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....

2008 Vol.21 Issue.4, Published 2008-08-01 Papers and Reports    Researches and Applications    Surveys and Reviews        Papers and Reports 425 Alternating Iterative One-against-One Algorithm LIU Bo, HAO Zhi-Feng, XIAO Yan-Shan One-against-one algorithm shows good performance in the multi-class classification algorithm based on SVMs. However, the existing middle unclassifiable region in the algorithm has a bad influence on its performance. To overcome this drawback, a method called alternating iterative one-against-one algorithm is proposed. And the validity analysis and computational complexity of the proposed algorithm are presented. Finally, one-against-one, fuzzy support vector machine (FSVM), decision directed acyclic graph (DDAG) and the proposed algorithm are compared on UCI datasets. The experimental results show that the proposed algorithm resolves the unclassifiable region problem effectively and its performance is better than that of the others. 2008 Vol. 21 (4): 425-431 [Abstract] ( 263 ) [HTML 1KB] [ PDF 365KB] ( 357 ) 432 Comparison between Two Kinds of PCA-Based Face Recognition Algorithmsby Sign Hypothesis Testing Strategy LI Le, ZHANG Yu-Jin PCA-based face recognition algorithms are actually classified into adaptive PCA based (APCA-based) algorithms and empirical PCA-based (EPCA-based) algorithms. The design principles and application characteristics of these two kinds of algorithms are analyzed. A new sign hypothesis testing strategy is designed to make objective comparisons between them on three common face databases. Two basic conclusions are drawn according to the comparison results. On one hand, as far as holistic performance is concerned, the difference between EPCA-based algorithms and APCA-based algorithms is relatively small if the training images have the same identity set as the gallery ones. Otherwise, the difference between them is very large. On the other hand, as far as the best realizable performance is concerned, there is no significant difference between them. Thus, some practical problems are analyzed and resolved. The conclusion provides a useful reference for deeply understanding and reasonably using PCA-based face recognition algorithms. 2008 Vol. 21 (4): 432-440 [Abstract] ( 231 ) [HTML 1KB] [ PDF 598KB] ( 378 ) 441 Rational User — A Sufficient Condition for Global Convergence in Interactive Evolutionary Computation HAO Guo-Sheng, HUANG Yong-Qing, ZHANG Yong, YAN Jun-Rong In interactive evolutionary computation (IEC), the relationship between user evaluation and user preference is an important factor of the convergence. Firstly, based on the dominated relationship between user evaluation and user preference, four kinds of users in IEC are put forward: absolute rational user, limited rational user, limited nonrational user and absolute nonrational user. Secondly, four theorems about the global convergence of IEC are proved. They illustrate the idea that the rational user is a sufficient condition for the global convergence of IEC. The theorems also point out that two kinds of elitist preservation strategies are necessary for the global convergence of IEC: fitness elitist preservation and satisfaction elitist preservation. Finally, the experimental results validate the above conclusion and show that it is a sufficient condition that as long as the user keeps rational, the algorithm convergence is ensured when other conditions are ready for the convergence. 2008 Vol. 21 (4): 441-445 [Abstract] ( 244 ) [HTML 1KB] [ PDF 609KB] ( 451 ) 446 An Average Reward Reinforcement Learning Algorithm with Tile Coding WANG Wei-Wei, CHEN Xing-Guo, GAO Yang Average reward reinforcement learning is an important undiscounted optimality framework. However, most of the work was based on discrete state space tasks. In this paper, how to combine function approximation with average reward learning is studied, and the parameter update condition is modified according to the continuous space. Besides, a close study on the performance of G-learning and its insensitivity to learning parameters is made. Finally, experimental results and relevant analysis are presented. The experimental results validate the solutions of R-learning and G-learning are prone to diverge when ε is relatively small. And the results also show that the Tile Coding is effective in function approximation as a feature extraction method and it can be taken as a comparative standard for other methods. 2008 Vol. 21 (4): 446-452 [Abstract] ( 316 ) [HTML 1KB] [ PDF 516KB] ( 667 ) 453 A Coevolutionary Algorithm Based on Dimension Identifying YANG Li-Ping, HUANG Hou-Kuan, YANG Xiao-Hong How to integrate the dimensional information of the problem domain into coevolution is studied. Through the analysis of the outcome characteristics of interactions between individuals, a strict dimension identifying method is proposed. Thus, an efficient and reliable coevolutionary algorithm is designed. It can automatically identify dimensions of the problem by the outcome characteristics between individuals with only the current highest test in each dimension maintained and monotonic progress on all dimensions sustained. In this algorithm, the archive can achieve minimum size to guarantee its practicability. Experimental comparisons demonstrate that the algorithm performs more efficiently than others. 2008 Vol. 21 (4): 453-461 [Abstract] ( 274 ) [HTML 1KB] [ PDF 530KB] ( 365 ) 462 Granularly Semantic Reasoning Based on Granular Computing in Granular Space YAN Lin, ZHANG Xue-Dong, WEI Yan-Tian, He Jian-Cang Based on a universal set, formulas are introduced and each formula corresponds to a granule. This gives rise to the definition of the granular space in which the granularly semantic reasoning is discussed. The reasoning is determined by the relationships among granules. By means of granular computing, it is proved that the granularly semantic reasoning satisfies each deductive rule in naturally deductive system of propositional logic. Thus, the soundness on the granular space is obtained, which means the formal reasoning can imply the granularly semantic reasoning. Although the completeness on the granular space is untenable, the results indicate that the granularly semantic reasoning is an expansion of the schemes of reasoning in propositional logic. Hence, the granularly semantic reasoning enlarges the scope of researches on reasoning as well as introduces a new method for the study of granular computing. Moreover, it promotes the integration of granular computing and logical reasoning. 2008 Vol. 21 (4): 462-468 [Abstract] ( 229 ) [HTML 1KB] [ PDF 333KB] ( 392 ) 469 An Expert Authority Computing Method Based on Real-Time Speech-Evaluation LI Min-Hua, DAI Ru-Wei, LI Yao-Dong To solve the problem of lacking effective method for group experts to interact in the hall for workshop of meta-synthetic engineering (HWME), a method for computing expert authority is proposed based on real-time speech-evaluation. The expert authority is obtained by computing speech quality and evaluation quality of the expert. The group interactive structure is built according to the speech-revaluation response relationships among experts. Experimental results demonstrate that the proposed method is feasible and it provides an ideal environment for experts to interact in HWME. 2008 Vol. 21 (4): 469-475 [Abstract] ( 240 ) [HTML 1KB] [ PDF 436KB] ( 391 ) 476 An Adaptive Anti-Collision Algorithm Based on Search Matrix DING Zhi-Guo, GUO Li, LIU Qi An adaptive anti-collision algorithm is proposed in the paper. To overcome the disadvantage of ABS algorithm, the information of collision bits is used and an EPC search matrix is determined in the proposed algorithm. The concept of collision stack is put forward.Based on the states of timeslots, the search paths can be adaptively adjusted. Theory and computer simulation results show that the proposed anti-collision algorithm is practical and can greatly decrease collision timeslots, idle timeslots and communication load. 2008 Vol. 21 (4): 476-481 [Abstract] ( 244 ) [HTML 1KB] [ PDF 558KB] ( 420 ) 482 LI Dun QIAO Bao-Jun, CAO Yuan-Da, WAN Yue-Liang With the development of internet and the rising of people's demand for information, the research on text orientation recognition is promising and challenging. As the basis of the text orientation recognition, the semantic orientation research of single word is presented in this paper. Using the integrated and detailed definitions of words in HowNet, two seed sets are built by the words with intense sentiment orientation. Then, with the context, the orientation similarity between seed words and common ones is calculated to recognize the sentiment orientation of the latter. The experimental results show that by the proposed method good results can be obtained in common word orientation recognition. Moreover, this method is of certain practical value for large granularity research. 2008 Vol. 21 (4): 482-487 [Abstract] ( 299 ) [HTML 1KB] [ PDF 422KB] ( 507 ) 488 Moving Object Detection Based on Omni-Directional Vision WANG Yu, WANG Yong-Tian, LIU Yue Based on omni-directional image characteristic, an algorithm is proposed to recognize and detect moving object with a static camera. First an omni-directional image is unwrapped through a fast unwrapping algorithm. Then the correction of the unwrapped image is performed based on a nonlinear distortion model. And an adaptive background modeling is built, which is real-time updated. Finally, the foreground is obtained to detect moving object. By the low resolution of the omni-directional correction image, the algorithm effectively solves problems of the noise and the shadow during the abstraction of the foreground. Experimental results show that the proposed algorithm is fast and effective. 2008 Vol. 21 (4): 488-493 [Abstract] ( 242 ) [HTML 1KB] [ PDF 810KB] ( 481 ) 494 Measurements of Discretization Schemes WANG Li-Hong, WU Geng-Feng Several measurements of the discretization schemes for continuous decision tables are discussed, including cut-point number, conditional entropy, granular entropy, class-attribute mutual information and interdependence redundancy. For consistent decision table, conditional entropy and class-attribute mutual information are both constants, and thus they can not offer more information for discretization schemes. The relationship between granular entropy and interdependence redundancy is analyzed. And it is proved that granular entropy increases when new cut points are added to the discretization scheme. A hybrid discretization algorithm is proposed to provide discretization schemes for testing. The simulation results show that the correlation coefficient between the cut-point number and classification accuracy is basically equal to that between granular entropy and classification accuracy, and both of them are correlated to datasets. 2008 Vol. 21 (4): 494-499 [Abstract] ( 259 ) [HTML 1KB] [ PDF 385KB] ( 283 ) 500 XIONG Wei-Qing In this paper, a new algorithm for classification rule mining is proposed, which is based on binary ant colony optimization algorithm. Aiming at the long computing time, a mutation operator is involved. To avoid the local optima problem, a disaster operator is also introduced. The algorithm is applied to the dataset from UCI machine learning repository, and the result shows that the forecasting accuracy is improved greatly. Moreover, by the mutation operator and disaster operator, the computing time can be effectively saved and the local optima can be avoided. 2008 Vol. 21 (4): 500-505 [Abstract] ( 266 ) [HTML 1KB] [ PDF 379KB] ( 319 ) Surveys and Reviews 506 A Survey of Differential Evolution Algorithms YANG QiWen, CAI Liang, XUE YunCan Differential evolution (DE) is a heuristic global optimization technique based on population. It is robust for real parameter optimization. To speed up the optimization and overcome the premature convergence of the heuristic optimization technique, many modifications are made to DE. The basic version of DE and its modifications are presented, and their advantages and disadvantages are also discussed. Some issues for further research on DE are addressed. 2008 Vol. 21 (4): 506-513 [Abstract] ( 433 ) [HTML 1KB] [ PDF 433KB] ( 1132 ) Researches and Applications 514 A Coarse-to-Fine Searching Method with Kernel Matching Based on Bhattacharyya Coefficients LI Liang-Fu, FENG Zu-Ren, CHEN Wei-Dong , ZHENG Bao-Zhong Mean shift is an efficient pattern match algorithm. Aiming at object tracking in large motion area, a mean shift algorithm is proposed based on coarse-to-fine searching with kernel matching. It can efficiently use a similarity measure function to realize the rough location of motion object. Then, the mean shift method is used to obtain the accurate local optimal value by iterative computing, and thus object tracking in large motion area is successfully realized. Experimental results show it has good performance in accuracy and speed compared with traditional algorithm. 2008 Vol. 21 (4): 514-519 [Abstract] ( 269 ) [HTML 1KB] [ PDF 685KB] ( 388 ) 520 Study of Data Mining in Intrusion Detection ZHOU Quan, ZHAO Feng-Ying, WANG Chong-Jun, CHEN Shi-Fu The solution based on multi-approaches of data mining involving k-means, C4.5, Nave Bayes, Bayes net and Co-training is proposed in order to deal with the major problems of intrusion detection dataset such as class balance, class overlapping, noise, distributions etc. The experiment results show its validity. 2008 Vol. 21 (4): 520-526 [Abstract] ( 218 ) [HTML 1KB] [ PDF 507KB] ( 331 ) 527 SAR Image Segmentation with Detail Preserving Based on Adaptive Neighborhoods TIAN Xiao-Lin, JIAO Li-Cheng, GOU Shui-Ping An adaptive neighborhood approach is proposed. The Markov random fields (MRF) segmentation approach with adaptive neighborhood systems is utilized to preserve detail features and border areas and to improve the segmentation effect. Bayesian inference is applied to integrate the different information sources of local image around the pixels. To improve the reliability of the belief value and the adaptivity, fuzzy c-means (FCM) clustering is introduced in Bayesian network. Thus, the selection of the neighborhood in the region label process need not depend on the known priori knowledge by applying the FCM. The neighborhood with the highest belief value in the threshold scope is chosen to compute the MRF region label process. Experimental results demonstrate that the segmentation effect of the proposed algorithm is superior to that of the classical MRF and hidden Markov random field with detail structures well preserved. 2008 Vol. 21 (4): 527-534 [Abstract] ( 245 ) [HTML 1KB] [ PDF 1431KB] ( 429 ) 535 An Improved Genetic Clustering Algorithm for Feature Extraction of Laser Scanner YU Jin-Xia, CAI Zi-Xing, DUAN Zhuo-Hua To automatically extract the environmental feature obtained by 2D laser scanner, an improved genetic clustering algorithm is presented. Firstly, a weighted fuzzy clustering algorithm is introduced to realize feature extraction of laser scanner after integrating the spatial neighboring information of range data into fuzzy clustering algorithm. Then, aiming at the unknown clustering number, the validities of different clustering algorithms are estimated by choosing a suitable index function for the fitness function of genetic algorithm. Moreover, to solve the local optimum of clustering algorithm, the genetic clustering algorithm is improved. The population diversity is increased and the genetic operators of elitist rule are improved to enhance the local search capacity and speed up the convergence. Compared with other algorithms, the effectiveness of the proposed algorithms is demonstrated. 2008 Vol. 21 (4): 535-540 [Abstract] ( 213 ) [HTML 1KB] [ PDF 572KB] ( 648 ) 541 Prediction of Speech Pauses Based on Punctuation Information and Statistical Language Model QIAN Yi-Li, XUN En-Dong Speech pauses are considered as punctuation marks of spoken language. People always insert different pauses at the boundaries of rhythmic phrases when communicating by language. Based on this characteristic, the speech pause of punctuation marks is investigated and the concept of predicting speech pauses using punctuation information is proposed. The punctuation-based and SLM-based methods are introduced to obtain training corpus and predict speech pauses. The influence of training corpus size on the performance of model is discussed. And the performance of punctuation-based corpus and manually-labeled corpus is compared. Experimental results show that the Chinese punctuation supplies valuable information on pause, and the method based on punctuation information can predict the Chinese speech pauses effectively. 2008 Vol. 21 (4): 541-545 [Abstract] ( 325 ) [HTML 1KB] [ PDF 387KB] ( 1083 ) 546 A Spatiotemporal Algorithm for Video Foreground and Shadow Segmentation CHU Yi-Ping , YE Xiu-Zi, HUANG Ye-Jue, ZHANG Yin, ZHANG San-Yuan Video segmentation is important for video object tracking, counting and recognition. Shadows are the factors that affect the accuracy of object segmentation. Efficiently detecting and removing the shadows can improve the quality of object segmentation. An algorithm for video-foreground and shadow segmentation is proposed in this paper. It models shadows with state machine and the shadows are removed according to the shadow models. The potential functions for the background, shadow and foreground are defined. The spatiotemporal neighboring relationships in the video sequence are constructed by using Markov random fields. Gibbs sampling algorithm is adopted to solve the MAP problem and thus the segmentation quality is improved . The correctness of the proposed algorithm is tested under different environments and the results demonstrate the validity of the algorithm compared with other algorithms. 2008 Vol. 21 (4): 546-551 [Abstract] ( 285 ) [HTML 1KB] [ PDF 729KB] ( 381 ) 552 Evidential Reasoning of Knowledge System Based on Fuzzy Rough Sets CHENG Yi, MIAO Duo-Qian, FENG Qin-Rong Fuzzy belief function and fuzzy plausible function are defined on Dubois fuzzy rough sets, Radzikowska fuzzy rough sets and fuzzy rough sets over two universes respectively. Then the equation relation between fuzzy belief function and lower approximation quality is proved. Thus, a model for evidential reasoning in fuzzy decision table is proposed. An example shows that the model is effective. 2008 Vol. 21 (4): 552-558 [Abstract] ( 222 ) [HTML 1KB] [ PDF 361KB] ( 315 ) 559 A Divisive Hierarchical Clustering Algorithm Based on Soft Hyperspheric Partition XIE Zhen-Ping, WANG Shi-Tong, WANG Xiao-Ming Hierarchical clustering is a classical data clustering method, but with two disadvantages—computational complexity and sensitivity to noises and outliers. To avoid these problems, a new divisive hierarchical clustering method is presented, called soft hyperspheric partition based divisive hierarchical clustering (SHPDHC). A new partitioning strategy, soft hyperspheric partition (SHP), is introduced. This strategy is derived from the possibilistic clustering method. SHPDHC has low computational complexity and has the ability of weakening the influence of outliers existing in the dataset, meanwhile, SHPDHC can easily produce the natural number of clusters. The theoretical analysis and experimental results on artificial datasets and real images demonstrate the effectiveness of the proposed method. 2008 Vol. 21 (4): 559-565 [Abstract] ( 294 ) [HTML 1KB] [ PDF 1586KB] ( 323 )

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