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....

2013 Vol.26 Issue.1, Published 2013-01-30 Orignal Article        Orignal Article 1 Speaker Clustering of Telephone Speech Based on Front-End Factor Analysis WU Kui,SONG Yan,DAI Li-Rong The existing speaker clustering methods based on Gaussian mixture model (GMM) mainly obtain clusters′ GMMs by adapting from universal background model (UBM). However,this adaptive method suffers from the lack of data and results in poor models. In this paper,two factor analysis modeling methods based on eigenvoice (EV) space analysis and total variability (TV) space analysis respectively are explored. The two methods greatly reduce the number of estimated parameters when clusters′ GMMs are estimated by modeling variability space. The experimental results on two speakers telephone data in 2008 NIST Speaker Recognition Evaluation show that the two proposed methods achieve considerable reduction in speaker error rate compared to the baseline system using MAP adaptation,and the method based on TV space analysis obtains lower speaker error rate compared to the method based on EV space analysis. 2013 Vol. 26 (1): 1-5 [Abstract] ( 773 ) [HTML 0KB] [ PDF 335KB] ( 1116 ) 6 Sparse Constrained Reconstruction for Parallel Magnetic Resonance Image Based on Variable Splitting Method LIU Xiao-Fang,YE Xiu-Zi,ZHANG San-Yuan,LI Xia In order to reduce the aliasing artifacts and noise in the reconstructed images due to under-sampling data,a sparse constrained image reconstruction algorithm is proposed for parallel magnetic resonance imaging. In this paper,first-order difference is viewed as the sparse project operator,and a parallel magnetic resonance image reconstruction algorithm restrained by anisotropic total variation minimization is researched. Meanwhile,a solution based on variable splitting method is proposed,and the effectiveness and robustness of the proposed algorithm are analyzing in some specified experimental environments. The results show that the quality of reconstructed images is evidently improved for parallel magnetic resonance imaging by the proposed method at a maximum acceleration factor. 2013 Vol. 26 (1): 6-13 [Abstract] ( 589 ) [HTML 0KB] [ PDF 953KB] ( 816 ) 14 Equity Multi-Instance Face Identification Based on Threshold Control and Fusion Feature DENG Jian-Xun,XIONG Zhong-Yang,ZENG Dai-Min Image retrieval based on multi-instance learning (MIL) has great value in the field of regional image retrieval. The traditional voting mechanism in MIL is prone to misunderstanding,because the local similarity does not mean the overall similarity in face identification. Firstly,instance equity concepts of equity MIL are presented. Each kind of instance has different equity and the training set has the similar features to the test set. Therefore,the classification attribution for different packets can be obtained by the sum of results from multiplying every discriminant result and its instance equity. Secondly,the overall characteristic is considered as a special instance,and the overall sample equity threshold is used to control equity ratio. At the same time,the abnormal conditions,such as two persons have similar facial features,are prevented by means of the feature fusion. And the recognition rate is improved by the use of threshold control. The experimental results on the ORL and FERET show that the algorithm is feasible and the performance is superior to other algorithms. 2013 Vol. 26 (1): 14-19 [Abstract] ( 535 ) [HTML 0KB] [ PDF 386KB] ( 741 ) 20 Double-Ring Mean Shift Algorithm for Visual Tracking XIA Yu,WU Xiao-Jun,WANG Hong-Yuan A tracking algorithm based on double-ring Mean Shift is proposed to solve the deficiency of target representation,template similarity measure and fixed kernel-bandwidth in traditional Mean Shift tracking algorithm. The feature extraction model based on universal elliptical region is used to reduce the influence of background feature and improve the quality of target model effectively. Double-ring descriptor is presented to emphasize the importance of target feature and improve the peak modality of matching function. The proposed method updates the bandwidth of kernel-function adaptively by the relationship of double-ring. The experimental results show that the proposed tracking approach is robust and invariant to scale,pose and partial occlusions. 2013 Vol. 26 (1): 20-27 [Abstract] ( 377 ) [HTML 0KB] [ PDF 1169KB] ( 703 ) 28 Structural Classification of Protein Domain Using Gabor Filter and Radon-Legendre Moment SHI Jian-Yu,ZHANG Yan-Ning Representing structural classification as image classification,an effective method of structural classification of protein domain is proposed. Firstly,the spatial structure of protein domain is mapped to its distance matrix which is regarded further as gray texture image. As a result,the secondary structure elements (SSE) and the topology of domain are transformed to local geometric structures with variant scales,orientations and the local-structure-composed shape in such image respectively. Then,Gabor filters are designed to segment these local structures out and extract the percentage feature which represents the composition of SSE. After that,Radon-Legendre moment is presented to characterize the local-structure-composed shape and is used as feature of the shape. Finally,the composition feature and the moment feature are combined to perform structural domain classification. The experimental results show that the proposed method achieves effective classification of protein domain and outperforms other methods in both classification accuracy and robustness of sample count. 2013 Vol. 26 (1): 28-33 [Abstract] ( 378 ) [HTML 0KB] [ PDF 487KB] ( 735 ) 34 Selection of Clustering Algorithms Based on Grid-MST LI Xiang-Yu,WANG Kai-Jun,GUO Gong-De To get better clustering results,it is necessary to choose a suitable clustering algorithm for the cluster structure of a given dataset. Selection of clustering algorithms based on Grid-MST is proposed to choose a suitable clustering algorithm for the data set automatically. The Grid-MST is constructed on the basis of the dataset by the proposed method,and the potential cluster structures are found by the number of trees. Then,a suitable clustering algorithm is selected to the discovered cluster structure. The experimental results on artificial datasets and real datasets show that the proposed method is efficient. 2013 Vol. 26 (1): 34-41 [Abstract] ( 443 ) [HTML 0KB] [ PDF 779KB] ( 831 ) 42 Efficient Incremental Updating Algorithm for Core AttributeBased on Information Entropy QIAN Wen-Bin,YANG Bing-Ru,XU Zhang-Yan,ZHANG Chang-Sheng Since the efficiency of the algorithms for core attribute based on information entropy is not well,the binary discernibility matrix from information view is defined. And it is proved theoretically that the core attribute based on the binary discernibility matrix is equivalent to that based on information entropy. The objects in the decision table are categorized into a consistent set and an inconsistent set,which effectively reduces the search space of algorithm for core attribute. Additionally,for dynamic decision table,the incremental updating mechanism for core attribute is discussed. Based on the mechanism,an efficient incremental updating algorithm for core attribute based on information entropy is proposed. The example analysis and experimental results show that the proposed algorithm outperforms other similar algorithms. 2013 Vol. 26 (1): 42-49 [Abstract] ( 644 ) [HTML 0KB] [ PDF 513KB] ( 634 ) 50 A Sub-Pattern Gabor Features Fusion Method for Single Sample Face Recognition WANG Ke-Jun,ZOU Guo-Feng To overcome the limitations of traditional face recognition methods for single sample face recognition,a sub-pattern Gabor features fusion method for single sample face recognition is proposed. Firstly,facial local features are extracted by Gabor wavelet transformation. Then,the Gabor face images are blocked to take full advantage of the spatial location information of facial organs,and the minimum distance classifiers are used for each sub-pattern. Finally,the recognition result is achieved by the fusion of the sub-pattern classifiers′ results at the decision level. According to the difference of sub-pattern construction and fusion method,two kinds of sub-pattern Gabor features integration programs are proposed. The experimental results and comparative analysis on ORL face database and CAS-PEAL-R1 face database show that the proposed method achieves better classification rate and improves the performance of single sample face recognition system. 2013 Vol. 26 (1): 50-56 [Abstract] ( 613 ) [HTML 0KB] [ PDF 436KB] ( 1182 ) 57 Soccer Robot Self-Localization by Combining Odometry and Visual Information LIU Zhan,DOU Hao-Bin,LUO Ding-Sheng A method is proposed which combines odometry and visual information for robot self-localization in soccer field. The method makes effective use of these two kinds of information by considering their characteristics simultaneously. On the one hand,odometry is used to effectively deal with the ambiguity which is prone to appear in landmark based visual localization. And on the other hand,the disambiguated visual localization results are useful to dynamically correct the accumulative odometry errors which are caused by robot motion. Finally,experiments are conducted on the Webots simulation platform and the results show the effectiveness of the proposed method. 2013 Vol. 26 (1): 57-62 [Abstract] ( 578 ) [HTML 0KB] [ PDF 541KB] ( 828 ) 63 Anti-Noise Iris Matching Method Based on Offset Hamming Distance Deviation CHENG Yu-Qi,FANG Wei,GE Wei The traditional iris matching method based on Hamming distance is studied,and the anti-noise iris matching method based on offset Hamming distance deviation (OHDD) is proposed. Firstly,the odd symmetry Gabor filters with single frequency and 2 directions are used to extract iris edge features. Then,the filtering results are encoded by zero-crossing detecting method. Finally,the OHDD parameter is constructed for iris matching. Comparison experiments of the traditional matching method and the proposed OHDD matching method are made on 6 iris datasets. The experimental results show that the equal error rate and the correct recognition rate of the OHDD matching method are better than those of the traditional matching methods consistently in all iris datasets and the OHDD matching method has strong anti eyelid and eyelash noise ability. 2013 Vol. 26 (1): 63-69 [Abstract] ( 547 ) [HTML 0KB] [ PDF 762KB] ( 791 ) 70 Spatial Relationship Representation of Objects in Images and Its Application to Image Retrieval YANG Tong-Feng,MA Jun For the image retrieval system based on spatial relationship of objects in images,it is hard to automatically recognize objects and their spatial relations correctly. Based on the outputs of object detection algorithms,a triple representation of the spatial relationship in images is proposed. Based on the representation,a method for indexing images,computing similarities and ranking results is proposed. A 2D user-match interface is also developed for users to express their needs in terms of retrieval keywords and spatial relationships,and a prototype is established. The representation is robust against errors of object detection. Incorporating the confidence given by object detection into the triple representation and ranking method,the impact of object detection errors on the performance of image retrieval is reduced. With the queries comprising explicit spatial relationship,the proposed approach gives more accurate results in experiments. It performs better than the existing systems in terms of NDCG@m,MAP and F@m. 2013 Vol. 26 (1): 70-75 [Abstract] ( 388 ) [HTML 0KB] [ PDF 462KB] ( 987 ) 76 Moving Horizon Estimation of Ego-Motion in Monocular Visual Systems YANG Dong-Fang,SUN Fu-Chun,WANG Shi-Cheng Estimating ego-motion in monocular visual systems from the input image sequence is a critical problem in computer vision. A moving horizon estimation (MHE) based algorithm is proposed to solve the pose estimation problem in most general application environment including buildings and trees. Firstly,different forms of epipolar constraints are analyzed. The time-space related constraints among the closed loop of image sequence are all involved in the global optimization model. In addition,the MHE is adopted to obtain the tradeoff between computation costs and estimation accuracy. Based on the general epipolar equations,the redundant epipolar constraints and the moving horizon constraints,the corresponding three referred pose estimation algorithms are performed comparatively,and the outdoor experimental results validate the effectiveness of the proposed method. 2013 Vol. 26 (1): 76-82 [Abstract] ( 515 ) [HTML 0KB] [ PDF 637KB] ( 953 ) 83 An Ensemble Detection Method of Pipeline Condition Based on Tabu Search WANG Yong-Xiong ,SU Jian-Bo To improve the recognition rate of pipe anomaly detection and real-time performance,an ensemble classification method based on Tabu search is proposed which combines semi-supervise K-means clustering and C4.5 decision tree. The cost-sensitive function is introduced in Tabu search to select the most discriminating feature subset and the best ensemble weights. Thus,the classification performance of the minority class in imbalance data is improved. The semi-supervise K-means approach partitions the features of samples into k clusters firstly. Then,a supervised C4.5 decision tree in each K-means cluster is trained to refine the decision boundaries by learning the subgroups within the cluster. The ensemble classification by cascading K-means and C4.5 alleviates the problems of imbalance data and improves the classification accuracy of imbalance data. The final decisions of the K-means and C4.5 methods are integrated based on the weighted sum rule,the nearest-neighbor rule,and the nearest consensus rule respectively. The experimental results show that the proposed system is effective in classifying imbalance data and has high performance in detecting the anomaly of pipeline. 2013 Vol. 26 (1): 83-89 [Abstract] ( 468 ) [HTML 0KB] [ PDF 554KB] ( 728 ) 90 An Improved Color Cooccurrence Matrix Texture Descriptor XU Shao-Ping,LI Chun-Quan,HU Ling-Yan,YANG Xiao-Hui,JIANG Shun-Liang Utilizing similarity measure defined by the fuzzy continuous t-norm operator to describe the degree of difference between pixels in color images,an improved color cooccurrence matrix texture descriptor is proposed. In accordance with the predefined interval of distances and directions,the multi-channel color information of original color image is effectively integrated and converted to pseudo-gray images. Then,gray level cooccurrence matrix (GLCM) texture method is used to extract feature vector for the pseudo-gray image. A large number of experiments are performed on the content-based image retrieval prototype platform and the results show that compared with other types of texture descriptors,the improved texture descriptor has the same feature vector dimension as the GLCM descriptor,while its description ability matches with all kinds of color cooccurrence matrix descriptor. The improved texture descriptor effectively achieves the integration of texture and color characteristics and improves the image retrieval performance. 2013 Vol. 26 (1): 90-98 [Abstract] ( 646 ) [HTML 0KB] [ PDF 1246KB] ( 1044 ) 99 Video Segmentation Algorithm Based on Homomorphic Filtering Inhibiting Illumination Changes ZHANG Xiao-Yu,HU Shi-Qiang The backgrounds can not be updated effectively by color difference histogram based video segmentation algorithm as major illumination changes. Therefore,the subsequent foreground can not be segmented from input images effectively. For the problem stated above,a video segmentation algorithm based on homomorphic filtering inhibiting illumination changes is proposed. Firstly,homomorphic filtering is used to rectify luminance component about both input and background images (RGB) in HSV space with the same parameters. Then,the rectified images are converted into RGB color space. Finally,the color difference histogram algorithm is used to segment the video. The proposed algorithm effectively solves the problem that the color difference histogram algorithm can not update the areas,in which illumination changes are large,into background. Hence,the proposed algorithm updates background in time and effectively and segments foreground robustly from subsequent input images. The simulation results of three sequences demonstrate that the proposed algorithm has a faster calculation speed and deals with illumination changes more robustly compared with Guassian mixture and Codebook algorithm. 2013 Vol. 26 (1): 99-105 [Abstract] ( 548 ) [HTML 0KB] [ PDF 903KB] ( 671 ) 106 A Feature Relevance Measure Based on Sparse Representation Coefficient GENG Yao-Jun,ZHANG Jun-Ying,YUAN Xi-Guo The feature relevance measures employed by current feature selection methods can effectively evaluate the relevance between two features,but they do not consider the influence of the other features on them. On the premise of considering feature interaction overall,sparse representation coefficient is proposed as a feature relevance measure. The difference between the proposed method and the existing relevance measures is that it reveals the relevance between feature and target under the influence of the other features,which reflects feature interaction. In order to verify the effectiveness of sparse representation coefficient to measure relevance of feature,the classification performance is compared among feature subsets selected by Relief F and the feature selection methods using sparse coefficient,symmetrical uncertainty and Pearson correlation coefficient as relevance measures respectively. The experimental results show that the classification performance of the features selected by the proposed method is higher and more stable. 2013 Vol. 26 (1): 106-113 [Abstract] ( 377 ) [HTML 0KB] [ PDF 498KB] ( 1052 ) 114 Chinese Geographic Entity Resolution Based on Markov Logic Network HU Yi-Min,SONG Liang-Tu ,CHEN Peng,WEI Yuan-Yuan ,SU Ya-Ru Markov Logic Network has the ability to handling the complex representation and the uncertainty of first-order logic and probabilistic graphical models. An entity resolution method based on Markov logic network and property extraction algorithm employing ontology and web search is proposed to improve the performance of named entity resolution for unstructured data based on Markov logic network. The method is then applied to the resolution of Chinese geographic names. The experimental result shows that the proposed method is effective in geographic entity resolution. 2013 Vol. 26 (1): 114-122 [Abstract] ( 404 ) [HTML 0KB] [ PDF 462KB] ( 1164 ) 123 SIFT Feature Matching Algorithm Based on Vector Angle WU Wei-Jiao,WANG Min,HUANG Xin-Han,MAO Shang-Qin An approximate nearest neighbor search method based on vector angle is proposed. Firstly,the vector angles between high dimensional vectors and a stochastic selected reference vector are computed,and these angles are sorted. Then,the angle of reference vector and the query vector is computed,and the angle is found in the sorted angles by binary search algorithm. Finally,taking the angle as the center,the approximate nearest neighbor of the query vector is searched in the setting range. The experimental results show that the scale invariant feature transform feature matching can be accelerated significantly without undermining the performance of feature matching. 2013 Vol. 26 (1): 123-128 [Abstract] ( 588 ) [HTML 0KB] [ PDF 373KB] ( 1356 )

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