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.4, Published 2012-08-25 Orignal Article        Orignal Article 557 Generalized Rough Set Model Based on Strong Symmetric Binary Relation MA Zhou-Ming, LI Jin-Jin The axiomatization is one of the most important topics in rough set theory. The definition of strong symmetric binary relation is presented by analyzing axiomatic characterizations of symmetric approximate operators. In contrast with the properties of equivalence relation, some important characteristics of the proposed binary relation are presented, and the necessary and sufficient condition for a symmetric binary relation becoming a strong symmetric one is given. The properties of corresponding generalized rough set are investigated, and the corresponding axiomatic group is studied. Utilizing of the correlation between these axioms and accurate sets, the characteristics of accurate sets in the generalized rough set based on a binary relation are discussed, and some assistance is provided to method and application of rough set theory. 2012 Vol. 25 (4): 557-563 [Abstract] ( 949 ) [HTML 1KB] [ PDF 307KB] ( 661 ) 564 Three-Dimensional Grid Semantic Map Building in Unstructured Indoor Environment WU Hao, TIAN Guo-Hui, WANG Jia-Chao, ZHOU Feng-Yu Aiming at the complicated unstructured indoor environment and the dynamic service task of robots, an environment cognition method based on quick response (QR) code technology is proposed. Under the premise of gaining depth information with binocular vision, the uncertain mathematic model of binocular vision information is established on Dezert-Smarandache theory (DSmT) evidence theory. And the three-dimensional grid map is formed which describes the occupied/free probability of the voxel. While the three-dimensional map is structured, the semantic labels for the large objects are added by QR code based object mark plastered on large objects. The occupied/free values of the corresponding voxels are updated based on the dimension of the large object. Then, the three-dimensional grid semantic map is formed which includes the function property and the attributive relation of the large objects. The experiments are carried out to compare several information fusion algorithms with the proposed method and analyze the recognition accuracy for the artificial object marks. The results prove the availability and feasibility of the proposed method. 2012 Vol. 25 (4): 564-572 [Abstract] ( 585 ) [HTML 1KB] [ PDF 1311KB] ( 839 ) 573 A Multi-Label Classification Method Based on Tree Structure of Label Dependency FU Bin, WANG Zhi-Hai In multi-label learning, the performance of a learning algorithm can be improved by discovering and making use of the dependencies within the labels. In this paper, an innovated algorithm for multi-label learning based on the classifier chain model is proposed. This algorithm mainly consists of two steps. The dependencies are quantified firstly using mutual information, and then a tree structure of labels is derived to depict the relationship within labels. Thus, the randomness of dependencies in classifier chain is weakened, and the linear dependency is generalized to a tree structure one. To further utilize the dependencies, ensemble technique is used to learn and aggregate multiple trees of labels. The experimental results show that the proposed algorithm is also competitive alternative and it improves the performance significantly after ensemble learning especially, hence it can learn the dependencies within labels more effectively. 2012 Vol. 25 (4): 573-580 [Abstract] ( 816 ) [HTML 1KB] [ PDF 420KB] ( 713 ) 581 Regularized Locality Preserving Discriminant Analysis GU Xiao-Hua, GONG Wei-Guo, YANG Li-Ping A regularized locality preserving discriminant analysis (RLPDA) for face recognition is proposed. Affected by the small sample size (SSS) problem and noises, zero eigenvalues and small eigenvalues of locality preserving within-class scatter matrix are inadequate. It degrades the performance of discriminant locality preserving projections (DLPP). In this paper, eigenvalues of locality preserving within-class scatter matrix are regularized by a reciprocal spectrum model, and the subspaces are weighted according to the regularized eigenvalues. Specifically, the face subspace is kept, the noise subspace is weakened, and the zero subspace is enhanced. Through the analysis of the distribution of discriminant information in data space, it is found that RLPDA utilizes the whole discriminant information. Hence, RLPDA improves the recognition accuracies and avoids the SSS problem in principal. The experimental results on FERET and UMIST face databases illustrate the effectiveness of the proposed RLPDA algorithm. 2012 Vol. 25 (4): 581-587 [Abstract] ( 297 ) [HTML 1KB] [ PDF 437KB] ( 504 ) 588 Image Retrieval Method Based on Selected Relation Embedding Algorithm LIU Li, TAO Dan, PENG Gang The key of image retrieval based on manifold learning and relevance feedback is to learn user semantics from a few feedbacks based on the low-level visual information, to get semantic subspace manifold. To get more real semantic subspace, the low-level visualization and the user feedback information are differentiated, and the relations of inter-class and intra-class are learned selectively from feedback information based on visualization feature. Then, a selected relation embedding algorithm for image retrieval is proposed. By this algorithm, more realistic semantic manifold structures are kept, and the retrieval precision in low-dimensional space is enhanced. The experimental results show that the proposed method mapps the image into a wider range of low-dimensional space, and improves the retrieval precision up to 16.3% after two feedback iterations. 2012 Vol. 25 (4): 588-594 [Abstract] ( 769 ) [HTML 1KB] [ PDF 640KB] ( 675 ) 595 A Classifier Ensemble Algorithm Based on Local Random Subspace YANG Ming, WANG Fei Classifier ensemble learning is one of the present research focuses in machine learning field. However, the classical method of completely random subspace selecting can not guarantee good performances of sub-classifiers for high dimension datasets. Therefore, a classifier ensemble algorithm based on local random subspace is proposed. The features are ranked by employing feature selection strategy firstly, and then the ranked feature list is partitioned into a few parts and the randomly feature is selected in each part according to the given sampling rate. Thus, the performances of sub-classifiers and their diversities are improved. Experiments are carried out on 5 UCI datasets and 5 gene datasets. The experimental results show that the proposed algorithm is superior to a single classifier, and in most cases it is better than those classical classifier ensemble methods. 2012 Vol. 25 (4): 595-603 [Abstract] ( 787 ) [HTML 1KB] [ PDF 390KB] ( 722 ) 604 Multi-Level Speech Emotion Recognition Based on Fisher Criterion and SVM CHEN Li-Jiang, MAO Xia, Mitsuru ISHIZUKA To solve the speaker independent emotion recognition problem, a multi-level speech emotion recognition system is proposed to classify 6 speech emotions, including sadness, anger, surprise, fear, happiness and disgust from coarse to fine. The key is that the emotions divided by each layer are closely related to the emotional features of speech. For each level, appropriate features are selected from 288 candidate features by Fisher ratio which is also regarded as input parameter for the training of support vector machine (SVM). Based on Beihang emotional speech database and Berlin emotional speech database, principal component analysis (PCA) for dimension reduction and Artificial Neural Network (ANN) for classification are adopted to design 4 comparative experiments, including Fisher+SVM, PCA+SVM, Fisher+ANN, PCA+ANN. The experimental results prove that Fisher rule is better than PCA for dimension reduction, and SVM is more expansible than ANN for speaker independent speech emotion recognition. Good cross-cultural adaptation can be inferred from the similar results of experiments based on two different databases. 2012 Vol. 25 (4): 604-609 [Abstract] ( 922 ) [HTML 1KB] [ PDF 456KB] ( 1350 ) 610 A Hybrid Particle Swarm and Multi-Population Cellular Genetic Algorithm LI Ming, JIE Li-Lin, LU Yu-Ming Cellular genetic algorithm (CGA) enhances global convergence rate via constraining individual interaction in its neighbor. However, it results in of low search efficiency. An algorithm, called hybrid particle swarm and multi-population cellular genetic algorithm (HPCGA), is proposed. Firstly, the whole population is divided into some sub-populations,the individuals in different sub-populations do not interact each other. Nevertheless different sub-populations can communicate with each other via immigrant and share the evolutionary information. Division of the population appropriately reduces the selection pressure, and thus the individual diversity is maintained more effectively. The mutation of CGA is replaced by particle swarm optimization to improve the ability of local search. The above two improvements balance the trade-off between global exploration and local exploitation. Selection pressure and individual diversity of the proposed HPCGA are also studied. Optimization of six typical functions is carried out by using the proposed HPCGA and CGA. The experimental results show that the performance of the proposed HPCGA is obviously superior to that of CGA in global convergence rate, convergence speed and stability. 2012 Vol. 25 (4): 610-616 [Abstract] ( 829 ) [HTML 1KB] [ PDF 431KB] ( 722 ) 617 An Improved KNN Algorithm Based on Variable Precision Rough Sets YU Ying, MIAO Duo-Qian, LIU Cai-Hui, WANG Lei K Nearest Neighbor (KNN) is a simple, stable and effective supervised classification algorithm in machine learning and is used in many practical applications. Its complexity increases with the number of instances, and thus it is not practicable for large-scale or high dimensional data. In this paper, an improved KNN algorithm based on variable parameter rough set model (RSKNN) is proposed. By introducing the concept of upper and lower approximations in variable precision rough set model, the instances of each class are classified into core and boundary areas, and the distribution of the training set is obtained. For a new instance, RSKNN firstly computes the area it belongs to. Then, according to the area information, the algorithm determines the category directly or searches k-nearest neighbors among the related areas instead of all areas. In this way, the computing cost is reduced and the robustness is enhanced. The experimental results for selected UCI datasets show that the proposed method is more effective than the traditional KNN with high classification accuracy. 2012 Vol. 25 (4): 617-623 [Abstract] ( 769 ) [HTML 1KB] [ PDF 376KB] ( 776 ) 624 Image Mosaicing Algorithm for Dynamic Scenes Using Multi-Scaled PHOG Feature and Optimal Seam ZOU Li-Hui, CHEN Jie, ZHANG Juan, LU Jing-Hua Aiming at the problems of registration error and synthetic movement ghost which are caused by moving objects in image mosaicing, a mosaicing algorithm for dynamic scene using multi-scale pyramid histogram of oriented gradients (PHOG) and optimal seam is proposed. Firstly, a new feature, multi-scaled PHOG, is generated by introducing PHOG to multi-scale space corner detections. The feature is used to align images for avoiding the local impact caused by moving objects in image registration. Then, an optimal seam, guaranteeing the minimum difference in geometry and gray value, is searched by graph cut algorithm through constructing an energy function to remove the movement ghost. The experimental results show that the proposed algorithm is efficient in dealing with the problems of image mosaicing with moving objects, and the mosaicing results are satisfactory with high precision. 2012 Vol. 25 (4): 624-631 [Abstract] ( 464 ) [HTML 1KB] [ PDF 2059KB] ( 741 ) 632 Argumentation-Based Non-Monotonic Reasoning of Agents LIAO Bei-Shui, DAI Jian-Hua Most of existing theories and methods for belief revision, deliberation, means-ends reasoning etc. are based on classical first order logic, and therefore effective mechanisms are absent in handling incomplete and inconsistent knowledge. Argumentation-based non-monotonic reasoning, including epistemic reasoning and practical reasoning, has become a promising theory to solve the above-mentioned problem. However, as an emerging research area, the basic notions, theories, methods, as well as the existing research problems, are still unclear. In this paper, after presenting the basic notions of argumentation, the recent development of argumentation-based non-monotonic reasoning of agents is analyzed. Finally, some challenging problems are discussed, and the possible future work is pointed out. 2012 Vol. 25 (4): 632-641 [Abstract] ( 658 ) [HTML 1KB] [ PDF 482KB] ( 760 ) 642 Anisotropic Diffusion Model Combined with Local Entropy ZHAO De, HE Chuan-Jiang, CHEN Qiang Perona-Malik (P-M) model is a classical anisotropic diffusion denoising model, but it is not able to preserve the important details effectively such as the texture of the image. To address this problem, an improved P-M model based on local entropy is proposed. The diffusion coefficient of the model not only depends on the image gradient, but also depends on the local region information described by local entropy. The experimental results show that the proposed model removes noises effectively, preserves the boundaries better, and maintains important details of the image well. 2012 Vol. 25 (4): 642-647 [Abstract] ( 447 ) [HTML 1KB] [ PDF 2358KB] ( 647 ) 648 Blind Super-Resolution Reconstruction Algorithm under Affine Motion Model ZHANG Xue-Song, JIANG Jing, PENG Si-Long An approach to the blind super-resolution (BSR) problem is proposed which yields a higher optical resolution image from a low-resolution (LR) image sequence with affine inter-frame motion. Firstly, an eigenvector-based method for constructing the null space of blurs is presented. It is used as the regularization constraint of the optimization procedure. Then, the iterative algorithm is developed for the triple-coupled problem. The proposed algorithm adopts a two-layer optimization strategy: in the first layer, the triple-coupled BSR problem is reduced to a quadratic form with respect to the blurs, and an nonlinear least squares (NLS) problem of the motion and the high-resolution (HR) image; in the second layer, the NLS problem is solved using a Gauss-Newton based method. The experimental results on synthetic data illustrate that the proposed BSR algorithm for affine transform, has better performance in terms of modeling the space-variant degradation process as well as restoring the local textures compared with the BSR algorithm for pure translation. Finally, the applicability of the proposed algorithm is demonstrated using real license plate images. 2012 Vol. 25 (4): 648-655 [Abstract] ( 484 ) [HTML 1KB] [ PDF 554KB] ( 684 ) 656 A Linear Reconstruction Method for Face Images under Normal Illumination XIONG Peng-Fei, LIU Chang-Ping, HUANG Lei Based on the stable relationships between the face representations under the certain and the normal illumination for different individuals, an approach to reconstruct face images under normal illumination is proposed. Firstly, to eliminate the impact of facial surfaces, an image deformation method in 3D domain is applied to achieve pixel-level alignment. Then, an illumination classification method based on image blocking is proposed to classify the images with the same lighting gradation. Finally, various linear reconstruction models of different illumination categories based on facial subspaces are trained from the preprocessing image pairs for face image reconstruction. The method effectively avoids the loss of the facial texture in image preprocessing and the distortion in image subspace. The experimental results of the proposed method on Extended Yale B demonstrate the performance in image representation and face recognition and verify the effectiveness in face alignment and illumination classification. 2012 Vol. 25 (4): 656-663 [Abstract] ( 708 ) [HTML 1KB] [ PDF 1621KB] ( 839 ) 664 The HMM-UBM Based Voice Print Password System ZHANG Zhao, GUO Wu, DAI Li-Rong A hidden markov model (HMM)-universal background model (UBM) algorithm for the voice print password system is proposed. Due to the sparseness of the enrollment data in the voice print password system, a mono-phone HMM-UBM is firstly trained by using the speaker-independent database. Then, the hypothesized speaker model is obtained by adapting the parameters of the UBM using the speaker’s training speech and the maximum a posteriori (MAP) estimation. The data sparseness problem is thus solved. The equal error rate (EER) of the proposed system is 6.8% on the IFIY-DESKTOP Ⅱ database. 2012 Vol. 25 (4): 664-668 [Abstract] ( 806 ) [HTML 1KB] [ PDF 393KB] ( 1028 ) 669 Dual-Modal Decision Fusion for Fingerprint and Finger Vein Recognition Based on Image Capture Quality Evaluation WANG Ke-Jun, MA Hui, GUAN Feng-Xun, LI Xue-Feng To overcome the influence of the image quality on the single modal recognition system, a fingerprint and finger vein dual-mode recognition method based on image quality evaluation is proposed. The image quality scores of the fingerprint are obtained by the proposed method. And according to its characteristics, the finger vein image quality assessment is designed for the first time to overcome the influence of image quality on identification results. Two classifiers are then designed for fingerprint and finger vein recognition respectively. The final recognition is achieved by the fusion of two quality scores and the recognition results of the classifiers at the decision level. The proposed method overcomes the limitations of image quality on single-modal biometrics, and effectively improves the recognition performance of the system. Thus, the validity of the method for multimodal biometric identification is demonstrated. 2012 Vol. 25 (4): 669-675 [Abstract] ( 845 ) [HTML 1KB] [ PDF 816KB] ( 837 ) 676 A Co-Evolutionary Algorithm for Clustering DONG Hong-Bin, YANG Bao-Di, LIU Jia-Yuan, HOU Wei A co-evolutionary algorithm for clustering is proposed. Firstly, the number of centers of clusters can be decided automatically with an improved mask code manner. The population is divided into two subpopulations which are constituted of the same size of individuals. The genetic algorithm is used in one subpopulation which is good at global search optimum ability, and the differential evolution algorithm is used in the other which has good local search ability to cluster. In the evolution process, different migration policies are utilized to exchange good individuals found by the two evolutionary algorithms between the twosubpopulations, which can balance the global and local search ability of the proposed algorithm. The experimental results show that the proposed method is effective through testing the number of the centers of clusters, performance and execution time on several datasets. 2012 Vol. 25 (4): 676-683 [Abstract] ( 730 ) [HTML 1KB] [ PDF 623KB] ( 797 ) 684 Compressive Sensing Based Neighborhood Embedding JIA Jiong, ZHENG Zhong-Long, YANG Jie How to construct local neighborhoods is one of the key points of spectral-manifold based algorithms. For example, locally linear embedding (LLE), one of the traditional manifold learning algorithms, constructs the local relationships through KNN or ε criterion. Motivated by compressive sensing theory, the strategy of neighborhood construction is proposed based on the linear combination of l2 and l1, which is called compressive sensing based neighborhood embedding (CSNE). The proposed strategy can not only be applied to LLE, but also to other spectral learning methods while neighborhoods need to be constructed. In addition, the semi-supervised CSNE algorithm is presented while the un-labeled data are taken into account. The results of visualization and classification experiments on several datasets demonstrates the competitive results of the proposed algorithm compared with PCA、LDA、LPP and S-Isomap. 2012 Vol. 25 (4): 684-690 [Abstract] ( 554 ) [HTML 1KB] [ PDF 1461KB] ( 645 ) 691 Path-Sensitive Multi-Variable Access Correlations Mining and Related Source Code Defects Detection YU Xiu-Mei, LIANG Bin, CHEN Hong, XIE Su-Bin, WANG Mei-Lin A large number of widespread source systems make the security of source code of software increasingly important. Multi-variable access correlation rules are mined by the path-sensitive method and the defects caused by inconsistent access to correlated variables are detected automatically in the large-scale source code systems. Multi-variable access correlations can be mined via logical information of source code and sensitive path information, which avoids fault from the insensitive path method. The effective solutions to the uneven distribution of path weight and the path explosion problems are presented. The presented method is verified by Linux system, and the experimental result shows that it mines correct multi-variable access correlations. 2012 Vol. 25 (4): 691-698 [Abstract] ( 572 ) [HTML 1KB] [ PDF 457KB] ( 803 ) 699 Sketch Face Recognition Based on Central Error Diffusion Local Binary Pattern DANG Li, KONG Fan-Rang The current research on sketch face recognition focuses on transformation between photos and sketches, which reduces the modality gap between features extracted from photos and sketches. An approach is proposed to reduce the modality gap at the feature extraction stage. A face encoding method based on central error diffusion local binary pattern is used to capture the same face modality and reduce the difference between photos and sketches. Under the background that sketch recognition is actually the problem of single sample, the sample amount is extended by using wavelet packet decomposition and central error diffusion local binary pattern. Then, PCA+LDA is used to extract features and recognize faces. The experimental results indicate that the proposed algorithm reduces the modality gap between photos and sketches obviously and it has a higher recognition rate and better performance than the methods based on pseudo-sketches synthesis. 2012 Vol. 25 (4): 699-704 [Abstract] ( 625 ) [HTML 1KB] [ PDF 996KB] ( 697 ) 705 Application Research on Complex Water Quality Prediction with Improved QGA-BP Model YU Li, WANG Jia-Quan Water quality prediction is a prerequisite for planning and managing of water environment and integrated controlling of water pollution. However, the construction of mechanism models is complex, the massive computation and data are required, prediction effects are not accurate enough and the further application of mechanism models is difficult. An improved QGA-BP model is constructed for predicting water quality of complex Huaihe river. The dynamic improvement strategy and catastrophe strategy are used as evolutionary operation guidelines in QGA to optimize the weight and the threshold of BP model. The past observation data are applied as the example to train the model. The comparison of experimental results show that the evolution generation, convergence speed and prediction precision of the improved QGA-BP model are improved. The model is applicable to solve the black box problem of water prediction and provides a new way for water environment management. 2012 Vol. 25 (4): 705-708 [Abstract] ( 527 ) [HTML 1KB] [ PDF 518KB] ( 522 ) 709 Variable Multigranulation Rough Set Model ZHANG Ming, TANG Zhen-Min, XU Wei-Yan, YANG Xi-Bei By analyzing the limitations of the optimistic multigranulation rough set and the pessimistic multigranulation rough set, the variable multigranulation rough set is proposed. The properties and measure relations of these kinds of rough sets are discussed. Furthermore, the acquisition of decision rules by the proposed variable multigranulation rough set is presented, and the heuristic algorithm of the attribute reduction based on the attribute significance is introduced. Finally, the experimental results show the effectiveness of the approach. 2012 Vol. 25 (4): 709-720 [Abstract] ( 380 ) [HTML 1KB] [ PDF 540KB] ( 706 )

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