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....

2011 Vol.24 Issue.2, Published 2011-04-30 Orignal Article        Articles 153 Kernel Based Slow Feature Analysis MA Kui-Jun, HAN Yan-Jun, TAO Qing, WANG Jue A kernelbased algorithm is proposed to solve the nonlinear exparsion problem of slow feature analysis (SFA) is proposed to solve this problem. By using the kernel trick, it avoids the difficulties of computing directly in high dimensional space. Because of the full use of nonlinear information of the data, its output is steady. Meanwhile, based on analysis of the objective of the algorithm, a formula is put forward to estimate the output slowness of the signal and utilize it as a guide line to choose parameters of the kernel functions. Experimental results show the effectiveness of the proposed algorithm. 2011 Vol. 24 (2): 153-159 [Abstract] ( 1089 ) [HTML 1KB] [ PDF 482KB] ( 1031 ) 160 An Emotion Cognitive Appraisal Model for Virtual Characters LIU Zhen, HE Shao-Hua, CHAI Yan-Jie Virtual characters are widely used in eeducation, cartoon games and ecommerce. However, the current emotion design for virtual characters is still in a handwork stage and needs a large amount of manpower. The behaviors of most virtual characters are simple and unbelievable. Based on psychology theory and ortong, clore, collins (OCC) model, the architecture of a virtual character is presented and a formalization of a virtual character's motivation is proposed. A virtual character's emotion state is driven by a set of fuzzy productive rules for emotion decision. Emotion intensities are calculated by a fuzzy inference system. The experimental result shows that the emotion model enhances the efficacy of friend interfaces for education software，and makes a virtual character to behave more humanlike. 2011 Vol. 24 (2): 160-167 [Abstract] ( 800 ) [HTML 1KB] [ PDF 562KB] ( 1202 ) 168 3D Facial Expressional Motion Tracking Algorithm Based on Online Model Adaptation and Updating YU Jun, WANG Zeng-Fu A 3D facial expressional motion tracking algorithm based on online model adaptation and updating is proposed. The algorithm constructs the online model using an adaptive statistic observation model. With the combination of adaptive state transition model and improved particle filter, statistic search and determinately search are proposed simultaneously. Multi-measurements are infused to decrease lighting influence and person dependence. With this algorithm, the global rigid motion parameters and local non rigid expressional parameters are obtained. Experiment result confirms the effective ness of the proposed algorithm. 2011 Vol. 24 (2): 168-175 [Abstract] ( 751 ) [HTML 1KB] [ PDF 633KB] ( 762 ) 176 Approach to Solving Computing the Attribute Reductions with Ant Colony Optimization YU Hong, YANG Da-Chun Attribute reduction is an important process in rough set theory. Minimal attribute reductions are expected to help clients make better decisions in some cases. In this paper, a heuristic approach for solving the minimal attribute reduction problem (MARP) is proposed based on the ant colony optimization (ACO) metaheuristic. Firstly, the MARP is trasformed into an assignment which minimizes the cost in a constraint satisfaction model. Then, a preprocessing step is introduced that removes the redundant data in a discernibility matrix through the absorbtion operator to favor a smaller exploration of the search space at a lower cost. Next, an algorithm, RACO, is developed to solve the MARP. Finally, the simulation results show that the proposed approach finds more minimal attribute reductions efficiently in most cases. 2011 Vol. 24 (2): 176-184 [Abstract] ( 659 ) [HTML 1KB] [ PDF 531KB] ( 803 ) 185 Solving TSP Problems with Estimation of Distribution Algorithm based on Superiority Pattern Junction HE Xiao-Juan, ZENG Jian-Chao An Estimation of Distribution Algorithm for TSP problems based on superiority pattern junction is proposed. The pairwise adjacent pattern matrix is constructed, then the junction blocks are built combining with superiority individual information. Each block is adjusted as a whole to avoid repeating search. Therefore, the disruption of superiority building blocks is solved and the search speed is improved. At the same time, the patterns within each block is made local adjustment under special conditions to enhance the local search ability. The simulation results show that the proposed algorithm has better efficiency in solving the TSP problems. 2011 Vol. 24 (2): 185-193 [Abstract] ( 726 ) [HTML 1KB] [ PDF 535KB] ( 725 ) 194 An Ontology ConceptBased Cluster Partition Approach for Computing the Semantic Distance between Concepts PENG Zhi-Ping, LI Xiao-Ming, KE Wen-De The semantic similarity computing between concepts is an important component in natural language processing etc., and the semantic similarity computing between concepts based on semantic distance is currently dominant technique. In this paper, the ontology based cluster partition approach for computing the semantic distance between concepts is proposed on the basis of the analysis of the lacks in the existing algorithms. The rules for computing the semantic distance between concepts are given under the situation of multiconcept clusters, and then the approach for computing the semantic distance between concepts within single cluster as well as crosscluster is put forward. In the proposed approach, the nonsymmetry of semantic similarities in the pairs of hyponymy concepts is worked out by introducing the forward semantic distance and the reverse semantic distance, and the other binary relationships of the pairs of nonhyponymy concepts are deal with by dynamically allocating the relation weights in the light of the locations of concept nodes. Experimented results shows that the proposed approach is effective and it is preferable to other typical similar ones. Guide:null 2011 Vol. 24 (2): 194-200 [Abstract] ( 732 ) [HTML 1KB] [ PDF 426KB] ( 642 ) 201 Research on Increasing the Performance of Evolutionary Algorithm in Searching Robust Optimal Solutions Based on Quasi Monte Carlo Method ZHU Yun-Fei, LUO Biao, ZHENG Jin-Hua, CAI Zi-Xing Robust optimal solution is of great significance in engineering application. It is one of the most important and difficult topics in evolutionary computation. Monte Carlo Integral (MCI) is generally used to approximate effective objective function (EOF) in searching robust optimal solution with evolutionary algorithm (EA). However, due to the low accuracy in existing crude Monte Carlo (C-MC) method, the performance of searching robust optimal solution with EA is unsatisfactory. Therefore, a QuasiMonte Carlo (Q-MC) method is proposed to estimate EOF. The experimental results demonstrate that the proposed Q-MC methods, SQRT sequence, SOBOL sequence and Korobov Lattice approximate EOF more precisely compared with CMC method, and consequently, the performance of searching robust optimal solution with EA is improved. Guide:null 2011 Vol. 24 (2): 201-214 [Abstract] ( 649 ) [HTML 1KB] [ PDF 493KB] ( 575 ) 210 A Covering Incremental Reduction Algorithm Based on Absolute Information Quantity LIN Guo-Ping, LI Jin-Jin A covering reduction algorithm is studied under condition that both the upper approximation operator and the under approximation operator are not changed. The absolute information quantity and information quantity are defined, and then the adjacency matrix is presented. An incremental reduction algorithm is presented for covering generalized rough sets which based on the absolute information quantity. The example shows that the proposed algorithm method is an effective technique to remove the redundant knowledge in the complex data sets. 2011 Vol. 24 (2): 210-214 [Abstract] ( 620 ) [HTML 1KB] [ PDF 268KB] ( 746 ) 215 Research and Development of Cognitive Computing in Mind WANG Zhi-Liang, ZHENG Si-Yi, WANG Xian-Mei, WANG Wei Cognitive computing in mind is an important part of intelligent humancomputer interaction (HCI), which attracts increasing attention from researchers in recent years. The development of the research advances in home and abroad is surveyed. Firstly, the conceptions correlated to cognitive computing in mind are introduced, and then the mechanisms and research contents in Mindreading are described in detail. Secondly, main neurobiological achievements of mental cognition are summarized and the research status of cognitive state in mind is compared to affective state. Moreover, the application trends of mental cognition in HCI are analyzed from two aspects of model establishment and mode extraction. Then, the primary frames that vision cognitive computing model of mental state with multilevel and multimode information fusion is put forward. The existing problems and the significance of cognitive computing in mind are firally discussed. 2011 Vol. 24 (2): 215-225 [Abstract] ( 849 ) [HTML 1KB] [ PDF 849KB] ( 1276 ) 226 Multi Feature Fusion Based Circular Traffic Sign Detection ZHANG Jing, HE Ming-Yi, DAI Yu-Chao, QU Xiao-Gang A multi feature fusion based circular traffic sign detection algorithm is proposed. Firstly, color segmentation and achromatic decomposition are used to extract area containing circular traffic signs based on color feature of circular traffic sign, and thus most of the background area is eliminated. Then, edge detection is applied, and the output edges are stored by chain code. Features, including length of contour circularity and aspect ratio are used to further eliminate background area. Finally, according to the shape feature of circular traffic sign, nonlinear least squares method is employed to detect the exact traffic signs. Effectiveness of the algorithm is demonstrated preliminarily on real images captured under various weather and illumination conditions. 2011 Vol. 24 (2): 226-232 [Abstract] ( 1033 ) [HTML 1KB] [ PDF 450KB] ( 700 ) 233 Multi Criteria Recommendation Algorithm Based on Widrow-Hoff Neural Network ZHANG Fu-Zhi, CHANG Jun-Feng, WANG Dong To solve the problem that the traditional collaborative filtering recommendation algorithm can not recommend with multiple criteria, a multicriteria recommendation algorithm based on Widrow-Hoff neural network is proposed by introducing the concept of multicriteria rating for extending the standard collaborative filtering algorithm. The Widrow-Hoff least mean square adaptive algorithm has the characteristics of hith accuracy fitting in the process of system identification. Based on that, an approach to compute user preferences eigenvector based on Widrow-Hoff LMS algorithm is proposed. Measuring users' similarity by adopting the user preferences eigenvector and spatial distance matrix so as to locate a neighbor set for the best recommendations. Experimental results show that the proposed algorithm improves the accuracy and the quality of recommendation. 2011 Vol. 24 (2): 233-242 [Abstract] ( 636 ) [HTML 1KB] [ PDF 638KB] ( 1146 ) 243 Ontology Mapping Method Based on Ontology Partition LI Zhi-Ming, LI Shan-Ping, YANG Chao-Hui, LIN Xin The mapping efficiency is key to the performane of dynamic ontology mapping in Semantic Web Service descovery, contextawareness in smart spaces and so on. The existing methods simplify the current methods of similarity computation to promotes the efficiency, nevertheless they fail in the case that the number of candidate mapping entity pairs increases when ontology gets larger. An efficient ontology mapping method based on ontology partition is proposed, which divides an ontology into a set of blocks through bottomup clustering. Then the blocks and candidate mapping entity pairs are mapped and selected from the block mapping result. The experimental results show that the proposed method promotes the efficiency of mapping significantly with 6 times faster than Falcon-AO. 2011 Vol. 24 (2): 243-248 [Abstract] ( 540 ) [HTML 1KB] [ PDF 356KB] ( 673 ) 249 Modified Katz Approach and Its Application in Speech Recognition Based on Lattice ZHANG Lei, LU Dong, XIANG Xue-Zhi In traditional Katz approach, the discount coefficients may be greater than 1, or can not be calculated in some serious conditions. To avoid the above problem, the smoothing idea in log domain of Simple GoodTuring combined with backoff model is proposed in the modified Katz approach. The proposed approach is further applied in speech recognition system based on lattice. The analysis of the effects on the structure and performance of lattice with different language model is given. Experiments show that compared with traditional Katz approach, the modified Katz approach can enhance the system performance. The best recognition rate can achieve 60.90% for the corpus from interview program. 2011 Vol. 24 (2): 249-254 [Abstract] ( 566 ) [HTML 1KB] [ PDF 351KB] ( 668 ) 255 Tensor Completion Algorithm and Its Applications in Face Recognition SHI Jia-Rong, JIAO Li-Cheng, SHANG Fan-Hua Missing data problems are commonly attributed to the matrix completion problem, and matrix completion is an important method of signal acquisitions following compressing sensing. The data examples have the property of multilinearity in applications, that is, the data set can be represented by higher order tensors. The tensor completion problem and its applications in face recognition are studied. Based on lowerdimensional Tucker decomposition of tensors, an iterative algorithm is proposed to complete tensors. And the distance between the estimating tensor and its Tucker approximation tensor is monotonically decreasing during the iterative procedure. Experimental results demonstrate the effectiveness and feasibility of the proposed method in completing tensor and face recognition. 2011 Vol. 24 (2): 255-261 [Abstract] ( 870 ) [HTML 1KB] [ PDF 386KB] ( 1106 ) 262 A Deep Web Query Interface Matching Approach Based on Evidence Theory and Task Assignment DONG Yong-Quan, LI Qing-Zhong, DING Yan-Hui, ZHANG Yong-Xin To solve the limitations of existing query interface matching which have the difficulties of weight setting of the matcher and the absence of the efficient processing of matching decision, a deep web query interface matching approach based on evidence theory and task assignment called Evidence Theory and Task Assignment based Query Interface Matching Approach(ETTA-IM) is proposed. Firstly, an improved DS evidence theory is used to automatically combine multiple matchers. In this way, the weight of each matcher is not required to be set by hand and human involvement is reduced. Then, a method is used to select a proper attribute correspondence of each source attribute from target query interface, which converts one to one matching decision to the extended task assignment problem. Finally, based on one to one matching results, some heuristic rules of tree structure are used to perform onetomany matching decision. Experimental results show that ETTA-IM approach has high precision and recall measure. Guide:null 2011 Vol. 24 (2): 262-271 [Abstract] ( 839 ) [HTML 1KB] [ PDF 629KB] ( 671 ) 272 A Topic Oriented Photo Browsing Method in Flickr Group ZHENG Nan, LI Qiu-Dan With the development of Web 2.0, social tagging systems, such as Flickr, become more and more popular. As a new developed user community, Flickr group not only attracts tens of thousands of online users to participate, but also becomes one of the research hotspot in recent years. In this paper, a topicoriented photo browsing method is proposed to deal with the photo browsing problems in Flickr groups which lack of clear themes. The proposed method consists of three main steps: matrix space representation of phototag relations, nonnegative matrix factorization based photo topic finding and a computational model of photo influence strength. The experiments on Flickr dataset show the proposed method is useful and effective. Guide:null 2011 Vol. 24 (2): 272-276 [Abstract] ( 614 ) [HTML 1KB] [ PDF 349KB] ( 791 ) 277 Image Segmentation Algorithm Using Generalized Integrated Squared  Error Based Eigenvector Selection ZHANG Da-Ming, FU Mao-Sheng, LUO Bin Not all of the top eigenvectors contain clustering information for the task of realworld data clustering. Since the noise exists, the distribution of elements of an eigenvector is complex and it is necessary to select eigenvectors for spectral clustering. In this paper, the integrated squared error (ISE) divergence is generalized and the proposed generalized integrated squared error (GISE) is used to estimate the multimodality of data distribution and measure the clustering information of eigenvector. Then, a spectral clustering algorithm based on eigenvector selection is proposed. The experimental results on varied natural images segmentation show that the proposed algorithm is simpler and more effective than pervious algorithms. Guide:null 2011 Vol. 24 (2): 277-283 [Abstract] ( 682 ) [HTML 1KB] [ PDF 450KB] ( 783 ) 284 Online Signature Verification System Based on Support Vector Data Description ZOU Jie, WU Zhong-Cheng An novel online signature verification system is proposed based on support vector data description (SVDD). Firstly, correspondences of the critical points in signatures are confirmed by bidirectional backwardmerging dynamic time wrapping algorithm. Then, subtle differences in the local are calculated by classical dynamic time wrapping algorithm. Feature selection principle based on mean and deviation minimization is proposed. Finally, the classifiers are designed using support vector data description (SVDD). To obtain better result, m-fold-cross validation and genetic algorithm are used to seek optimal parameters of SVDD. The average equal error rate for skill forge signatures in SVC2004 signatures database is 4.25%. 2011 Vol. 24 (2): 284-290 [Abstract] ( 549 ) [HTML 1KB] [ PDF 434KB] ( 653 ) 291 A Randomized Corner Detection Algorithm 吕Na , FENG Zu-Ren There is no parametric formulation of corner feature. Therefore, the conventional Hough transform can not be employed to transform the corner detection into maximum search in parametric space. A randomized Hough transform in Monte Carlo framework is presented, which detects the corner by searching for the local maximum in the intersection point cumulative space instead of parametric space. The proposed intersection point cumulative space is a concept based on the fact that the corner is the intersection point of two lines. The algorithm is demonstrated and the computing procedures are given. The algorithm is isotropic, robust to image rotation, insensitive to noise and not susceptible to diagonal edge. Experimental results show that the proposed algorithm outperforms Harris detector, Shen & Wang algorithm, and SIFT feature detection algorithm. Guide:null 2011 Vol. 24 (2): 291-298 [Abstract] ( 516 ) [HTML 1KB] [ PDF 1064KB] ( 767 ) 299 Hand Gesture Recognition Based on Online PCA with  Adaptive Subspace YAO Ming-Hai, QU Xin-Yu The learning method for hand gesture recognition system based on vision is commonly offline, which results in repeated offline learning when new hand gestures come. Its realtime performance, expansibility and robustness are poor. In this paper, a method named online PCA with adaptive subspace is proposed for hand gesture recognition. The subspace is updated online by calculate PCA of sample coefficients. The subspace updating strategy is adjusted according to the difference degree between new sample and learned sample. The algorithm is able to adapt different situations and reduce the cost of oalculation and storage. The incrementally online learming and recognition of hard gestures are realized by the proposed algorithm. Experimental results show that the proposed method solves the unknown hand gesture problem, realizes online hand gesture accumulation and updating and improves the recognition performance of system. 2011 Vol. 24 (2): 299-301 [Abstract] ( 617 ) [HTML 1KB] [ PDF 394KB] ( 633 ) Orignal Article 153 Kernel-Based Slow Feature Analysis MA Kui-Jun, HAN Yan-Jun, TAO Qing, WANG Jue A kernel-based algorithm is proposed to solve the nonlinear expansion problem of slow feature analysis (SFA). By using the kernel trick, the difficulties of computing directly in high dimensional space are avoided. Because of the full use of nonlinear information of the data, its output is steady. Meanwhile, based on the objective analysis of the proposed algorithm, a formula is put forward to estimate the output slowness of the signal and it is utilized as a guide line to select parameters of the kernel functions. Experimental results show the effectiveness of the proposed algorithm. 2011 Vol. 24 (2): 153-159 [Abstract] ( 1222 ) [HTML 1KB] [ PDF 482KB] ( 1130 ) 160 An Emotion Cognitive Appraisal Model for Virtual Characters LIU Zhen, He Shao-Hua, Chai Yan-Jie Virtual characters are widely used in e-education, cartoon games and e-commerce. However, the current emotion design for virtual characters is still in a handwork stage and needs a large amount of manpower. The behaviors of most virtual characters are simple and unbelievable. Based on psychology theory and ortony, clore, collins (OCC) model, the architecture of a virtual character is presented and a formalization of a virtual characters motivation is proposed. A virtual characters emotion state is driven by a set of fuzzy productive rules for emotion decision. Emotion intensities are calculated by a fuzzy inference system. The experimental result shows that the emotion model enhances the efficacy of friend interfaces for education software,and makes a virtual character behave more humanlike. 2011 Vol. 24 (2): 160-167 [Abstract] ( 772 ) [HTML 1KB] [ PDF 562KB] ( 800 ) 168 3D Facial Expressional Motion Tracking Algorithm Based on Online Model Adaptation and Updating YU Jun, WANG Zeng-Fu A 3D facial expressional motion tracking algorithm based on online model adaptation and updating is proposed. It constructs the online model using an adaptive statistic observation model. With the combination of adaptive state transition model and improved particle filter, statistic search and determinate search are proposed simultaneously. Multi-measurements are infused to decrease lighting influence and person dependence. With the proposed algorithm, the global rigid motion parameters and local non rigid expressional parameters are obtained. Experiment result confirms the effectiveness of the proposed algorithm. 2011 Vol. 24 (2): 168-175 [Abstract] ( 691 ) [HTML 1KB] [ PDF 633KB] ( 878 ) 176 Approach to Solving Attribute Reductions with Ant Colony Optimization YU Hong, YANG Da-Chun Attribute reduction is an important process in rough set theory. Minimal attribute reductions are expected to help clients make better decisions in some cases. In this paper, a heuristic approach for solving the minimal attribute reduction problem (MARP) is proposed based on the ant colony optimization (ACO) metaheuristic. Firstly, the MARP is transformed into an assignment which minimizes the cost in a constraint satisfaction model. Then, a preprocessing step is introduced that removes the redundant data in a discernibility matrix through the absorption operator to favor a smaller exploration of the search space at a lower cost. Next, an algorithm, R-ACO, is developed to solve the MARP. Finally, the simulation results show that the proposed approach finds more minimal attribute reductions efficiently in most cases. 2011 Vol. 24 (2): 176-184 [Abstract] ( 633 ) [HTML 1KB] [ PDF 531KB] ( 737 ) 185 Solving TSP Problems with Estimation of Distribution Algorithm Based on Superiority Pattern Junction HE Xiao-Juan, ZENG Jian-Chao An estimation of distribution algorithm for TSP problems based on superiority pattern junction is proposed. The pairwise adjacent pattern matrix is constructed, then the junction blocks are built combining with superiority individual information. Each block is adjusted as a whole to avoid repeating search. Therefore, the disruption of superiority building blocks is solved and the search speed is improved. Meanwhile, the patterns within each block is made local adjustment under special conditions to enhance the local search ability. The simulation results show that the proposed algorithm has better efficiency in solving the TSP problems. 2011 Vol. 24 (2): 185-193 [Abstract] ( 757 ) [HTML 1KB] [ PDF 535KB] ( 653 ) 194 An Ontology Concept-Based Cluster Partition Approach for Computing the Semantic Distance between Concepts PENG Zhi-Ping, LI Xiao-Ming, KE Wen-De The semantic similarity computing between concepts is an important component in natural language processing etc., and the semantic similarity computing between concepts based on semantic distance is currently dominant technique. In this paper, the ontology based cluster partition approach for computing the semantic distance between concepts is proposed on the basis of the analysis of the lacks in the existing algorithms. The rules for computing the semantic distance between concepts are given under the situation of multi-concept clusters, and then the approach for computing the semantic distance between concepts within single cluster as well as cross-cluster is put forward. In the proposed approach, the non-symmetry of semantic similarities in the pairs of hyponymy concepts is worked out by introducing the forward semantic distance and the reverse semantic distance, and the other binary relationships of the pairs of non-hyponymy concepts are deal with by dynamically allocating the relation weights in the light of the locations of concept nodes. Experimented results shows that the proposed approach is effective and it is preferable to other typical similar ones. 2011 Vol. 24 (2): 194-200 [Abstract] ( 735 ) [HTML 1KB] [ PDF 426KB] ( 735 ) 201 Research on Increasing the Performance of Evolutionary Algorithm in Searching Robust Optimal Solutions Based on Quasi-Monte Carlo Method ZHU Yun-Fei, LUO Biao, ZHENG Jin-Hua, CAI Zi-Xing Robust optimal solution is of great significance in engineering application. It is one of the most important and difficult topics in evolutionary computation. Monte Carlo Integral (MCI) is generally used to approximate effective objective function (EOF) in searching robust optimal solution with evolutionary algorithm (EA). However, due to the low accuracy in existing crude Monte Carlo (C-MC) method, the performance of searching robust optimal solution with EA is unsatisfactory. Therefore, a Quasi-Monte Carlo (Q-MC) method is proposed to estimate EOF. The experimental results demonstrate that the proposed Q-MC methods-SQRT sequence, SOBOL sequence and Korobov Lattice approximate EOF more precisely compared with C-MC method, and consequently, the performance of searching robust optimal solution with EA is improved. 2011 Vol. 24 (2): 201-209 [Abstract] ( 626 ) [HTML 1KB] [ PDF 493KB] ( 624 ) 210 A Covering Incremental Reduction Algorithm Based on Absolute Information Quantity LIN Guo-Ping, LI Jin-Jin A covering reduction algorithm is studied under the condition that both the upper approximation operator and the under approximation operator are not changed. The absolute information quantity, information quantity and the adjacency matrix are defined. An incremental reduction algorithm is presented based on the absolute information quantity for covering generalized rough sets. The example shows that the proposed algorithm is an effective technique to remove the redundant knowledge in the complex datasets. 2011 Vol. 24 (2): 210-214 [Abstract] ( 604 ) [HTML 1KB] [ PDF 268KB] ( 589 ) 215 Research and Development of Cognitive Computing in Mind WANG Zhi-Liang , ZHENG Si-Yi, WANG Xian-Mei , WANG Wei Cognitive computing in mind is an important part of intelligent human-computer interaction (HCI), which attracts increasing attention from researchers in recent years. The development of the research advances at home and abroad is surveyed. Firstly, the conceptions correlated to cognitive computing in mind are introduced, and then the mechanisms and research contents in mind-reading are described in detail. Secondly, main neurobiological achievements of mental cognition are summarized and the research status of cognitive state in mind is compared to affective state. Moreover, the application trends of mental cognition in HCI are analyzed from the aspects of model establishment and mode extraction. Then, the primary frame of vision cognitive computing model of mental state with multi-level and multi-mode information fusion is put forward. The existing problems and the significance of cognitive computing in mind are finally discussed. 2011 Vol. 24 (2): 215-225 [Abstract] ( 938 ) [HTML 1KB] [ PDF 849KB] ( 1337 ) 226 Multi-Feature Fusion Based Circular Traffic Sign Detection ZHANG Jing, HE Ming-Yi, DAI Yu-Chao, QU Xiao-Gang Based on color and shape features of circular traffic sign, a multi-feature fusion based circular traffic sign detection algorithm is proposed. Firstly, color segmentation and achromatic decomposition are used to extract area containing circular traffic signs, and thus most of the background area is eliminated. Then, edge detection is applied, and the output edges are stored by chain code. Features, including length of contour, circularity and aspect ratio, are used to further eliminate background area. Finally, according to the shape feature of circular traffic sign, nonlinear least squares method is employed to detect the exact traffic signs. Effectiveness of the proposed algorithm is demonstrated preliminarily on real images captured under various weather and illumination conditions. 2011 Vol. 24 (2): 226-232 [Abstract] ( 930 ) [HTML 1KB] [ PDF 450KB] ( 841 ) 233 Multi-Criteria Recommendation Algorithm Based on Widrow-Hoff Neural Network ZHANG Fu-Zhi, CHANG Jun-Feng, WANG Dong To solve the problem that the traditional collaborative filtering recommendation algorithm can not recommend with multiple criteria, a multi-criteria recommendation algorithm based on Widrow-Hoff neural network is proposed by introducing the concept of multi-criteria rating for extending the standard collaborative filtering algorithm. The Widrow-Hoff least mean square (LMS) adaptive algorithm has the characteristics of high accuracy fitting in the process of system identification. Based on that, an approach to compute user preferences eigenvector based on Widrow-Hoff LMS algorithm is proposed. The user preferences eigenvector and spatial distance are adopted to measure user similarity and then a neighbor set for the best recommendations is located. Experimental results show that the proposed algorithm improves the accuracy and the quality of recommendation. 2011 Vol. 24 (2): 233-242 [Abstract] ( 615 ) [HTML 1KB] [ PDF 638KB] ( 990 ) 243 Ontology Mapping Method Based on Ontology Partition LI Zhi-Ming, LI Shan-Ping, YANG Chao-Hui, LIN Xin The mapping efficiency is key to the performance of dynamic ontology mapping in semantic web service discovery, context-awareness in smart spaces and so on. The existing methods simplify the current methods of similarity computation to promotes the efficiency, nevertheless they fail in the case that the number of candidate mapping entity pairs increases when ontology gets larger. An efficient ontology mapping method based on ontology partition is proposed, which divides an ontology into a set of blocks through bottom-up clustering. Then, the blocks are mapped and candidate mapping entity pairs are selected from the block mapping result. The experimental results show that the proposed method promotes the efficiency of mapping significantly with 6 times faster than it of Falcon-AO. 2011 Vol. 24 (2): 243-248 [Abstract] ( 520 ) [HTML 1KB] [ PDF 356KB] ( 630 ) 249 Modified Katz Approach and Its Application in Speech Recognition Based on Lattice ZHANG Lei, LU Dong, XIANG Xue-Zhi In traditional Katz approach, the discount coefficients may be greater than 1 or can not be calculated in some serious conditions. The idea of smoothing in log domain of couple occurrence number in simple good-turing is adopted. The modified Katz approach is proposed combined with back-off model. The proposed approach is further applied in speech recognition system based on lattice. The analysis of the effects on the structure and performance of lattice with different language models is given. Experiments show that the modified Katz approach enhances the system performance compared with traditional Katz approach. The best recognition rate achieves 60.90% for the corpus from interview program. 2011 Vol. 24 (2): 249-254 [Abstract] ( 591 ) [HTML 1KB] [ PDF 351KB] ( 581 ) 255 Tensor Completion Algorithm and Its Applications in Face Recognition SHI Jia-Rong, JIAO Li-Cheng, SHANG Fan-Hua Missing data problems are commonly attributed to the matrix completion problem, and matrix completion is an important method of signal acquisitions following compressing sensing. The data examples have the property of multi-linearity in applications, that is, the data set can be represented by higher order tensors. The tensor completion problem and its applications in face recognition are studied. Based on lower-dimensional Tucker decomposition of tensors, an iterative algorithm is proposed to complete tensors. And the distance between the estimating tensor and its Tucker approximation tensor monotonically decreases during the iterative procedure. Experimental results demonstrate the effectiveness and feasibility of the proposed method in completing tensor and face recognition. 2011 Vol. 24 (2): 255-261 [Abstract] ( 972 ) [HTML 1KB] [ PDF 386KB] ( 1237 ) 262 A Deep Web Query Interface Matching Approach Based on Evidence Theory and Task Assignment DONG Yong-Quan, LI Qing-Zhong, DING Yan-Hui, Zhang Yong-Xin To solve the limitations of existing query interface matching which have the difficulties of weight setting of the matcher and the absence of the efficient processing of matching decision, a deep web query interface matching approach based on evidence theory and task assignment is proposed called evidence theory and task assignment based query interface matching approach(ETTA-IM). Firstly, an improved D-S evidence theory is used to automatically combine multiple matchers. Thus, the weight of each matcher is not required to be set by hand and human involvement is reduced. Then, a method is used to select a proper attribute correspondence of each source attribute from target query interface, which converts one-to-one matching decision to the extended task assignment problem. Finally, based on one-to-one matching results, some heuristic rules of tree structure are used to perform one-to-many matching decision. Experimental results show that ETTA-IM approach has high precision and recall measure. 2011 Vol. 24 (2): 262-271 [Abstract] ( 252 ) [HTML 1KB] [ PDF 629KB] ( 622 ) 272 A Topic-Oriented Photo Browsing Method in Flickr Group ZHENG Nan, LI Qiu-Dan With the development of Web 2.0, social tagging systems, such as Flickr, become more and more popular. As a new developed user community, Flickr group attracts tens of thousands of online users to participate and becomes one of the research hotspot in recent years. In this paper, a topic-oriented photo browsing method is proposed to deal with the photo browsing problems in Flickr groups which lack of clear themes. The proposed method consists of three main steps: matrix space representation of photo-tag relations, non-negative matrix factorization based photo topic finding and a computational model of photo influence strength. The experimental results on Flickr dataset show the proposed method is useful and effective. 2011 Vol. 24 (2): 272-276 [Abstract] ( 656 ) [HTML 1KB] [ PDF 349KB] ( 647 ) 277 Image Segmentation Using Generalized Integrated Squared Error-Based Eigenvector Selection ZHANG Da-Ming, FU Mao-Sheng, LUO Bin Not all of the top eigenvectors contain clustering information for the task of real-world data clustering. Since the noise exists, the distribution of elements of an eigenvector is complex and it is necessary to select eigenvectors for spectral clustering. In this paper, the integrated squared error (ISE) divergence is generalized and the proposed generalized integrated squared error (GISE) is used to estimate the multimodality of data distribution and measure the clustering information of eigenvector. Then, a spectral clustering algorithm based on eigenvector selection is proposed. The experimental results on varied natural images segmentation show that the proposed algorithm is simpler and more effective than pervious algorithms. 2011 Vol. 24 (2): 277-283 [Abstract] ( 706 ) [HTML 1KB] [ PDF 450KB] ( 614 ) 284 Online Signature Verification SystemBased on Support Vector Data Description ZOU Jie, WU Zhong-Cheng An online signature verification system is proposed based on support vector data description (SVDD). Firstly, correspondences of the critical points in signatures are confirmed by bidirectional backward-merging dynamic time wrapping algorithm. Then, subtle differences in the local are calculated by classical dynamic time wrapping algorithm. Feature selection principle based on mean and deviation minimization is proposed. Finally, the classifiers are designed using SVDD. To obtain better result, m-fold cross validation and genetic algorithm are used to seek optimal parameters of SVDD. The average equal error rate for skill forge signatures on SVC2004 signatures database is 4.25%. 2011 Vol. 24 (2): 284-290 [Abstract] ( 546 ) [HTML 1KB] [ PDF 434KB] ( 881 ) 291 A Randomized Corner Detection Algorithm L Na, FENG Zu-Ren There is no parametric formulation of corner feature. Therefore, the conventional Hough transform can not be employed to transform the corner detection into maximum search in parametric space. A randomized Hough transform in Monte Carlo framework is presented, which detects the corner by searching for the local maximum in the intersection point cumulative space instead of parametric space. The intersection point cumulative space is a concept based on the fact that the corner is the intersection point of two lines. The proposed algorithm is demonstrated and the computing procedures are given. The proposed algorithm is isotropic, robust to image rotation, insensitive to noise and not susceptible to diagonal edge. Experimental results show that it outperforms Harris detector, Shen Wang algorithm, and SIFT feature detection algorithm. 2011 Vol. 24 (2): 291-298 [Abstract] ( 468 ) [HTML 1KB] [ PDF 1064KB] ( 665 ) 299 Hand Gesture Recognition Based on Online PCA with Adaptive Subspace YAO Ming-Hai, QU Xin-Yu The learning method for hand gesture recognition system based on vision is commonly off-line, which results in repeated off-line learning when new hand gestures come. Its real-time performance, expansibility and robustness are poor. In this paper, a method named online principle component analysis (PCA) with adaptive subspace is proposed for hand gesture recognition. The subspace is updated online by calculating PCA of sample coefficients. The subspace updating strategy is adjusted according to the degree of difference between new sample and learned sample. The algorithm is able to adapt to different situations and reduce the cost of calculation and storage. The incremental online learning and recognition of hand gestures are realized by the proposed algorithm. Experimental results show that the proposed method solves the unknown hand gesture problem, realizes online hand gesture accumulation and updating and improves the recognition performance of system. 2011 Vol. 24 (2): 299-304 [Abstract] ( 619 ) [HTML 1KB] [ PDF 394KB] ( 720 )

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