ISSN 1000-1239 CN 11-1777/TP

Table of Content

15 September 2005, Volume 42 Issue 9
Research on Incremental Learning of Bayesian Network Structure Based on Genetic Algorithms
Wang Fei, Liu Dayou, and Wang Songxin,
2005, 42(9):  1461-1466. 
Asbtract ( 579 )   HTML ( 1)   PDF (316KB) ( 617 )  
Related Articles | Metrics
Because of great difference in constructed model and changes in the dynamics of the domains, it is necessary to improve the performance and accuracy of a Bayesian network as new data is observed. A genetic algorithm is introduced to refine Bayesian networks in which both parameters and structure are expected to change. The genetic operators iteratively refine a Bayesian network based on ‘goodness’ evaluated by fitness function. This function contains three parts, exactitude that a Bayesian network respectively matches the old data and the new data, and conciseness of the model. To make computation feasible, the old incomplete data is converted into complete data based on expectation theory via the learned Bayesian network from the old data, and the new incomplete data is completed based on the available best Bayesian network in the last genetic iteration. Besides, the old data is compressed to expected sufficient statistics instead of some Bayesian networks. It not only economizes storage but also reduces the complexity of computation. Experimental results show this algorithm can make good choice between quality of the result and quantity of storage, can effectively refine Bayesian network structure from incomplete data.
MLSVM4—An SVM Fast Training Algorithm Based on Multi-Lagrange Multiplier
Ye Ning, Sun Ruixiang, and Dong Yisheng
2005, 42(9):  1467-1471. 
Asbtract ( 476 )   HTML ( 0)   PDF (303KB) ( 482 )  
Related Articles | Metrics
Sequential minimal optimization (SMO), as a popular effective approach to train the support vector machine for large data set has some drawbacks. Since during every iteration it selects the two samples violating KKT conditions most with the help of random function to train support vector machine, the randomness makes it unable to converge steadily. Based on the new analytical method proposed before, the which incorporates multiple Lagrange multipliers to optimize support vector machine, a new algorithm MLSVM4 with multiplier 4 is proposed without the help of random function. Because it can more accurately select the samples used during the iteration, it can converge much faster than the other methods proposed before, especially in the case of support vector machine with linear kernel. Experiment on a large range of standard data sets, such as Adult, Web and handwriting digital data, shows that MLSVM4 performs better with the factor of 3 to 42 times than SMO methods.
Layer Allocation Algorithms in Layered Peer-to-Peer Streaming with Source Server's Participation
Liu Yajie and Dou Wenhua
2005, 42(9):  1472-1477. 
Asbtract ( 515 )   HTML ( 0)   PDF (325KB) ( 572 )  
Related Articles | Metrics
Peer-to-peer streaming is cost-effective for it can capitalize the resources of peer nodes to provide service to other receivers. But in peer-to-peer streaming, as the outbound bandwidths of supplying peers or the bandwidths they willing to contribute are normally constrained and limited, the source media server can still become overloaded when the system's scale is very large. Considering layered peer-to-peer media streaming, it is needed to optimally allocate layers among peers to maximally save the source server's bandwidth consumption, and the optimal allocation problem is NP-hard. Two algorithms are analyzed for this optimization problem: the first one is an approximation algorithm which is based on the principle of multiple objective optimization, and the algorithm's approximation ratio is also analyzed; The second one is a branch and bound accurate algorithm, and by calculating a constructed bigraph's maximum flow it can identify the corresponding branch's upper bound and can thus decide whether to prune this branch or not. Performance study based on simulation is carried out. The results show that the two proposed algorithms have better performance than the related algorithm, and the accurate algorithm's branch and cut principle is high-efficient.
A Temporal Logic Semantics for UML Activity Diagrams
Zhu Xueyang and Tang Zhisong
2005, 42(9):  1478-1484. 
Asbtract ( 584 )   HTML ( 1)   PDF (438KB) ( 653 )  
Related Articles | Metrics
UML activity diagrams can be used to describe the control flow of different abstract levels and are very suitable for modeling system behaviors. However, the lack of precise semantics makes it difficult to analyze properties of the system they describe. In this paper, a variant of UML activity diagrams—XYZ activity diagrams which can be translated easily from activity diagrams is given, and a data structure of directed graph used to represent the activity diagram is defined. Its semantics then is interpreted using the executable linear temporal logic language XYZ/E, which can represent both dynamic semantics and static semantics. Compared with other formalizations such as CSP, ASM, and FSP, the XYZ/E semantics for UML activity diagrams is more intuitive. And the formalized activity diagram can be analyzed within a unified logical framework. An example is included.
Research on Decomposition Problem of Temporal Elementary Key Normal form and Temporal Simple Normal Form in Temporal Database with Multiple Time Granularities
Hao Zhongxiao, and Li Yanjuan
2005, 42(9):  1485-1492. 
Asbtract ( 502 )   HTML ( 0)   PDF (557KB) ( 520 )  
Related Articles | Metrics
The purpose of a good database logical design is to eliminate data redundancy and insertion, deletion and update anomalies. Temporal database is the same case. In this paper, the notions of temporal elementary functional dependency, temporal elementary key, temporal simple key are introduced. On this basis, the normalization of temporal database is studied by using constraints of temporal functional dependency(TFD) with multiple time granularities; the concept of temporal elementary key normal form(TEKNF) and temporal simple normal form(TSNF) is introduced; the proof that the normalization degree of both normal form is between T3NF and TBCNF and the normalization degree of TEKNF is lower than that of TSNF is given. Decomposition algorithms that give lossless, dependencypreserving, TEKNF decompositions and lossless, dependencypreserving, TSNF decompositions and the proof for its termination and correction are also given.
An Effective Distributed k-Means Clustering Algorithm Based on the Pretreatment of Vectors' Inner-Product
Ni Weiwei, Lu Jieping, and Sun Zhihui
2005, 42(9):  1493-1497. 
Asbtract ( 548 )   HTML ( 0)   PDF (349KB) ( 949 )  
Related Articles | Metrics
Clustering is an important research in data mining. Clustering in large data sets becomes a nut with the accumulating of the data. Despite its simplicity and its linear time, a serial k-Means algorithm's time complexity remains expensive when it is applied to a large data set. Distributed clustering is an effective method to solve this problem. In this paper, the knowledge of vectors' inner product inequation is adopted to improve efficiency of the existing parallel k-Means algorithm(k-DMeans), and an effective distributed k-Means clustering algorithm k-DCBIP is proposed. Theoretical analysis and experimental results testify that k-DCBIP outperforms the algorithm k-DMeans, and it is effective and efficient.
Finding Outliers in Distributed Data Streams Based on Kernel Density Estimation
Yang Yidong, Sun Zhihui, and Zhang Jing,
2005, 42(9):  1498-1504. 
Asbtract ( 469 )   HTML ( 0)   PDF (497KB) ( 813 )  
Related Articles | Metrics
Recently, there has been occurring more and more applications based on data stream models. Data mining in data stream, such as clustering, classifying, etc, becomes a hot research field. This paper presents an algorithm for outlier detection in distributed data streams. The data stream on every distributed node is taken for a subset of the global data stream, which consists of data on all distributed nodes. Because of huge network traffic, it is impossible to send all data to a central node and do detection. Based on the communication of distribution information between distributed nodes and the central node, the algorithm maintains the density estimation for the union of all streams. On every distributed node, global outliers can be detected by the estimation. Details of communication schedule and outlier detection are also discussed in this paper. Experimental results show promising availabilities of the approach.
Bisecting Grid-Based Clustering Approach and Its Validity
Yue Shihong and Wang Zhengyou
2005, 42(9):  1505-1510. 
Asbtract ( 423 )   HTML ( 0)   PDF (374KB) ( 630 )  
Related Articles | Metrics
A new grid-based clustering approach is presented. By hierarchically bisecting each grid into two volume-equal new grids, this approach can use a new criterion to measure the dissimilarity among all grids and find these candidates of all prototypes. Therefore, the proposed approach can overcome the parameter-sensitive defects in most conventional grid-based clustering approaches, and work well in such dataset with arbitrary-shaped and density-skewed clusters at linear computational complexity. Two experiments are used to verify its clustering effectiveness.
pepReap: A Peptide Identification Algorithm Using Support Vector Machines
Wang Haipeng, Fu Yan, Sun Ruixiang, He Simin, Zeng Rong, and Gao Wen,
2005, 42(9):  1511-1518. 
Asbtract ( 580 )   HTML ( 1)   PDF (505KB) ( 599 )  
Related Articles | Metrics
Protein identification plays an important role in proteomics. An algorithm for peptide identification using support vector machines (SVM), pepReap, which consists of two-layered scoring scheme, is designed and implemented. First, a list of peptide candidates is obtained by coarse scoring calculated from total intensity and number of matched peaks, and peptide length. Second, the above preliminary peptide candidates are evaluated by an SVM-based scoring scheme using other important factors, such as correlations between ions, average match error, peptide sequence information, to improve the reliability of peptide identifications. Matthews correlation coefficient is used to measure the classification performance in the SVM training process in order to accommodate to unbalanced datasets. Experiments on a public dataset of tandem mass spectra demonstrate that the pepSeap algorithm outperforms the popular software SEQUEST which uses threshold evaluation in terms of identification sensitivity with comparable precision.
Multimodal Particle Swarm Optimization for Neural Network Ensemble
Liu Yu, Qin Zheng, Lu Jiang, and Shi Zhewen
2005, 42(9):  1519-1526. 
Asbtract ( 524 )   HTML ( 0)   PDF (478KB) ( 625 )  
Related Articles | Metrics
The multimodal particle swarm optimization algorithm is proposed for neural network ensemble. In every iteration, particles are dynamically partitioned into several clusters, and then the difference between the best particles of clusters in output space is calculated. When the best particles of the two clusters have little difference, the two clusters are combined. Therefore, finally several clusters with much difference not only in network weight space but also in output space are achieved. Each cluster is responsible for one component network training. The best particle in a cluster corresponds to a component network. The number of component networks in the ensemble is automatically determined. The proposed method can effectively control the diversity of the networks. Compared with evolutionary ensembles with negative correlation learning algorithm, bagging and boosting, the experimental results show that the proposed method greatly improves the generalization ability of the neural network ensemble.
Research on Explanation Function for Reason Conclusions with Bayesian Network
Wang Ronggui, Zhang Yousheng, Gao Jun, and Peng Qingsong
2005, 42(9):  1527-1532. 
Asbtract ( 455 )   HTML ( 0)   PDF (431KB) ( 469 )  
Related Articles | Metrics
In this paper, an explanation function about Bayesian network is presented. With it, evidences' effect degree, direction and paths on reason conclusion can be explained. Necessity factor and sufficiency factor are designed as a measure approach, to valuate evidences' effect degree on posteriori distributions. By the way of qualitatively analysis the character of network structure, notes relative to reason conclusion are find out. Based on those notes, and combined with the quantitatively analysis, sub chains which consist of effect paths are found out, too. Those sub chains are valuated to create and explain the effect paths. Experiment results show the effectiveness of the explain function.
Ensemble-Based Manifold Learning for Visualization
Zhan Dechuan and Zhou Zhihua
2005, 42(9):  1533-1537. 
Asbtract ( 696 )   HTML ( 0)   PDF (324KB) ( 779 )  
Related Articles | Metrics
Manifold learning is helpful to the discovery of the intrinsic distribution and geometry structure of data. Current manifold learning algorithms are usually sensitive to noise and input parameters. The appearance of noise and the change of input parameters usually produce significantly different learning results. In this paper, a new method is proposed based on the manifold learning algorithm Isomap through introducing ensemble learning technique, which enlarges the value range that the input parameters can take to generate good visualization effect and reduces the sensitivity to noise.
A Method of Autonomous Robot Navigation in Dynamic Unknown Environment
Meng Wei, Huang Qingcheng, Han Xuedong, and Hong Bingrong
2005, 42(9):  1538-1543. 
Asbtract ( 528 )   HTML ( 3)   PDF (354KB) ( 625 )  
Related Articles | Metrics
An autonomous navigation method of robot in dynamic unknown environment is presented. Human's assistance is used instead of particular description of map. There are two phases in this method, which are human's guiding phase and autonomous navigation phase. In phase of human's guiding, a rough polar coordinate map of local environment is obtained by multi sensor information fusion and through it the global map is obtained. The method of eliminating sensor data error is also given here. According to map obtained in guiding phase, a robot moves in dynamic environment in the phase of autonomous navigation. Constraint condition of robot's motion control and dynamic obstacle avoiding method are also presented. Robot can dispose of emergent obstacles in this method, and optimization of path is also attained. Experiment results show the effectiveness of this method.
Study of Key Techniques of the Inference Machine Model for Function-Structure Project of the New Instrument Product Development Based on Genetic Algorithm(GA)
Shang Jiandong
2005, 42(9):  1544-1549. 
Asbtract ( 464 )   HTML ( 0)   PDF (346KB) ( 477 )  
Related Articles | Metrics
Virtual prototype reality design (VPRD) is a new technique that has developed recently. Here studied are the problem of product modeling as well as the problem of integration product and process design based on virtual environment simulation. The inference machine model for function-structure project is one of important modules which consist of the framework of VRPD. According to the complexity of function-module combination on project optimization issue, the realizing techniques on building the inference machine model for function-structure project based on genetic algorithm(GA) are given. The algorithms of double-stranded exclusive OR crossover operator and adaptive communication selection to crossover operation are proposed. The advantage of GA-model is verified by the empirical test of instruments. The presented techniques have been applied to a decision support system (DSS) for the new product development of instruments.
Adaptive RSVP Path Fast Handoff Scheme in Wireless/Mobile Networks
Jiang Aiquan, Wu Jiagao, and Ye Xiaoguo
2005, 42(9):  1550-1557. 
Asbtract ( 459 )   HTML ( 0)   PDF (582KB) ( 537 )  
Related Articles | Metrics
Among the schemes of mobile IP and RSVP integration, common router's computation is the key to reduce RSVP path handoff delay and end-to-end cost. In this paper RSVP path handoff schemes in IP networks are studied, an adaptive common router selection algorithm(ACRS) is represented, which satisfies application's quality of service (QoS) requirement, and balance between RSVP path handoff delay and end-to-end cost is kept via control parameter λ. In ACRS, optimum common router (OCR) is selected by two steps. Firstly, the node in previous path, which is nearest to new arrived node, is selected as common router, if the nodes are more than one; the nearest node to correspondence host is selected as CR. The RSVP path through CR between new arrived node and correspondence host must satisfy application's QoS requirement. Secondly, the RSVP path through CR between new arrived node and correspondence host is inspected, if there exists an zigzag path, then CR is re-selected, otherwise the selected CR is OCR. Through the performance comparison with other RSVP path handoff schemes, the results of experiment show that ACRS can support fast and adaptive RSVP path handoff.
Secure Access Control for Group Communication on Multi-Autonomous Domains Collaborative Environment
Zhang Yu, Zhang Wenyi, Li Xianxian, and Huai Jinpeng
2005, 42(9):  1558-1563. 
Asbtract ( 347 )   HTML ( 1)   PDF (396KB) ( 512 )  
Related Articles | Metrics
Secure communication environment of multiple autonomous domains collaboration is the basis of large-scale distributed applications, group communication with the character of high efficiency and flexibility is the basic communication mode. However, these collaborative applications lack central control, and in addition their users and resources belong to different autonomous domains. Users in collaborative environments expect to join/leave group, access domain resources dynamically, which leads to large numbers of new security challenges and access control problem. In view of the heterogeneous and dynamic character of multiple autonomous domains collaboration, role-based access control with distributed trust management is complemented and a role-based distributed trust management framework is proposed, thus resolving dynamic joint authorization and attribute-based delegation authorization. Meanwhile, an infrastructure is presented, which includes security policy negotiation, credentials issue, proof-of-compliance for the credentials and access control policy, and reasoning about users' access rights. A more flexible, reliable, secure access control model is provided for the collaborative environment of multi-domains group communication.
Forwarding State Reduction Scheme Based on Interface Format for Sparse Mode Multicast
Huang Kui, Wu Yichuan, Zheng Jianping, and Wu Zhimei
2005, 42(9):  1564-1570. 
Asbtract ( 373 )   HTML ( 0)   PDF (419KB) ( 466 )  
Related Articles | Metrics
Forwarding state proportional to the number of multicast groups may cause the scalability problem of multicast routing protocols. Existing intra-group forwarding state reduction approaches for sparse mode multicast, which aim to remove forwarding states at non-branching routers, are analyzed, and some problems appearing in these schemes are presented. A new forwarding state reduction scheme is proposed, which reduces storage requirement of multicast forwarding state by dynamically modifying the format of each interface in the multicast forwarding table entries according to the location and number of branching points on the multicast tree.
Modeling and Analysis of Non-Repudiation Protocols by Using Petri Nets
Li Botao and Luo Junzhou
2005, 42(9):  1571-1577. 
Asbtract ( 522 )   HTML ( 0)   PDF (437KB) ( 471 )  
Related Articles | Metrics
Since Petri nets is a mature and widely-used tool for the description and analysis of concurrent actions, it has been widely used in many fields in computer science, including security protocols. As a special kind of security protocol, non-repudiation protocols have been analyzed with many formal methods in recent years. However, there is no published research on using Petri nets to analyze non-repudiation protocols. For the advantage of Petri nets, it is attractive to adopt it to analyze non-repudiation protocols. Techniques used in normal security protocols, however, are not all suitable for non-repudiation protocols. Therefore, a Petri nets based modeling and analysis approach is given, which can describe and analyze some non-repudiation properties that can not be described by some other methods. A fair non-repudiation protocol proposed by J. Zhou and D. Gollmann in 1996 is modeled and analyzed on CPN tools using this method and, a known flaw of the protocol that has not been discovered by many other formal methods is discovered.
Network Intrusion Detection and Attack Analysis Based on SOFM with Fast Nearest-Neighbor Search
Zheng Jun, Hu Mingzeng, Yun Xiaochun, and Zhang Hongli
2005, 42(9):  1578-1586. 
Asbtract ( 512 )   HTML ( 0)   PDF (584KB) ( 466 )  
Related Articles | Metrics
Owing to computer attacks becoming more complex, more and more machine learning algorithms are increasingly proposed to solve the problems of intrusion detection. But these algorithms have wide gap when applied in network intrusion detection systems(NIDS), especially in high-speed networking environments. In this paper, An NIDS based on self-organizing feature map (SOFM) is proposed. And to achieve more efficiency and usability, the vector elimination nearest-neighbor search (VENNS) algorithm is implemented for the NIDS, where the final aim is to reduce the system computational cost of training and detection. Using the DARPA Intrusion Detection Evaluation Data Set, the performance evaluation and comparison analysis are implemented. It is shown that network attacks are detected with the higher detection rates and relatively the lower false positive rates. The performance and efficiency of NIDS are improved greatly: the training time cost the detection time cost can be shortened about by four times and seven times respectively.
Security Protocol and Scheme for Inter-Realm Information Accessing
Peng Shuanghe, Han Zhen, and Shen Changxiang
2005, 42(9):  1587-1593. 
Asbtract ( 497 )   HTML ( 0)   PDF (381KB) ( 521 )  
Related Articles | Metrics
In order to improve the security of Intranet, application boundary security devices must be set. In order to access resources in different application areas on Internet in a security way,authentication is the first key step. Kerberos is an authentication protocol that is widely used. It is applied in application boundary security devices such as socks5. But there exists some limitation. In the processing of authentication between application boundary security devices, the object authenticated by application boundary security device at resource realm is not client which requests the resource, but application boundary security device at principal realm. So the object audited by application boundary security device at resource realm isn't the really one. A new inter-realm authentication protocol and a new identity-passing protocol based on Kerberos v5 inter-realm authentication protocol are presented in this paper. The proposed protocols can not only supply the security audit for user's access requests at application boundary security devices but also improve the efficiency of communication system because it needs only two connections between realms and the connection is setup not by subjects and objects but by application boundary security device. The proposed scheme can solve the problem of security information transferring between enterprise networks which will expand its application boundary including current enterprise network.
A DDoS Attack Detection Method Based on Hidden Markov Model
Zhou Dongqing, Zhang Haifeng, Zhang Shaowu, and Hu Xiangpei
2005, 42(9):  1594-1599. 
Asbtract ( 652 )   HTML ( 1)   PDF (332KB) ( 754 )  
Related Articles | Metrics
The statistical characteristics of the selected data packets show anomalies under distributed denial of service (DDoS) attacks. The detection of the anomalies is an important task. Some detection methods are based on the hypothesis of data packet rates. This hypothesis, however, is unreasonable in some situations. Other detection methods are based on the statistics of IP addresses and the length of data packets, but their detection accuracy declines rapidly under the IP spoofing attack. In this paper, an HMM-based detection method of DDoS attacks is presented. The method integrates four different detection models against different type attacks. The models are established based on selected normal network data packet attributes, which are the flag bits of TCP packets, the ports of UDP packets, and the type and code of ICMP packets. These packets are from normal audit data. The models simulate the statistical characteristics of normal network data packets. The models are then used to detect the DDoS attacks by processing selected target audit data packets. Experimental results show that this method outperforms other methods reported on the DDoS attacks in adaptability and detection accuracy.
A Learning-Based Peer-to-Peer Search Algorithm
Chen Haitao, Gong Zhenghu, and Huang Zunguo
2005, 42(9):  1600-1604. 
Asbtract ( 353 )   HTML ( 3)   PDF (257KB) ( 531 )  
Related Articles | Metrics
Content search is an essential function, but it presents a very difficult and challenging problem for large-scale peer-to-peer systems. In this paper, a new learning-based algorithm——SmartSearch is introduced. SmartSearch learns passively interest similarity between nodes from history search results, divides nodes into interest groups, and constructs friend relations between nodes with similar interest which can be used to locate content effectively. Simulation results show that, compared to the Gnutella algorithm, SmartSearch improves query efficiency by up to ten times without a significant increases in load.
Digital Audio Watermarking Based on Support Vector Machine (SVM)
Wang Jian and Lin Fuzong
2005, 42(9):  1605-1611. 
Asbtract ( 730 )   HTML ( 2)   PDF (455KB) ( 702 )  
Related Articles | Metrics
A novel digital audio watermarking technique based on SVM(support vector machine) is proposed in this paper. The main idea of the method is that the retrieval of embedded watermark can be considered as a two-class problem, and SVM can be used to learn the characteristics of the embedded watermark in audio. Because the SVM possesses the learning and adaptive capabilities, it almost exactly recovers the watermark from the watermarked audio. The extensive experimental results show that the embedded watermark is robust against audio signal processing, such as MPEG audio coding, cropping, filtering, resampling and requantizing. The watermarking method does not require the use of the original signal for watermarking detection.
A Multi-View Face Detection Based on Real Adaboost Algorithm
Wu Bo, Huang Chang, Ai Haizhou, and Lao Shihong
2005, 42(9):  1612-1621. 
Asbtract ( 861 )   HTML ( 3)   PDF (752KB) ( 1248 )  
Related Articles | Metrics
In this paper, a multi-view face detection method based on real Adaboost algorithm is presented. Human faces are divided into several viewpoint categories according to their poses in 3D, and for each of these categories a form of weak classifiers in look-up-table (LUT) type is designed using Haar-like features that have confidences in real values as their outputs, and correspondingly its space of weak classifiers is constructed, from which the cascade face detector is learnt by using real Adaboost algorithm. For speed up, multi-resolution searching and pose prediction strategies are introduced. For frontal face detection, the experiments on CMU+MIT frontal face test set result in a correct rate of 94.5% with 57 false alarms; for multi-view face detection, the experiments on CMU profile face test set result in a correct rate of 89.8% with 221 false alarms. The average processing time on a PⅣ 2.4GHz PC is about 80 ms for a 320×240-pixel image. It can be seen that the proposed method is very efficient and has significant value in application.
Fingerprint Matching Based on Delaunay Triangulation
Yin Yilong, Zhang Hongwei, and Liu Ning
2005, 42(9):  1622-1627. 
Asbtract ( 572 )   HTML ( 1)   PDF (419KB) ( 937 )  
Related Articles | Metrics
In this paper, a method for fingerprint matching is proposed based on Delaunay triangulation in computational geometry. First, minutiae taken from the template and the query fingerprint images are triangulated using the Delaunay's rule. Then reference minutiae pairs are obtained by searching two DT nets. Finally the query fingerprint image is adjusted according to the template fingerprint image with parameters computed from reference minutiae pairs and match score is calculated using a simple match algorithm. The experiments conducted on BVC2004 confirm the effectiveness of the proposed algorithm.
Matrix-Pattern-Oriented Ho-Kashyap Classifier with Regularization Learning
Tian Yongjun and Chen Songcan
2005, 42(9):  1628-1632. 
Asbtract ( 932 )   HTML ( 3)   PDF (307KB) ( 573 )  
Related Articles | Metrics
Linear classifier is a common method in statistical pattern recognition. The modified Ho-kashyap with square approximation of the misclassification errors (MHKS) is a linear classifier designed similarly as the support vector machine to maximize the separation margin. However, the conventional linear classifiers are almost based on vector patterns, i.e., before applying them, any non-vector pattern should be first vectorized into a vector pattern. But, such a vectorization will bring three potential problems at least: ①Some implicit structural or contextual information may be corrupted; ②the higher the dimension of input pattern, the more space are needed for storing weight vector; ③When the dimension of pattern is very high and the sample size is small, linear classifier tends to be overstrained. In this paper, a new classifier design method based on matrix patterns, called two-sided linear classifier (MatMHKS), is proposed by modifying the MHKS algorithm. This method can overcome above problems. Experiment results on real dataset show that MatMHKS is more powerful than MHKS.
Strategy of Combining Multiple Models Based on Image Sequence
Cao Zhiqing, Shi Jiaoying, Zhang Shiming, Sun Xin, and Liu Peijun
2005, 42(9):  1633-1639. 
Asbtract ( 499 )   HTML ( 1)   PDF (443KB) ( 590 )  
Related Articles | Metrics
A single image or a couple of images always include enough geometry information for reconstructing a portion of model. But it is not sufficient when it comes to the whole model reconstruction, and the error accumulation is also a problem to get the intact model. Based on the structural regularity, a new strategy is proposed, which integrates the whole model by combining separate portions of model reconstructed from each of series images. The strategy is composed of four procedures: constructing the linked list of the view coordinates, converting the coordinates, merging the vertex, and integrating the whole model. The linked list of the view coordinates offers the method to construct the relationship for all the view models. More important, the epipolar line restriction is used to merge the vertex and integrate the whole model. The experimental results show that each step of the strategy is effective.
Feature Fusion Based on the Average Precision in Image Retrieval
Ru Liyun, Ma Shaoping, and Lu Jing
2005, 42(9):  1640-1646. 
Asbtract ( 585 )   HTML ( 1)   PDF (367KB) ( 559 )  
Related Articles | Metrics
In content-based image retrieval, image has various inherent aspects which reflect its contents, therefore how to organize and utilize these contents effectively to improve the retrieval performance is a valuable research topic. In this paper, a method of measuring the complementarities between two feature spaces is proposed, based on which the fusion feature set can be selected effectually, and the experimental results are positive. At the same time, a linear fusion method based on the average precision of features is proposed. Extensive comparisons against several methods, such as flat model and rank-based linear fusion are performed. Experiments are carried out on a large-size heterogeneous image collection consisting of 12000 images and the results demonstrate the effectiveness of the proposed method.