模式识别与人工智能
Home      About Journal      Editorial Board      Instructions      Ethics Statement      Contact Us                   中文
Pattern Recognition and Artificial Intelligence
22 Judgement and Disposal of Academic Misconduct Article
22 Copyright Transfer Agreement
22 Proof of Confidentiality
22 Requirements for Electronic Version
More....
22 Chinese Association of Automation
22 National ResearchCenter for Intelligent Computing System
22 Institute of Intelligent Machines,Chinese Academy of Sciences
More....
 
 
2010 Vol.23 Issue.1, Published 2010-02-25
Orignal Article   
   
Orignal Article
1 Time Complexity Analysis of Ant Colony Algorithm on First Order Deceptive Problem
CHEN Ling,SUN Hai-Yin
The time complexity of ant colony optimization (ACO) on the deceptive systems is analyzed. Based on the study of the n-bit trap problem which is a first order deceptive problem of ACO, It is proved that time complexity for n-bit trap problem is O(n2mlnn) by max-min ant system (MMAS), which is an ACO with limitations on the pheromone on each edge. Here, n is the size of the problem and m is the number of artificial ants. The experimental results confirm the correctness of the conclusions.
2010 Vol. 23 (1): 1-6 [Abstract] ( 475 ) [HTML 1KB] [ PDF 347KB] ( 1033 )
7 Correlation Minimized Based 2-Dimensional Principal Component Analysis
YAN Hui,JIN Zhong,YANG Jing-Yu
It is indicated that two components belonging to the feature vector are correlated and the corresponding mathematical expression (2DPCA) is presented. The correlation minimized based 2-dimensional principal component analysis is proposed. It maximizes the total scatter of the feature vectors meanwhile minimizes the correlations of arbitrary two components belonging to the feature vector. The experimental results on Yale face database indicate that the proposed method has powerful ability of feature extraction and higher face recognition rates than 2DPCA and DiaPCA.
2010 Vol. 23 (1): 7-10 [Abstract] ( 345 ) [HTML 1KB] [ PDF 249KB] ( 602 )
11 A Fuzzy Cluster Validity Index Based on Clustering Mistake Measures
BEN Sheng-Lan,SU Guang-Da
The mistakes in fuzzy clustering can be categorized into two types: classifying data originated from different classes into one cluster and classifying data originated from the same class into different clusters. In this paper, intra-class non-consistency and inter-class overlapping are defined to measure the two kinds of mistakes respectively. A good fuzzy partition is expected to have few clustering mistakes and large compactness. Based on the two mistake measures and cluster compactness, a cluster validity index is proposed to evaluate the clustering results. Experimental results show the effectiveness and the robustness of the proposed validity index in determining optimal number of clusters.
2010 Vol. 23 (1): 11-16 [Abstract] ( 300 ) [HTML 1KB] [ PDF 376KB] ( 505 )
17 Structural Similarity Sparse Coding and Image Feature Extraction
LI Zhi-Qing,SHI Zhi-Ping,LI Zhi-Xin,SHI Zhong-Zhi
The structural similarity is introduced into sparse coding model, and a sparse coding model based on structural similarity is proposed. Then, the model is employed to extract the image sparse coding feature. The experimental results show that the improved sparse coding model is consistent with human visual system for its capacity of structural information preservation. Furthermore, compared with the standard sparse coding model, the proposed model attains the reconstructed image which preserves better structural information of the original image.
2010 Vol. 23 (1): 17-22 [Abstract] ( 358 ) [HTML 1KB] [ PDF 391KB] ( 822 )
23 Face Recognition Using Neighborhood Preserving Maximal Margin Analysis of Kernel Ridge Regression
LI Yong-Zhou,LUO Da-Yong,LIU Shao-Qiang
Neighborhood preserving embedding is a linear approximation to locally linear embedding, and it emphasizes preserving the local structure of the data manifold. The modified maximal margin criterion focuses on the discriminant and geometrical structure of the data manifold, and it improves the classification performance of the data. An algorithm is proposed called neighborhood preserving maximal margin analysis of kernel ridge regression. It preserves the local structure of the manifold and maximizes margins between the data of different classes to construct the objective function. As the data manifold is highly nonlinear, the kernel ridge regression is adopted to calculate the transformation matrix. The mapped results of the data samples are obtained by the proposed algorithm in the kernel subspace firstly, then the kernel subspace is obtained. The experimental results on the standard face database demonstrate that the proposed algorithm is correct and effective. Moreover, it achieves better performance than the popular manifold learning algorithms.
2010 Vol. 23 (1): 23-28 [Abstract] ( 311 ) [HTML 1KB] [ PDF 367KB] ( 728 )
29 Directional Evolutionary Algorithm Based on Fitness Gradient of Individuals
ZHAO ZHi-Qiang,GOU Jin,WANG Jing
The evolutionary direction is proposed based on the fitness gradient between the individuals of the current population and its parent. The evolutionary direction is analyzed qualitatively. The optimal evolutionary direction is proposed based on the gradient. The directional evolutionary algorithm (DEA) based on gradient of individuals is put forward under the description of evolutionary direction and optimal evolutionary direction. Two different reproduction strategies are proposed for DEA to generate individuals of next generation. The efficiency of DEA is validated theoretically. The experimental results show that the proposed algorithm has a high quality of precision, stability and convergence rate. Moreover, the improved evolutionary algorithm overcomes the shortcoming of low efficiency in traditional evolutionary algorithms to a certain extent.
2010 Vol. 23 (1): 29-38 [Abstract] ( 479 ) [HTML 1KB] [ PDF 599KB] ( 588 )
39 Based on Dynamic Programming along Region Boundary
LIU He-Wei,WANG Zeng-Fu

A region based stereo matching algorithm is presented. The presented algorithm uses regions in images as elements for disparity computation, which is different from conventional stereo matching algorithms based on dynamic programming along scan lines. Once the initial disparities are obtained, a multiple seeds based dynamic programming process along the region boundaries of the image improves and refines the imperfect disparities. The experimental results show that the proposed algorithm is fast and effective.
2010 Vol. 23 (1): 39-44 [Abstract] ( 379 ) [HTML 1KB] [ PDF 542KB] ( 606 )
45 Keyword Extraction Based on Lexical Chains for Chinese News Web Pages
HU Xue-Gang, LI Xing-Hua, XIE Fei, WU Xin-Dong
A lexical chain is an external performance consistency by semantically related words of a text, and it is the representation of the semantic content of a text. Based on the word ambiguity resolution, a method for keyword extraction from Chinese news web pages is proposed by using lexical chains combined with frequency features, location features and cohesion features. The document is represented as lexical chains by the relationship between phrases and the key phrases are extracted from the lexical chains. The proposed method is tested on the corpus of Chinese news web pages and journal articles. The experimental results show that the proposed method improves the quality of the keywords extraction.
2010 Vol. 23 (1): 45-51 [Abstract] ( 453 ) [HTML 1KB] [ PDF 515KB] ( 1206 )
52 A Proof of New Formula for 3D Images Euler Number
LIN Xiao-Zhu,JI Jun-Wei,HUANG Step-Hen,YANG Jian-Hua
The Euler number for digital images is one of the most important features of the digital topology parameter. The method for calculating the Euler number has been constantly explored to understand the nature of the Euler number for three-dimensional images better and to conveniently calculate the 3D image Euler number. Through the in-depth study of the connectivity of three-dimensional images, a new formula to locally calculate the Euler number for 3D images is proposed based on the two basic definitions of a 3D foreground run and a 3D neighbor number. Equivalence between the new formula and the global formula is proved by the induction method. A new way to locally calculate the Euler number for 3D images is provided which is unlike the description of the previous pixels and connectivity.
2010 Vol. 23 (1): 52-58 [Abstract] ( 409 ) [HTML 1KB] [ PDF 424KB] ( 709 )
59 SVM Speaker Verification Method of Mismatch Compensation Based on Factor Analysis
WU De-Hui ,LI Hui ,LIU Qing-Song ,DAI Bei-Qian
The poor performance of speaker verification system results from the channel mismatch and the lack of distinction between statistical models. A text-independent speaker verification method is proposed which combines the channel compensation based on factor analysis and the discriminative support vector machine (SVM) model. Gaussian mixture model (GMM) is used to make the speech parameter clustered and ascended, then the channel information of GMM mean super-vectors is wiped off by using factor analysis. The parameters, which are used as inputting parameters, are employed for the construction of SVM speaker verification system. The proposed method solves the problems of large samples, dimension raising and channel mismatch effectively when SVM is used for the text-independent speaker verification. Experimental results on NIST 06 male speaker corpus show that the proposed method improves system performance. Compared with the baseline system Gaussian mixture model-universal background model (GMM-UBM), GMM-SVM without channel compensation, the system improves the equal error rate (EER) more than 50%.Compared with the system factor analysis (FA)-GMM-UBM which uses channel compensation based on factor analysis without discriminative models, it also gets the improvement of EER by 15.8%.
2010 Vol. 23 (1): 59-64 [Abstract] ( 373 ) [HTML 1KB] [ PDF 424KB] ( 595 )
65 Semantic Web Service Discovery Based on Context and Action Reasoning
NIU Wen-Jia,CHANG Liang,Wang Xiao-Feng,HAN Xu,SHI Zhong-Zhi
Context around service itself and user is one of the most significant factors for semantic Web service discovery. Aiming at the drawbacks of the existing discovery approaches, a semantic Web service discovery based on context and action reasoning is proposed. An action-based context model is built to characterize the static and the dynamic context of Web services. And the action reasoning mechanism in dynamic description logic is used to realize context reasoning. Then, a context-aware discovering algorithm is implemented. The case study and the related work comparison show that the proposed approach has better performance on both context characterization and reasoning. Moreover, the results of experimental evaluation show that the proposed approach provides appropriate results of service discovery for user while reasonable time cost and space cost of logic reasoning are added.
2010 Vol. 23 (1): 65-71 [Abstract] ( 353 ) [HTML 1KB] [ PDF 503KB] ( 910 )
72 A Topic Model Based on CRP and Word Similarity
ZHANG Xiao-Ping,ZHOU Xue-Zhong,HUANG Hou-Kuan,FENG Qi,CHEN Shi-Bo
The topic model can extract the topics hided in documents to make the dimensions of documents reduced and the documents be classified and retrieved according to their topics. It is a research focus on information classification and retrieval fields. Aiming at the problem that the number of topics cannot be automatically determined in LDA topic model, a latent topic model is proposed by combining the similarity between words and Chinese restaurant process (CRP). It can adaptively update the contents and determine the rational number of topics. Meanwhile, a novel method of setting the hyperparameters during updating topics is put forward. The experimental results on traditional Chinese medicine (TCM) clinical dataset show the proposed model has good analysis results accepted by TCM expert.
2010 Vol. 23 (1): 72-76 [Abstract] ( 414 ) [HTML 1KB] [ PDF 322KB] ( 671 )
77 MW(2D)2PCA Based Face Recognition with Single Training Sample
LI Xin,WANG Ke-Jun,BEN Xian-Ye
Traditional methods get low recognition accuracy in the condition of only one training sample, therefore, it is a great challenge for face recognition. In this paper, a two-directional two-dimensional principal component analysis((2D)2PCA) is developed to solve this problem. An improved arithmetic combining weight and block called modular weighted (2D)2PCA is proposed for efficient local feature extraction. Besides, the fuzzy theory is introduced to classify the single sample face recognition. Experimental results on ORL and a subset of CAS-PEAL face databases show that the presented method achieves a high recognition accuracy.
2010 Vol. 23 (1): 77-83 [Abstract] ( 409 ) [HTML 1KB] [ PDF 467KB] ( 630 )
84 A Belief Propagation Algorithm for Stereo Matching
LU A-Li,TANG Zhen-Min,YANG Jing-Yu
Large-scale computing and high matching error rate are two disadvantages of the existing algorithms based on belief propagation. An efficient global optimal algorithm for dense disparity mapping is presented. Firstly, according to the feature of match cost and the disparity discontinuities accompanying the intensity changes, the adapted data constraint and the smoothness constraint are defined, and the messages are passed after the smoothness constraint is adjusted in every level. Then, the redundancy in message passing iteration process is discussed, and the message convergence is checked to decrease the running time. Finally, match symmetry is used to detect occlusions according to the analysis of matching errors in belief propagation algorithm. After the data term is reconstructed, a greedy method is used to iteratively refine the disparity result to propagate disparity information from the stable pixels to the unstable ones. The experimental results show the proposed algorithm computes a disparity map accurately with relative less time.
2010 Vol. 23 (1): 84-90 [Abstract] ( 304 ) [HTML 1KB] [ PDF 480KB] ( 744 )
91 Influences of Perturbations of Training Pattern Pairs on Fuzzy Morphological Associative Memory Networks
ZENG Shui-Ling,XU Wei-Hong,YANG Jing-Yu
Two kinds of morphological associative memory networks are analyzed by scholars and they have almost the same good fault-tolerant abilities of erosive/dilative noise. In this paper, the investigation reveals that two kinds of networks have different robustness to perturbation of training patterns. The robustness is good to one kind of fuzzy morphological associative memory network, and the property of the other kind is relatively poor. The conclusions offer guidance for the learning algorithm selection of the neural networks and the acquisition process of the previous training patterns.
2010 Vol. 23 (1): 91-96 [Abstract] ( 356 ) [HTML 1KB] [ PDF 353KB] ( 548 )
97 Modified Discrete Particle Swarm Optimization Algorithm Based on Inver-Over Operator
ZHENG Dong-Liang,XUE Yun-Can,YANG Qi-Wen,LI Fei
Though the discrete particle swarm optimization (DPSO) can make the best of the local and global optima of particles, it converges slowly with low precision. The Guo Tao algorithm converges with fast high precision, but it is blindfold to learn from the other particles. A modified discrete particle swarm optimization algorithm is presented based on the inver-over operator (IDPSO). To prevent premature convergence, the local sub-optimum particle swarm is introduced into IDPSO. Particles learn from the particles in the local sub-optimum particle swarm instead of their local optima. Three new parameters are introduced into IDPSO. Learning selection probability is introduced to select the particle to be learned. A generation threshold is introduced to define when to learn from the global particle. Local sub-optimum particle swarm ratio is introduced to define the size of the sub-optimum particle swarm. Selecting principles of these parameters is detailed discussed and the general reference scopes are given. Experiments are carried out on the traveling salesman problem and the results show that the modified IDPSO achieves good results compared with the Guo Tao algorithm and the general DPSO. The proposed algorithm improves both the convergence speed and solution precision.
2010 Vol. 23 (1): 97-102 [Abstract] ( 422 ) [HTML 1KB] [ PDF 428KB] ( 695 )
103 An Improved Character String Pattern Matching Algorithm
HU Jin-Zhu,XIONG Chun-Xiu,SHU Jiang-Bo,ZHOU Xin,CHENG Wen-Tao
By analyzing a variety of improved pattern matching algorithms, an pattern matching algorithm is improved. Firstly, the text string is preprocessed. Two kinds of characters are made up. One does not exist in the pattern string, and the other is the characters that appear the least. Secondly, through matching both the first character and the last character of the pattern string, the number of the marked characters which appear the least is reduced. If the matching fails, the pattern string directly slides to the next marked character which appears the least. Finally, experimental results show that the frequencies of the movement and the comparison decreased greatly by the improved algorithm, and the additional cost of space is less than the length of the pattern string. Moreover, the efficiency of the pattern matching is improved.
2010 Vol. 23 (1): 103-106 [Abstract] ( 348 ) [HTML 1KB] [ PDF 279KB] ( 597 )
107 Signature Verification AlgorithmBased on HMM/DTW Two-Level Architecture
YU Hong-Bin,WU Zhong-Cheng,SHEN Fei
A method for online hand-writing signature verification is proposed based on hidden Markov model (HMM) combined with dynamic time wrapping (DTW). To satisfy the inter-class consistency of signature segmentation and status division, the critical points in signatures are defined and its features are calculated. The difference values measured by DTW using two-dimension signature shape are used as observations of HMM to build the two-level architecture of HMMs for handwriting signature verification. The accuracy rate reaches about 93%.
2010 Vol. 23 (1): 107-114 [Abstract] ( 350 ) [HTML 1KB] [ PDF 504KB] ( 740 )
115 An Improved Algorithm for IR Object Recognition
LIU Jin,JI Hong-Bing
An improved fast independent component analysis (ICA) feature extraction algorithm is proposed based on one dimension search and distance function. Aiming at the problem that the selection of the initial value influences the convergence in fast ICA algorithm, the improved algorithm ensures the good convergence result by using one dimension search. Meanwhile, a rule based on distance function is designed to select the optimal features for the recognition according to the characteristics of the infrared image. Thus, the problem of the recognition rate and the robustness decreasing with the increasing number of training image samples is resolved. Compared with the traditional methods, the experimental results of real infrared images show that the proposed algorithm reaches a lower error classification rate with fewer infrared object features, and the error classification rate of the proposed method is robust in different kinds of classes.
2010 Vol. 23 (1): 115-119 [Abstract] ( 355 ) [HTML 1KB] [ PDF 0KB] ( 134 )
120 Collaborative FCPM Fuzzy Clustering Algorithm
QI Hong-Yu,WU Xiao-Jun,WANG Shi-Tong,YANG Jing-Yu
Fuzzy clustering with proportional membership (FCPM) algorithm can solve the clustering problem from different viewpoints. With the collaborative relations among different feature subsets, the collaborative fuzzy clustering is combined with other clustering algorithms to make its clustering result better than that of the one with the original algorithm. An improved fuzzy clustering algorithm is proposed based on the combination of FCPM and collaborative fuzzy clustering. The experimental results on the Wine and Iris datasets show the effectiveness of the proposed method.
2010 Vol. 23 (1): 120-126 [Abstract] ( 373 ) [HTML 1KB] [ PDF 364KB] ( 712 )
127 Modification of Two-Dimensional Entropic Thresholding Method and Its Fast Iterative Algorithm
WU Cheng-Mao,TIAN Xiao-Ping,TAN Tie-Niu
A modified method for two-dimensional entropic thresholding method and its fast iterative algorithm are proposed. Aiming at the disadvantage of high computational complexity of the classical two-dimensional entropic thresholding and its recursive algorithm, the two-variables probability distribution of two dimensional histogram is firstly modified and a new two-dimensional entropic thresholding method is obtained. Then, the fast iterative algorithm for the new modified two-dimensional entropic thresholding method is educed on the assumption that the modified two-variables probability distribution of two-dimensional histogram is continuous and differentiable. Experimental results show that the modified two-dimensional entropic thresholding method and its fast iterative algorithm are feasible, and the computational time of the fast iterative algorithm is much less than that of its recursive algorithm to a certain extent.
2010 Vol. 23 (1): 127-130 [Abstract] ( 347 ) [HTML 1KB] [ PDF 619KB] ( 745 )
模式识别与人工智能
 

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