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.5, Published 2012-10-25 Orignal Article        Orignal Article 721 L1/2 Regularized Logistic Regression ZHAO Qian, MENG De-Yu, XU Zong-Ben A Logistic L1/2 regularization model with its efficient solution algorithm is proposed. By the proposed model, which is constructed on the basis of the L1/2 regularization theory, the variable selection capability is enhanced and the over-fitting problem of the traditional model is alleviated. The proposed algorithm with high computational efficiency is designed by the coordinate descent technique. The experimental results on synthetic and real datasets indicate that the proposed method outperforms the traditional Logistic regression and the L1 regularized Logistic regression on both variable selection and tendency prediction. 2012 Vol. 25 (5): 721-728 [Abstract] ( 574 ) [HTML 1KB] [ PDF 409KB] ( 4425 ) 729 Technical Architecture and Implementation of Intelligent System for Agriculture Domain XIONG Fan-Lun A series of studies on the technical architecture of intelligent system for agriculture domain is elaborated in this paper, including key technologies such as the expert system, knowledge representation, inference mechanism, knowledge acquisition, development platform, intelligent computing, machine learning, data mining, ontology, artificial life and their implementations to decision support, control, forecasting, retrieval , simulation in fertilization of crops, diagnosis and treatment of plant diseases and insect pests, cultivation, horticulture, livestock and aquaculture etc. It shows that application of intelligent and information technology into agricultural domain is of broad prospect. 2012 Vol. 25 (5): 729-736 [Abstract] ( 800 ) [HTML 1KB] [ PDF 1072KB] ( 802 ) 737 Monocular Pose Determination from Three Perpendicular Lines LIU Chang, ZHU Feng, OU Jin-Jun Monocular vision based pose determination is one of the typical problems in computer vision. If an object has three lines which are perpendicular to each other and intersect at two points, the pose parameters between a calibrated camera and the object can be calculated using 2D-3D correspondences of these three lines. A method is presented to find out the closed-form solutions of the problem. It is proved that the pose solution number depends on the location of camera’s optical center from the three lines. If the optical center locates between two planes, the problem has two symmetry solutions. Otherwise, it has a unique solution. The symmetry property of the solutions can be used to distinguish the real one from two possible solutions. Three perpendicular lines are three edge lines of a cuboid and the cuboids exist widely in real world. Therefore, the above results are useful to design or select cooperative targets in real applications using line features. 2012 Vol. 25 (5): 737-744 [Abstract] ( 688 ) [HTML 1KB] [ PDF 516KB] ( 780 ) 745 A Semi-Supervised Rough Set Model for Classification Based on Active Learning and Co-Training GAO Can, MIAO Duo-Qian, ZHANG Zhi-Fei, LIU Cai-Hui Rough set theory, as an effective supervised learning model, usually relies on the availability of an amount of labeled data to train the classifier. Howerer, in many practical problems, large amount of unlabeled data are readily available, and labeled ones are fairly expensive to obtain because of high cost. In this paper, a semi-supervised rough set model is proposed to deal with the partially labeled data. The proposed model firstly employs two diverse semi-supervised reducts to train its base classifiers on labeled data. The unlabeled ramified samples for two base classifiers are selected to be labeled based on the principle of active learning, and then the updated classifiers learn from each other by labeling confident unlabeled samples to its concomitant. The experimental results on selected UCI datasets show that the proposed model greatly improves the classification performance of partially labeled data, and even the best performance of dataset is obtained. 2012 Vol. 25 (5): 745-754 [Abstract] ( 568 ) [HTML 1KB] [ PDF 604KB] ( 829 ) 755 A Vocal Password System Based on Multi-Dimension Feature Classifier in Score Domain PAN Yi-Qian, WEI Si, Dai Li-Rong, LIU Qing-Feng As providing the score of different types of data the same weight, the average likelihood ratio verification measure used in GMM-UBM vocal password system brings decline in the system performance. Based on the different distinguished capacity between data types, a method is proposed in the score domain which classifies the test data by UBM, combines the likelihood ratio score of each class to form new multi-dimension feature, and then implements speaker verification by SVM. By use of the proposed strategy, the traditional likelihood ratio test is converted into a two-class classification problem in the multi-dimension feature space. The equal error rate of the proposed system is relatively 41.25%, 33.33%, 37.49% and 26.03% less than that of text-dependent GMM-UBM system in the co-channel experiments on four telephone corpuses respectively. The improvement of performance is also demonstrated through the cross-channel experiments. 2012 Vol. 25 (5): 755-761 [Abstract] ( 610 ) [HTML 1KB] [ PDF 472KB] ( 644 ) 762 Reduction Algorithm Based on Attribute Correlation YIN Lin-Zi, YANG Chun-Hua, WANG Xiao-Li, ZHOU Wei-Kang To obtain an optimal reduct in heuristic reduction methods, the attraction and repulsion correlations of attributes are analyzed and a definition of attribute significance is presented. On this basis, a heuristic reduction method based on attribute correlation is proposed to calculate an optimal reduct, which integrates the discernibility ability of the single attribute and the correlation among attributes. The experimental results show that the proposed method employs less heuristic calculations than the similar methods and the method based on attribute frequency, and it is more effective to obtain the optimal reduct. 2012 Vol. 25 (5): 762-767 [Abstract] ( 532 ) [HTML 1KB] [ PDF 348KB] ( 717 ) 768 Biogeography-Based Optimization Based on Hybrid Migration Strategy BI Xiao-Jun,WANG Jue To improve the performance of biogeography-based optimization (BBO), a biogeography-based optimization algorithm with hybrid migration strategy(HMBBO) is proposed. In this algorithm, the emigrated individuals are dynamically selected to balance the selection pressure for the solution set searching. A migration mechanism based on hybrid migration strategy is applied to enhance search capability and avoid premature convergence. The chaotic mutation mechanism is applied to improve the convergence precision for individuals. The experimental results on benchmark functions show that the HMBBO algorithm effectively avoids the premature convergence and improves convergence property and robustness compared to BBO algorithm. 2012 Vol. 25 (5): 768-774 [Abstract] ( 501 ) [HTML 1KB] [ PDF 461KB] ( 689 ) 775 Message Family Propagation Algorithm for Ising Graphical Model Based on Mean Field Computing Tree CHEN Ya-Rui A message family propagation algorithm based on mean field computing tree for the Ising graphical model is proposed. Firstly the concepts of mean field computing tree and mean field pruned computing tree are defined to describe the iteration computation process of the mean field inference of the Ising graphical model. Next, the message family propagation algorithm based on the mean field computing trees is designed. The proposed algorithm propagates message families from bottom to top in the computing tree and computes the marginal distribution families of root random variables. Then, the marginal distribution bound theorem is proved, which shows that the marginal distribution families computed by the algorithm in the pruned computing tree contain the exact marginal distributions. Finally, the theoretical and experimental results show that the message family propagation algorithm is valid and the marginal distribution bounds are tight. 2012 Vol. 25 (5): 775-782 [Abstract] ( 318 ) [HTML 1KB] [ PDF 461KB] ( 622 ) 783 Survey of Local Feature Extraction on Range Images GUO Yu-Lan, LU Min, TAN Zhi-Guo, WAN Jian-Wei Three dimensional (3D) object recognition is a hot research topic in computer vision. Local feature extraction is a key stage for 3D object recognition with the presence of occlusion and clutter. Firstly, range images and their representations are described. The differential geometric attributes are introduced, including the surface normal, the curvature and the shape index. Then, the local feature detection methods are classified into fixed scale method and adaptive scale method. And the local feature description methods are classified into depth value based, point spatial distribution based and geometric attributes distribution based methods. These methods with their merits and demerits are described. Finally, the existing methods are summarized and several challenges and future research directions are pointed out. 2012 Vol. 25 (5): 783-791 [Abstract] ( 1028 ) [HTML 1KB] [ PDF 478KB] ( 19614 ) 792 Boosting-Based k-NN Learning for Software Defect Prediction HE Liang,SONG Qin-Bao,SHEN Jun-Yi Timely identification of defective modules improves both software quality and testing efficiency. A software metrics-based ensemble k-NN algorithm is proposed for software defect prediction. Firstly, a set of base k-NN predictors is constructed iteratively from different bootstrap sampling datasets. Next, the base k-NN predictors estimate the software module independently and their individual outputs are combined as the composite result. Then, an adaptive threshold training approach is designed for the ensemble to classify new software modules. If the composite result is greater than the threshold value, the software module is recognized as defective, otherwise as normal. Finally, the experiments are conducted on NASA MDP and PROMISE AR datasets. Compared with a widely referenced defect prediction approach, the results show the considerable improvements of the ensemble k-NN and prove the effectiveness of software metrics in defect prediction. 2012 Vol. 25 (5): 792-802 [Abstract] ( 470 ) [HTML 1KB] [ PDF 496KB] ( 942 ) 803 Human Detection Method Based on Multi-Part Detector and Multi-Instance Learning DING Jian-Hao, GENG Wei-Dong, WANG Yi-Gang Part-based detection methods can deal with large articulated pose variations of human target and partial occlusions. Multi-instance learning is employed in content-based image retrieval and scene understanding, because it is good at handling the inherent ambiguity of images.A human detection method based on multi-part and multi-instance learning methods is presented. Firstly, the training samples are partitioned into several regions containing multi-instance according to body structure. Then, the part detectors are trained by using multiple instance learning method based on AdaBoost algorithm. After that the responding scores from the training samples tests are obtained by using the individual part detector when predicting on the positive and negative training bags. Therefore, the training samples are converted to feature vectors composed of part scores. The final assemble detector is learned using a linear SVM method. The experimental results on INRIA database show that the proposed approach improves the detection performance in single instance learning and the influence of the three different multi-part divisions on detection performance is evaluated. 2012 Vol. 25 (5): 803-809 [Abstract] ( 533 ) [HTML 1KB] [ PDF 577KB] ( 825 ) 810 Robust Object Tracking Algorithm by Particle Filter Based on Human Memory Model QI Yu-Juan, WANG Yan-Jiang Particle filter (PF) fails when the tracked object is occluded by other objects or its appearance changes. In this paper, human memory model is introduced into the template updating process of particle filter, which is inspired by the human memory mechanism, and a memory-based particle filter (MPF) algorithm is proposed. Each template is processed and transferred through ultra-short time memory space, short time memory space and long time memory space. The proposed memory-based model can remember what the template used to be, which helps the model adapt to the variation of object’s appearance more quickly. The experimental results show the effectiveness of the proposed method. 2012 Vol. 25 (5): 810-816 [Abstract] ( 466 ) [HTML 1KB] [ PDF 1280KB] ( 845 ) 817 Direction Visible Division in a Polygon ZHANG Yun-Hui, GAO Man-Tun, WU Jian-Jun, WANG Shu-Xia To achieve line burning trajectory calculations within a simple polygon, the concept of direction visible is proposed under the condition of line view, eight types of visible lines are found firstly, and seven kinds of bridge structure models are summarized to achieve direction visible division for a simple polygon. Polygonal interior is divided into two point visible areas and two line visible areas under direction projection and the bridges are constructed to complete boundary direction division by using the block relationship between main line and secondary line. Then, combined with point visible division algorithm, the polygon is divided into deep direction visible sub-polygons. Using these sub-polygons, the shortest path from any point to any line is derived within original polygon. Finally, the proposed algorithm is applied to line burning trajectory calculation in a polygon and obtains a good performance. 2012 Vol. 25 (5): 817-825 [Abstract] ( 440 ) [HTML 1KB] [ PDF 743KB] ( 642 ) 826 Clustering Uncertain Data Streams Based on Immune Principle XIAO Dan-Ping, YE Dong-Yi An algorithm based on immune principle, named IUMicro, is proposed to cluster uncertain data streams. IUMicro applies a dynamically updated immune model to adapt to the data streams. An effective B-cell feature vector and updating strategy are used to collect statistical information of data streams on line by this model. To choose the optimal candidate cluster for each increasing tuple in the data stream, IUMicro defines a probability radius of a B-cell’s recognition zone to address both uncertainty and distance metric. The offline clustering is an arbitrary-shape unsupervised clustering based on immune B-cells’ spatial relationship between regions. The experimental results show that IUMicro effectively suppresses noise and gains better clustering quality at a high processing speed. 2012 Vol. 25 (5): 826-834 [Abstract] ( 533 ) [HTML 1KB] [ PDF 736KB] ( 545 ) 835 Mining Method for Data Quality Detection Rules LIU Bo, GENG Yin-Rong Data quality rules are key to the database quality detection. To discover data quality rules from relational databases automatically and detect the error or abnormal data based on them, the form and evaluation measures of data quality rules are studied, and criterions of computing data quality rules are presented based on data item groups and the confidence threshold. The algorithms of mining minimal data quality rules and the main idea of detecting data errors using data quality rules are also given. The new form of data quality rules makes use of confidence mechanism of association rules and the expression of conditional functional dependencies to describe functional dependencies, conditional functional dependencies and association rules in the same format. It can be concluded that this kind of data quality rules has the properties of conciseness, objectivity, completeness and accuracy of detecting the error or abnormal data. Compared with other related research work, the proposed algorithms have lower temporal complexity, and the discovered quality rules improve the detecting rate. The effectiveness and correctness of the proposed methods are proved by the experiments. 2012 Vol. 25 (5): 835-844 [Abstract] ( 636 ) [HTML 1KB] [ PDF 434KB] ( 1411 ) 845 Contour Detection Model Based on Mechanisms of Visual Perception in Environment of Low Contrast CHEN Jian-Jun, REN Yong-Feng, ZHENG Guo-Yong The traditional contour detection model based on mechanisms of visual perception loses some of the object contours at low local contrast. And it augments the noise or breaks around the corner of the local high curvature contour in the collinear enhancement. To solve these problems, a contour detection model based on the mechanism of visual perception in the environment of low contrast is proposed. Based on the traditional model, the adjustment mechanism of stimulus contrast is introduced, which adjusts the size of receptive field and the strength of lateral inhibition. According to Gestalt principles of perceptual organization, the disinhibition region is divided into adjacent disinhibition region and peripheral disinhibition region which are given different enhancement mechanisms respectively. And the acquired binary image is morphologically filtered to eliminate noises. The experimental results for natural images show that the model has significant effects on above issues and it improves the performance of contour extraction in natural background. 2012 Vol. 25 (5): 845-850 [Abstract] ( 320 ) [HTML 1KB] [ PDF 681KB] ( 654 ) 851 An Ensemble Learning Method Based on CCA with Pairwise Constraints GUO Yun, ZHANG Dao-Qiang, SONG Tong The diversity among base classifiers is crucial for ensemble learning, and intuitively resampling pairwise constraints get better diversity than resampling instances. The supervision information in the form of pairwise constraints is introduced for feature extraction of samples to generate new training data based on canonical correlation analysis (CCA). In this algorithm, the spirit of ensemble learning is embodied in the way to select constraints. The constraints are resampled randomly to get the diverse base classifiers on multiview data. The experiments are carried out on multiple feature database and Yale and AR facial databases, and the results show that the proposed ensemble method achieves better performance than the conventional ensemble learning methods. 2012 Vol. 25 (5): 851-858 [Abstract] ( 629 ) [HTML 1KB] [ PDF 627KB] ( 692 ) 859 A Dictionary Learning Based Kernel Sparse Representation Method for Face Recognition ZHU Jie, YANG Wan-Kou,TANG Zhen-Min Inspired by Metafaces, a dictionary learning based kernel sparse representation method for face recognition is presented. Firstly, a kernel sparse representation classifier is proposed by extending sparse representation classifier to high dimensional space via kernel functions. Then, the kernel dictionary bases are learned based on Metafaces framework. Finally, the samples are reconstructed by kernel dictionary and the face images are classified according to the residual. The experimental results on AR, ORL and Yale face databases show that the proposed method works well. 2012 Vol. 25 (5): 859-864 [Abstract] ( 836 ) [HTML 1KB] [ PDF 451KB] ( 2471 ) 865 Linear Hybrid of Generative and Discriminative Classifiers SHI Hong-Bo, LIU Ya-Qin The generative approaches and the discriminative approaches are two kinds of paradigms for solving classification problems. To exploit the advantages of these approaches, a linear hybrid of generative/discriminative model (LHGD) is proposed, and a learning algorithm of LHGD based on genetic algorithms (LHGD_GA) is designed. LHGD_GA regards hybrid parameter learning of the linear hybrid classification model as an optimization problem, and utilizes genetic algorithms to find the best hybrid parameters of linear hybrid classification model. The experimental results show that the linear hybrid generative/discriminative classifier is better than or similar to the better classifier of two base classifiers on most datasets. 2012 Vol. 25 (5): 865-873 [Abstract] ( 457 ) [HTML 1KB] [ PDF 424KB] ( 893 ) 874 Convergence Analysis and Convergence Rate Estimate of Cellular Genetic Algorithms LI Jun-Hua, LI Ming Cellular genetic algorithms (cGAs) are a class of evolutionary algorithms (EAs) with a decentralized population in which the tentative solutions evolve in overlapped neighborhoods. However, there are few theoretical researches for the convergence and the convergence speed of cGA. A Markov chain that models canonical cGA is constructed. Then, the convergence of canonical cGA is deduced based on the absorbing state Markov chain. Next, the convergence rate of canonical cGA is studied. The upper and lower bounds for the number of iterations that canonical cGA gets a globally optimal solution are estimated. 2012 Vol. 25 (5): 874-878 [Abstract] ( 559 ) [HTML 1KB] [ PDF 305KB] ( 536 ) 879 Application of Improved Genetic Algorithms in Real-Time Differential Image Motion Picture Processing DU Zhuo-Ming, GENG Guo-Hua, XU Peng, WANG Jian-Ye Differential image motion picture processing is an important part of measurement of atmospheric coherence length, and the above measurement needs real-time data. Therefore, the measurement capabilities of the instrument correlate with the processing speed of the differential image motion picture. To improve the real-time measurement capabilities of the instrument of measuring the atmospheric coherence length, an improved genetic algorithm is devised based on the characteristics of the differential image motion picture to identify the target quickly. In order to speed up the rate of convergence, the crossover operator is canceled. Meanwhile, to avoid the premature convergence, the structure of the chromosome is improved to assure the global search capability of the improved algorithm when the mutation operator is used only. The results of simulation experiment on Schaffer function illustrate the feasibility of the improved algorithm. The new instrument of measuring the atmospheric coherence length achieves the target of the real-time measurement. 2012 Vol. 25 (5): 879-884 [Abstract] ( 594 ) [HTML 1KB] [ PDF 931KB] ( 616 )

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