve descriptor. These feature descriptors are invariant to image rotation and linear intensity changes. Constructing descriptors for both line and curve explores a new way for line and curve matching. Experimental results show that the point descriptor performs well on image transformations, and the line and curve descriptors are effective in real image matching as well. 2009 Vol. 22 (4): 505-513 [Abstract] ( 274 ) [HTML 1KB] [ PDF 2353KB] ( 579 ) 514 Serial Factor Analysis in Speaker Recognition GUO Wu, DAI Li-Rong, WANG Ren-Hua A serial loading matrix training method is proposed in the factor analysis based speaker recognition. In the loading matrix training process, the eigenvoice matrix, the diagonal matrix(residual) and the channel matrix are calculated serially. In the speaker enrollment process, the above three matrixes are assembled, and then the factors are calculated through the joint factor analysis. Thus, the saturation problem in factor analysis is solved. On the NIST SRE 2006 core test corpus, the equal error rate of the proposed system is 3.65%. 2009 Vol. 22 (4): 514-518 [Abstract] ( 309 ) [HTML 1KB] [ PDF 308KB] ( 450 ) 519 Repair Strategies for Multiobjective 0/1 Knapsack Problem in MOEA HUANG Lin-Feng, LUO Wen-Jian, WANG Xu-Fa A repair strategy is often adopted to guarantee feasibility of the multiobjective evolutionary algorithms for multiobjective 0/1 knapsack problem (MOKP). In this paper, impacts of each item on all knapsacks are much considered and two novel repair strategies are proposed based on the knapsack capacities and constraint violations, respectively. The two novel strategies are applied to SPEA2 to solve MOKP. The experimental results on 9 standard test cases of MOKP demonstrate that SPEA2 with the proposed repair strategies has better convergence to the Pareto-optimal front. 2009 Vol. 22 (4): 519-526 [Abstract] ( 377 ) [HTML 1KB] [ PDF 394KB] ( 1058 ) 527 Structure and Performance Analysis of Open Domain QA System DU Yong-Ping, HUANG Xuan-Jing Open domain question answering (QA) has drawn much attention from the natural language processing communities. A pattern based question answering system is introduced, and the deep performance analysis and the evaluation of the QA system are presented. The impact of parameter tuning and training set size on system performance is discussed as well. Meanwhile, the t-test results denote the significance of the performance improvement by different factors. Natural language processing tools are used in the QA system, such as the syntax parser and the named entity recognition tool. Analysis results indicate that these tools play an important role in the QA system. 2009 Vol. 22 (4): 527-531 [Abstract] ( 300 ) [HTML 1KB] [ PDF 442KB] ( 546 ) 532 An Enhanced Tableau Algorithm Based on Clause Reorganization GU Hua-Mao, GAO Ji, WANG Xun, WU Hai-Yan Taking advantage of the expandability of acyclic concept definitions, an enhanced Tableau algorithm based on clause reorganization is presented. It substitutes the succinctest conjunctive clause of concepts for sub-concepts set in labeling nodes of complete tree/graph. To build a new complete tree/graph, a set of reasoning rules is designed, which avoids tremendous description redundancy caused by ∩-rules and ∪-rules of the traditional Tableau. Thus, the spatial performance of judging the satisfiabilities of acyclic concept definitions is improved greatly. In this papers only reasoning rules and proofs designated for SI language are provided. However, the idea of enhancement mode is suitable for various DL languages as well, which makes it much valuable. 2009 Vol. 22 (4): 532-540 [Abstract] ( 342 ) [HTML 1KB] [ PDF 507KB] ( 575 ) 541 Regularized Possibilistic Linear Models Based Adaptive Filter for Image Restoration GE Hong-Wei, WANG Shi-Tong Median filter is widely used to remove impulsive noise but it distorts the fine structure of signals. To improve the median filter, an adaptive filter controlled by regularized possibilistic linear models is proposed. The proposed filter achieves good results through a summation of the input signal and the output of median filter. The weights are set based on regularized possibilistic linear models according to the states of the input signal sequence. The experimental results of image denoising show this filter effectively suppresses impulsive noises and simultaneously preserves image details. Moreover, the proposed filter has excellent robustness to various percentages of impulse noise in the testing examples. 2009 Vol. 22 (4): 541-547 [Abstract] ( 334 ) [HTML 1KB] [ PDF 1506KB] ( 505 ) 548 Support Vector Data Description with Manifold Embedding CHEN Bin, LI Bin, PAN Zhi-Song, CHEN Song-Can Geodesic distance is a good metric to approximate the underlying global geometry. However, support vector data description (SVDD) with geodesic distance cannot be directly optimized. A framework for manifold-based classifier is designed. The Euclid distance in the feature space induced by isometric feature mapping (ISOMAP) dimension reduction is approximated by the geodesic distance in the input space, and implicitly conducts the former learning algorithm (with Euclid distance) after the ISOMAP process. Next, the proposed method is extended to SVDD and a SVDD derivate with ISOMAP manifold embedding (mSVDD) is developed. Experimental results on USPS handwritten digital dataset show that compared with traditional Euclid distance based SVDD, mSVDD significantly increases the performance for one-class classification. 2009 Vol. 22 (4): 548-553 [Abstract] ( 276 ) [HTML 1KB] [ PDF 506KB] ( 651 ) 554 An Algorithm for Fingerprint Ridge Distance Estimation REN Chun-Xiao, YIN Yi-Long, MA Jun, ZHAN Xiao-Si The average ridge distance of a fingerprint image is an important characteristic of fingerprint texture attribute. As an important parameter, the precision of estimation seriously affects the results of segmentation, enhancement and classification of fingerprint recognition. An algorithm is proposed to compute the average ridge distance of fingerprint accurately by using discrete Fourier transform, discrete information entropy theory, and weighted Euclidean distance. To evaluate the performance of the proposed algorithm, an experimental scheme is put forward based on man-made experimental datasets and typical fingerprint images. The experimental results show that the proposed algorithm estimates the average ridge distance accurately. 2009 Vol. 22 (4): 554-559 [Abstract] ( 324 ) [HTML 1KB] [ PDF 1563KB] ( 1034 ) 560 A Bi-Fuzzy Progressive Transductive Support Vector Machine Algorithm PENG Xin-Jun, WANG Yi-Fei Transductive support vector machine (TSVM) is a well-known algorithm that integrates transductive learning into support vector machine. In this paper, a bi-fuzzy progressive transductive support vector machine (BFPTSVM) is constructed by introducing the bi-fuzzy memberships and sample-pruning strategy for the temporary labeled samples. BFPTSVM is capable of degrading the computational complexity and the store memory of TSVM. Simulation results show that BFPTSVM has better classification and convergence performance compared with other learning algorithms. 2009 Vol. 22 (4): 560-566 [Abstract] ( 275 ) [HTML 1KB] [ PDF 442KB] ( 436 ) 567 A Fast Data-Oriented Algorithm for Principal Component Analysis YU Ying, WANG Bin, ZHANG Li-Ming Principal components analysis (PCA) for high-dimensional data is a difficult problem because the computational time and the space complexity rapidly increase as the data dimensions increase. A data-oriented and covariance-free PCA algorithm is proposed, inspired by the idea that the updated eigenvector in iteration is the weighted average of all samples. In a stationary environment or the condition that all training samples are available, the proposed algorithm is capable of overcoming the shortage of the conventional batch or incremental approaches. Furthermore, the convergence of the proposed algorithm is proved mathematically. Experimental results show that the most accurate solution is converged in a few iterations by the proposed algorithm. 2009 Vol. 22 (4): 567-573 [Abstract] ( 456 ) [HTML 1KB] [ PDF 621KB] ( 943 ) 574 Semi-Supervised Dimensionality Reduction Algorithm of Tensor Image ZHU Feng-Mei, ZHANG Dao-Qiang Traditionally, an (n1*n2) image is represented by a vector in the Euclidean space R(n1*n2), thus the spatial relationships between pixels in an image are ignored. In this paper, the images are presented as points in the tensor space Rn1Rn2. Then, a semi-supervised dimensionality reduction algorithm is put forward based on pairwise constraints (must-link and cannot-link)between the images. The data in the reduced space preserve the local structure of the data manifold well. Finally, experimental results on face datasets validate the effectiveness of the proposed algorithm. 2009 Vol. 22 (4): 574-580 [Abstract] ( 313 ) [HTML 1KB] [ PDF 693KB] ( 545 ) Surveys and Reviews 581 Object Recognition Models Based on Primate Visual Cortices: A Review YAO Xing-Zhong, LU Tong-Wei, HU Han-Ping In the past decade, a large number of experimental data have been accumulated in the primate visual cortices. Many visual system models have been developed based on the experimental data of the primates, and they have been used for computer object recognitions. In this paper, the main experimental conclusions on the primate visual system are summarized. Then, the frequently used models based on the mechanism of the primate visual system are reviewed. The MIT cortex hierarchical model is mainly discussed. 2009 Vol. 22 (4): 581-588 [Abstract] ( 274 ) [HTML 1KB] [ PDF 681KB] ( 891 ) Researches and Applications 589 A Multi-Objective Evolutionary Algorithm Based on Spatial Distance LI Mi-Qing, ZHENG Jin-Hua, XIAO Gui-Xia, XIE Jiong-Liang To improve the convergence of multi-objective evolutionary algorithm, a measure based on distance is proposed. A density estimation metric-tree crowding distance is defined. The individual distance and the tree crowing distance are used as the selection criteria when the non-dominated front is considered. When the size of non-dominated solution set exceeds that of the population, a method based on neighboring distance is employed to truncate population. By examining of five performance metrics on six test problems, the proposed algorithm is demonstrated to be more competitive in uniformity and spread and performs better in converging to the pareto front, compared to NSGA-II and SPEA2. 2009 Vol. 22 (4): 589-596 [Abstract] ( 268 ) [HTML 1KB] [ PDF 455KB] ( 738 ) 597 An Efficient Discrete Particle Swarm Optimization Algorithm for Multi-Criteria Minimum Spanning Tree GUO Wen-Zhong, CHEN Guo-Long A discrete particle swarm optimization (DPSO) algorithm is developed. To obtain a better approximation of true Pareto front, the phenotype sharing function of the objective space is applied in the fitness function. Inspired by the physics of genetic algorithm (GA), the principles of mutation and crossover operator in GA are incorporated into the proposed PSO algorithm to achieve better diversity and break away from local optima. The global convergence of the proposed algorithm is proved by the theorem of Markov chain. The experimental results show that DPSO is efficient and has good performance to problems with increased size. 2009 Vol. 22 (4): 597-604 [Abstract] ( 286 ) [HTML 1KB] [ PDF 552KB] ( 720 ) 605 Rapid Eye Localization Based on Projection Peak DAI Jing-Wen, LIU Dan, SU Jian-Bo An eye localization method based on projection peak is presented. Firstly, according to the proportional relationships of face features, the eye region from an face image is segmented by setting appropriate candidate window. Then, by means of histogram analysis of the eye region image, a threshold is obtained and binary transform is performed to segment the eyes out of the eye region. Next, a series of projection peak is derived from vertical and horizontal gray projection curves of the binary image. Through comparison and analysis, the exact coordinates of the eyes is finally confirmed. The presented method has no need of any previous knowledge and training process. Experimental results on three face databases show that the proposed method is effective, accurate and rapid in eye localization and it meets the requirement of real-time face recognition system entirely. 2009 Vol. 22 (4): 605-609 [Abstract] ( 282 ) [HTML 1KB] [ PDF 1063KB] ( 598 ) 610 Feature Extraction Method of Maximum Scatter Difference Based on Preserving Projection WANG Jian-Guo, YANG Wan-Kou, YANG Jing-Yu Firstly, the unsupervised discriminant projection (UDP) criterion is modified. Then, the feature extraction method of the maximum scatter difference based on preserving projection is proposed on the basis of the modified discriminant criterion. The proposed method adopts the difference of both nonlocal scatter and local scatter as discriminant criterion. Thus, the singular problem of local scatter caused by small sample size problem in UDP linear discriminant analysis is avoided. Finally, experimental results on Yale and FERET face databases demonstrate the effectiveness of the proposed method. 2009 Vol. 22 (4): 610-613 [Abstract] ( 291 ) [HTML 1KB] [ PDF 295KB] ( 560 ) 614 An Improved Collaborative Filtering Algorithm Based on Manifold Alignments ZHANG Fu-Zhi, ZHANG Qi-Feng Collaborative filtering by the learning manifold alignments provides a new way for cross system personalization. It uses similarity between users to compute reconstruction weights. However, inaccurate similarity often leads to inaccurate weights and poor recommendation quality. By combining the topology and geometry structures of data set to calculate weight matrix, an improved collaborative filtering algorithm based on manifold alignments is proposed. The proposed algorithm removes the effect of similarity error on recommendation quality effectively. The experimental result indicates that the improved algorithm has better recommendation quality than that of the original algorithm. 2009 Vol. 22 (4): 614-618 [Abstract] ( 313 ) [HTML 1KB] [ PDF 539KB] ( 433 ) 619 Feature Extraction and Personalized Feature Selection for Online Signature Verification ZHANG Da-Hai, WANG Zeng-Fu An online signature verification algorithm is presented based on feature extraction and feature selection. A novel digital tablet, called F-Tablet, is used to capture the signature information. The tablet can capture both shape series and five-dimensional forces. Total 188 features are extracted from each signature and then divided into three classes. Then, the weight function of features F is defined and the 188 features are sorted according to the F values. With different thresholds, different feature sets are obtained. The SVM is used to train the selected feature sets in the training process and the signatures are verified by the trained models. The proposed algorithm achieves false rejection rate (FRR) of 1.2% and false acceptance rate (FAR) of 3.7%. 2009 Vol. 22 (4): 619-623 [Abstract] ( 279 ) [HTML 1KB] [ PDF 333KB] ( 682 ) 624 Attribute Reduction Methods of Decision Table Based on Concept Lattice HU Xue-Gang, XUE Feng, ZHANG Yu-Hong, ZHANG Jing The exitsing reduction methods mainly use basic algorithm or heuristic algorithm based on discernibility matrix. However, the former can only be applied to the small dataset and the latter can not guarantee completeness. On the basis of studying the mapping relation between equivalence class and extension, the relevant solution of rough set based on concept lattice is mainly studied. Moreover, a complete reduction algorithm is proposed based on concept lattice and the test results show that the proposed algorithm enhances the performance of time and space. 2009 Vol. 22 (4): 624-629 [Abstract] ( 245 ) [HTML 1KB] [ PDF 420KB] ( 422 ) 630 A Robust Algorithm for Detecting Image Copy-Move Forgery WANG Jun-Wen, LIU Guang-Jie, DAI Yue-Wei, WANG Zhi-Quan Most of the existing region duplication detection algorithms are not robust to the post processing and have high time complexity. Therefore, an efficient and robust algorithm is proposed for detecting and localizing malicious tampering. The dimension of image is firstly reduced by DWT, and the geometric moment is applied to the fixed sized overlapping blocks of low-frequency in wavelet sub-band. Then, the eigenvectors are lexicographically sorted. Finally, the experimental threshold value and mathematical morphology operations are performed to locate the tampered part. The experimental results show that the proposed algorithm successfully detects the tampered image with various post region duplication image processing, including noise contamination and JPEG compression. Furthermore, the proposed algorithm reduces the total number of blocks to decrease time complexity. 2009 Vol. 22 (4): 630-634 [Abstract] ( 271 ) [HTML 1KB] [ PDF 1358KB] ( 482 ) 635 Face Recognition Based on Circularly Symmetrical Gabor Transforms and Weighted PCA WANG Jin-Jun, WANG Hui-Yuan, WU Xiao-Juan Circularly symmetrical Gabor transforms (CSGT) has two advantageous properties:reduced redundancy and rotation invariance. A face recognition method is proposed based on circularly symmetrical Gabor transforms and weighted PCA feature. The face image is transformed to CSGT, and then the discriminant features for face recognition are extracted using weighted PCA. Detailed theoretical analysis is presented and simulation results on three face databases are given. Comparative experiments of various face recognition schemes are carried out. The experimental results show the feasibility and advantages of the proposed method. 2009 Vol. 22 (4): 635-638 [Abstract] ( 257 ) [HTML 1KB] [ PDF 0KB] ( 94 ) 639 Interrelation Analysis Method of Celestial Spectrum Data Based on Constrained FP Tree ZHANG Ji-Fu, ZHAO Xu-Jun To search the unknown laws of celestial bodies is one of the objectives of human exploration of the universe. Utilizing the association rules is an effective way to find out the inherent and unknown interrelationships between characteristics of the celestial spectrum data and its physical and chemical properties. Using the national science project LAMOST as application background, constrained FP tree and its constructing algorithm are presented by taking first order predicate logic as knowledge representation technique of celestial spectrum data. Consequently, the interrelation analysis efficiency and the pertinence of celestial spectrum data are greatly improved. Thus, an interrelation analysis method of celestial spectrum data is proposed. The experimental results validate that the proposed method is feasible and valuable. 2009 Vol. 22 (4): 639-646 [Abstract] ( 291 ) [HTML 1KB] [ PDF 1252KB] ( 448 ) 647 Physicomimetic Method for Swarm Robots Search XIE Li-Ping, ZENG Jian-Chao Inspired by physicomimetic approach, a physicomimetics framework for swarm robots search is presented. The virtual forces among robots are defined by Newton's law of gravity. The relationships are constructed between robots' sensing intensity of the target signal and their virtual masses, and the virtual interaction rules are established among robots. The simulation results indicate the superiority of the proposed approach in search efficiency and precision. 2009 Vol. 22 (4): 647-652 [Abstract] ( 267 ) [HTML 1KB] [ PDF 441KB] ( 425 ) 653 Global Particle Swarm Based Cooperative Artificial Immune Network for Optimization LIU Li, XU Wen-Bo, WU Xiao-Jun A cooperative artificial immune network model is proposed. Inspired by global particle swarm intelligence, a cooperative artificial immune network, namely gpso-CoAIN, is developed for optimization. Due to the added global swarm cooperative operator, memory cells with particle swarm behavior are capable of sharing search experience. Furthermore, the clone selection procedure with variable step size of the artificial immune network is improved to adapt to fine optimal search. Experimental results of function optimization show that gpso-CoAIN outperforms several algorithms in optimal searching ability and running speed. The dynamic analysis illustrates the good diversity of the memory cells of the gpso-CoAIN network in the network population. 2009 Vol. 22 (4): 653-659 [Abstract] ( 282 ) [HTML 1KB] [ PDF 426KB] ( 554 ) 660 Ordered Decision Rules Extraction Algorithm Based on Granular Computing XU Jiu-Cheng, SHI Jin-Ling, ZHANG Qian-Qian An algorithm for extracting ordered decision rules based on granular computing is proposed to extract the most compact ordered decision rule from the ordered decision table. Firstly, an ordered decision table is transformed into the form of the ordered matrix by defining the concept of the ordered matrix and the λ-rank granular base about ordered decision table. Then, the ordered matrix and granular bases are studied and analyzed from different granularity level. Moreover, the algorithm is implemented for extracting the ordered decision rules, which satifies user expectation, as many as possible from the lower rank granular base with the search criteria of the lowest limitation of rule coverage and confidence. Finally, the validity for the algorithm is proved by analyzing examples. 2009 Vol. 22 (4): 660-665 [Abstract] ( 256 ) [HTML 1KB] [ PDF 337KB] ( 498 ) 666 Mean Shift Tracking Algorithm Based on Multi-Feature Space YU Dan, WEI Wei, ZHANG Yuan-Hui Due to the single feature space used in standard mean-shift algorithm, the confusion caused by the similarity object in its vicinity is hard to deal with. A variety of local pixel-level features are summarized, and a method for measuring the discrimination of various features is presented to make feature selection adaptive. Based on the analysis of deriving the weights during the iteration, an improved mean shift algorithm is proposed based on the discrimination measurement of various features through multi-feature space. It monitors the saliency of each feature effectively to compensate each other and improves the robustness to the confusion caused by the outlier. Experimental results indicate the proposed algorithm is real-time and robust and it has good tracking performance on object tracking. 2009 Vol. 22 (4): 666-672 [Abstract] ( 251 ) [HTML 1KB] [ PDF 1786KB] ( 447 )

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