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.3, Published 2010-06-30
Orignal Article   
   
Orignal Article
289 Algebraic Properties of Constrained Concept Lattice and Its Completeness of Knowledge Representation
ZHANG Ji-Fu,ZHANG Su-Lan,JIANG Yi-Yong
The constrained concept lattice is one type of concept lattice. It is constructed according to the background knowledge of usersinterest and knowledge on dataset. In this paper, the algebraic system of the constrained concept lattice is presented by defining operations of supremum and infimum among the constrained concept lattice nodes, and its algebra properties and completeness of knowledge representation are proved, thus the theoretical basis is established for the application of the constrained concept lattice.
2010 Vol. 23 (3): 289-299 [Abstract] ( 345 ) [HTML 1KB] [ PDF 591KB] ( 534 )
300 Data-Dependent Kernel Function Based Kernel Optimization Algorithm
LI Jun-Bao, GAO Hui-Jun
To solve the selection problem of kernel function and its parameters in kernel learning to enhance the performance of the algorihtm, data-dependent kernel function based kernel optimization method is proposed in this paper. The optimal objective function is built through the maximum margin criterion to solve the optimal parameter of data-dependent kernel. Experimental results show that the proposed algorithm can effectively increase the performance of kernel learning machine.
2010 Vol. 23 (3): 300-306 [Abstract] ( 287 ) [HTML 1KB] [ PDF 430KB] ( 615 )
307 A Method of Identifying Important Events from Text Collection Using Event Influence Relationship
ZHONG Zhao-Man,LIU Zong-Tian
A large amount of research results show that events objectively exist in a lot of texts, having essential inherent connections between them and different event has different importance. The matrix of event influence factor is constructed to depict the associative strengths between events of text collection. Based on the matrix of event influence factor, a method of identifying important events from text collection is elaborated by using event influence relations. This method utilizes the special timed transition relations between events and synthetically considers both hubs and authorities of events to compute event importance, abbreviated to HARank (Hubs-Authorities Rank). The experimental results show that the proposed algorithm can achieve significantly better ranking results for events over the classical PageRank and Reverse PageRank algorithms.
2010 Vol. 23 (3): 307-313 [Abstract] ( 307 ) [HTML 1KB] [ PDF 459KB] ( 585 )
314 Quantum-Inspired Estimation of Distribution Algorithm Based on Comprehensive Learning
TAN Li-Xiang, GUO Li
Quantum Inspired Evolutionary Algorithm(QIEA) adopts the framework that multiple simple probabilistic models parallel explore and it can improve its exploration ability consequently by introducing more effective multi-model learning mechanism. In this paper, the idea of comprehensive learning is introduced into the learning of multiple quantum probabilistic models, and Comprehensive Learning Quantum-inspired Estimation of Distribution Algorithm(CLQEDA) is proposed. In CLQEDA, the comprehensive learning is implemented by allowing each component of model to learn from different target solutions. It makes the quantum probabilistic model possible to relatively comprehensively extract knowledge from the known better solutions, then as comprehensively as possible to describe the good area in solution space and effectively improve the performance of the algorithm on complicated optimization problems. The advance and effectiveness of CLQEDA are testified via comparison experiments on classical 0-1 knapsack problems.
2010 Vol. 23 (3): 314-319 [Abstract] ( 423 ) [HTML 1KB] [ PDF 405KB] ( 595 )
320 A Method of Clustering Feature Vectors via Incremental Iteration
HUANG Rui,SANG Nong,LIU Le-Yuan,LUO Da-Peng,TANG Qi-Ling
A method of clustering in feature space is proposed in this paper via a kind of organization of data points. Firstly, those feature data points with higher densities which are relatively easy to be clustered are picked out as the initial seed data set. Then, the k-nearest neighbors of data in seed set are selected from the remained data points in feature space, and the data points in seed set and their k-nearest neighbors are transformed into a new space. In this space those data points are re-clustered, and the k-nearest neighbors are merged into current seed set. The above steps are iterated, and the clustering method will not terminate until there are no k-nearest points of the seed set to be found. Experimental results show that the clustering method performs better than the traditional clustering methods such as K-means, mean shift and spectral clustering.
2010 Vol. 23 (3): 320-326 [Abstract] ( 290 ) [HTML 1KB] [ PDF 511KB] ( 543 )
327 Ranking Model of Optimized Multiple Hyperplanes Using Order Relations
SUN He-Li,FENG Bo-Qin,HUANG Jian-Bin,ZHAO Ying-Liang,LIU Jun
A ranking model utilizing the multiple hyperplanes optimized by the order relations is proposed based on RankSVM in this paper. Firstly, the multiple hyperplanes are built based on the order relations between the ranks for training data in this model. Then, the ranking list generated by multiple hyperplanes is aggregated to gain the final ranking results. The proposed model is tested on LETOR OHSUMED dataset, some typical indices in Information Retrieval field being applied to evaluate its performance and the method being compared with other methods such as RankSVM. The experimental results show that the model not only has better ranking performance but also shorten the training time evidently.
2010 Vol. 23 (3): 327-334 [Abstract] ( 334 ) [HTML 1KB] [ PDF 476KB] ( 620 )
335 Super-Resolution through Dictionary Learning and Sparse Representation
PU Jian,ZHANG Jun-Ping
The overcomplete dictionary extracted from large scale dataset and sparse representation have been widely applied in image denoise, deblocking and inpainting in recent years. However, this technique can not be directly employed to deal with heterogeneous low resolution and high resolution image patches and relevant image reconstruction with super-resolution as well. The method to yield the sparse representation meeting two overcomplete dictionaries of different scales at the same time is proposed in this paper and the super-resolution reconstruction of image sparse representation is implemented by it. To further improve the super-resolution effect of color images, the UV chroma super-resolution reconstruction based on super-resolution luminance information is put forward as well. The experimental results show the method in this paper obtain better outcome no matter in visual effects or in root mean squared (RMS) error.
2010 Vol. 23 (3): 335-340 [Abstract] ( 367 ) [HTML 1KB] [ PDF 402KB] ( 1164 )
341 TP 301Similarity Measures on Vague Values and Three-Parameter Vague Values
LAN Rong,FAN Jiu-Lun
As an extension of vague values, the concept of three-parameter vague values is introduced in this paper, and a general similarity measure expression between three-parameter vague values is presented. Based on this expression, a corresponding general similarity measure between vague values is researched and a relatively general expression is given. When the parameters of the proposed similarity measure between vague values are selected as some special values, some existent similarity measure formulas between vague values can be obtained. The applications of proposed similarity measure in pattern recognition and medical diagnosis are shown.
2010 Vol. 23 (3): 341-348 [Abstract] ( 303 ) [HTML 1KB] [ PDF 458KB] ( 568 )
349 Quick Algorithm for Certain Rule Acquisition Based on Divide and Conquer Method
HU Feng,WANG Guo-Yin
Value reduction is a very important issue in rough set theory. Many efficient algorithms have been developed, however, few of them can process huge data sets quickly. In this paper, a quick algorithm for certain rule acquisition based on divide and conquer method is developed by dividing universe objects in attribute space. The proposed algorithm is illustrated by a case research as well. A certain rule set can be got quickly from a discrete decision table in this algorithm. If the data set is in uniform distribution, the time complexity of the algorithm is less than n2, which is fit to process large data sets efficiently. Experiment results show its high efficiency.
2010 Vol. 23 (3): 349-356 [Abstract] ( 276 ) [HTML 1KB] [ PDF 534KB] ( 752 )
357 Analysis of Stability for Equilibriumpoint Mean Square Exponential in Stochestic Celluler Neural Networks
LI Yu
The mean square exponential stability research on equilibriumpoint of stochastic delay cellular neural networks with pulse is mainly concerned by utilizing Lyapurov function. The primary theorems are deduced majorly by virtue of inequality and stochastic analysis theory. Finally one numerical example is presented to demonstrate its validity.
2010 Vol. 23 (3): 357-361 [Abstract] ( 290 ) [HTML 1KB] [ PDF 225KB] ( 561 )
362 A Survey of Research on Logic Model of Argumentation
XIONG Cai-Quan,SUN Xian-Bin,OUYANG Yong
The logic model of argumentation is the topic in many fields such as philosophy, logic, and artificial intelligence, and is widely applied in many areas such as non-monotonic reasoning, legal reasoning, group decision-making and interaction of multi-agent system (MAS). In this paper, the basic concepts and logic models of argumentation are firstly expounded. Then the present researches both on modeling for argumentation and on modeling with argumentation are summarized, and the characteristics and problems of current influential logic models of argumentation are analyzed. Finally, the existing challenges and potential research directions of the logic model of argumentation are pointed out.
2010 Vol. 23 (3): 362-368 [Abstract] ( 386 ) [HTML 1KB] [ PDF 528KB] ( 586 )
369 A Latent Variable Model Based on Local Preservation
WANG Xiu-Mei,GAO Xin-Bo,ZHANG Qian-Kun,SONG Guo-Xiang
Latent variable model (LVM) is a kind of efficient nonlinear dimensionality reduction algorithm through establishing smooth kernel mappings from the latent space to the data space. However, this kind of mappings cannot keep the points close in the latent space even they are close in data space. A LVM is proposed based on locality preserving projection (LPP) which can preserve the locality structure of dataset. The objective function of LPP is considered as a prior of the variables in the Gaussian process latent variable model (GP-LVM). The proposed locality preserving GP-LVM is built with the constrained term of the objective function. Compared with the traditional LPP and GP-LVM, experimental results show that the proposed method performs better in preserving local structure on common data sets.
2010 Vol. 23 (3): 369-375 [Abstract] ( 336 ) [HTML 1KB] [ PDF 477KB] ( 683 )
376 KNN and RVM Based Classification Method: KNN-RVM Classifier
ZHANG Lei,LIU Jian-Wei, LUO Xiong-Lin
Aimming at the problems of relevance vector machine (RVM) classification such as low precision and difficulty in kernel parameter selection, a concept called critical sliding threshold is presented in this paper. A classifier combining RVM with K nearest neighbour (KNN) called KNN-RVM classifier is constructed. In theory, three theorems is proposed and proved. The first is that the process of KNN-RVM classification is equivalent to an implementation of soft margin SVM. The second is that KNN-RVM classifier is equivalent to a 1NN classifier in which only one representative point is selected for each class. The last is the result of KNN-RVM classification is superior to that of RVM classification. The sliding and critical characteristics of critical sliding threshold are proved using three different datasets. The veracity, adaptability and global optimality of KNN-RVM classifier are proved as well. The KNN-RVM classifier improves the classification precision, reduces the reliance of algorithm on the kernel parameter, and thereby is proved to be an effective and excellent classifier.
2010 Vol. 23 (3): 376-384 [Abstract] ( 524 ) [HTML 1KB] [ PDF 581KB] ( 888 )
385 An α-Expansion Stereo Algorithm Based on Segment-Constraint
LU A-Li,TANG Zhen-Min

An improved stereo algorithm based on graph cuts is developed. Firstly, a robust and adaptive energy function is defined and its graph-representability is proved. The function employs rank transform to reduce noises of data term and utilizes adaptive truncated liner model based on color similarity to preserve the discontinuities of disparity. Secondly, the complexity of the existing graph cuts based stereo algorithms is analyzed, and an α-expansion operation based on segment-constraint is presented. According to disparity smoothness in color connectivity regions, the search range for every pixel is reduced and distance transform is introduced to get candidate correspondences of the disparity α as vertices of the constructed graph in each expansion, thus the computation cost of the maximum flow on graphs is reduced. Finally, α-expansion is operated in descending sequence of the disparity distribution to reduce the total number of iteration. The experimental results show that the improved algorithm can effectively raise the computation efficiency and matching accuracy of stereo ones based on graph cuts.
2010 Vol. 23 (3): 385-395 [Abstract] ( 333 ) [HTML 1KB] [ PDF 717KB] ( 591 )
396 Video Feature Extraction Based on Improved Locality Preserving Projections
XIAO Yong-Liang,XIA Li-Min
A method to extract video feature is introduced. To solve the problems related to the projection dimension and nearest neighbor K in locality preserving projections (LPP), the method to determine the optimal projection dimension based on structure error between dimension reduction before and after is proposed in this papers. The nearest neighbor K is dynamically selected combining with the neighbor statistical character of each data. On the basis of the above an optimal projection matrix of video feature is obtained by using LPP, and then the high dimension feature of new video is reduced to a lower one through the projection matrix. The comparison of experimental results show that the feature based on LPP is more favorable for shot segmentation than the other features.
2010 Vol. 23 (3): 396-401 [Abstract] ( 284 ) [HTML 1KB] [ PDF 370KB] ( 520 )
402 Expecting Association Rule Set and Quantitative Analysis for Its Base Number
LI Kai-Li,WANG Li-Hong,TONG Xiang-Rong
It is deserved to foresee how many association rules will be mined from a given database without taking support and confidence in consideration. Hence the concept of expecting association rule is proposed in this paper to turn the above problem into how to calculate the base number of an expecting association rule set. The categorical and continuous computing formulas are presented respectively. The exclusive property of items in itemset is discussed after the transformation of continuous data into categorical data. An expanding matrix and an expanding method is deduced by the exclusive property. This method is used to calculate the base number of an expecting association rule set of continuous dataset in a brief way. The analysis and test results show that the size of an expecting association rule set decrease as the amount of exclusive items increase. These results are helpful to understand the essence of association rule mining and furtherly develop more highly efficient mining algorithm.
2010 Vol. 23 (3): 402-407 [Abstract] ( 313 ) [HTML 1KB] [ PDF 391KB] ( 582 )
408 Multi-Touch Gesture Recognition Based on Petri Net and Back Propagation Neural Networks
WANG De-Xin,SHI Chong-Lin,ZHANG Mao-Jun
To recognize gestures of multi-touch system, a framework including gesture description and recognition is proposed. Multi-touch gesture can be decomposed into atomic gestures and composite gestures. For gesture description, the back propagation neural networks (BPNN) is used to model the atomic gesture. Users motions are mapped into composite gestures combined with atomic gestures logic, temporal and spatial relations. Petri nets (PN) introduced with logical, temporal, spatial descriptors are used to model the composite gesture. For recognition, BPNN is used as a classifier to recognize atomic gestures and the recognition results trigger the transition of Petri nets for composite gestures to realize the recognition. Experimental results show that the proposed method is robust to different users customs and can recognize multi-touch gestures effectively.
2010 Vol. 23 (3): 408-413 [Abstract] ( 310 ) [HTML 1KB] [ PDF 407KB] ( 576 )
414 A Kernel Fisher Linear Discriminant Analysis Approach Aiming at Imbalanced Data Set
YIN Jun-Mei,YANG Ming,WAN Jian-Wu
In practical real applications lots of classification questions are aiming at imbalanced data sets, while these unbalanced data will lead to the descending of the classification performance of many classifiers. In this paper the classification mechanism based on kernel fisher linear discriminant analysis (KFDA) is introduced, and then the reasons that the unbalanced data cause KFDA to turn ineffective is analyzed. Therefore, a weighted kernel fisher linear discriminant analysis (WKFDA) method is proposed. The method balances the contributions from kernel covariance matrices of two classes of sample to the kernel within-class scatter matrix and can constrain the influence of unbalanced data on classification performance. The experiments on 7 UCI datasets are performed to further test the performance of our algorithm. The experimental results show that the developed approach can effectively improve the classification performance of the proposed classifier.
2010 Vol. 23 (3): 414-420 [Abstract] ( 441 ) [HTML 1KB] [ PDF 389KB] ( 731 )
421 Abnormal Behavior of Pedestrian Detection Based on Fuzzy Theory
ZHANG Jun,LIU Zhi-Jing
To automatically identify pedestrian abnormal movement in Intelligent Monitoring System, a simplified articulated model of human body is presented. A fuzzification function using the variety in trunk and limbs contour angles of pedestrian is designed. Then an abnormal behavior discrimination algorithm based on fuzzy theory is proposed. The algorithm applies fuzzy membership of the trunk and limbs of pedestrian to get the overall degree of anomaly. In the system realization, a method of combining center of mass trajectory and fuzzy discriminant is proposed to discriminate the anomaly of pedestrian. Fuzzy discriminant can implement active analysis of pedestrian behavior in visual surveillance and thereby detect irregularities to recognize abnormal behavior and alarm. The experimental results show that the proposed algorithm has a higher recognition rate.
2010 Vol. 23 (3): 421-427 [Abstract] ( 400 ) [HTML 1KB] [ PDF 444KB] ( 869 )
428 An Alternative Ranking Method Based on Deviation Degree of Interval Numbers
XU Yan-Xia,LI De-Yu,HU Jian-Long
For the alternative ranking problem with attribute values being interval numbers, the concept of deviation degree of two interval numbers is presented to construct the deviation degree matrix for alternatives and a method to determine weight vector based on the total deviation degree optimization is proposed as well in this paper. By using the proposed concept of relative closeness degree of interval number vectors, the degree of an alternative close to another is depicted, and thereby the merit of alternatives are ranked. The proposed measures of this method are intuitive and easy to understand, and an illustrative example is employed to demonstrate its feasibility and practicability.
2010 Vol. 23 (3): 428-433 [Abstract] ( 302 ) [HTML 1KB] [ PDF 340KB] ( 561 )
434 Active Exploration of Visual SLAM Based on Comprehensive Mutual Information
SUN Feng-Chi,KANG Ye-Wei,HUANG Ya-Lou,LIU Guang
The active Simultaneous Localization and Mapping(SLAM) problem based on a single camera is discussed in this paper. Regarding the defect of available research results based on mutual information that neglect costs by different motion activities, the exploration strategy based on comprehensive mutual information that consider motion costs is proposed. The costs of movements including displacement distance and rotation degree of the camera are taken into account to choose movement activities. Each activity is assigned to a certain weight according to the extent that it affects the reliability of landmark matching. Then the activity obtaining largest information gain under unit cost is selected. The experimental result shows that without affecting location precision and real time property, the SLAM of single camera based on the proposed exploration strategy can choose feasible motion activity scientifically and avoid the shortage of the exploration mode pursuing information benefits only.
2010 Vol. 23 (3): 434-440 [Abstract] ( 449 ) [HTML 1KB] [ PDF 482KB] ( 680 )
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