Loading...

Table of Content

      
    16 October 2010
    Volume 40 Issue 5
    Articles
    Method of feature generation and selection for network traffic classification
    YANG Ai-min1, ZHOU Yong-mei1, DENG He2, ZHOU Jian-feng3
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  1-7. 
    Abstract ( 281 )   PDF (1195KB) ( 2846 )   Save
    Related Articles | Metrics

    In the System of Network Traffic Classification based on machine learning method, feature generation and feature selection directly affects the speed and accuracy of classification. To solve this problem, in feature generation aspect, we analyze the packet’s attributes (size, count, time, flag) and flow’s attributes (time) from the information of Packet-Level and FlowLevel, and 37 statistical features are generated. In feature selection aspect, we proposes a method of feature selection integrating Filter model and Wrapper model, to decrease the dimension of features. Experiments show the proposed methods improve the accuracy of classification.

    Fast antigen detect method based on detection tree
    SUN Qiu-li, HAN Fang-xi, WANG Xiao-lin
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  8-11. 
    Abstract ( 345 )   PDF (1012KB) ( 1310 )   Save
    Related Articles | Metrics

    To overcome the problems of low detection efficiency such as duplicate detection,bit by bit comparison on detecting the antigen’s legitimacy with negative selection algorithm, a legitimacy detection method of antigen based on detection tree was proposed. It constructs a detection tree with detectors using step by step manner. The legitimacy of antigens could be detected by path searching in the detection tree. Experimental results show that the testing method based on detection tree is better than the method by traversing detector collection and bit by bit comparisons, which could meet the requirements of realtime detection. This provides an efficient way for anomaly detection based on artificial immune principle.

    An optimization approach to grid workflow scheduling using improved SPEA2 algorithm
    LI Jin-zhong1, XIA Jie-wu1, ZENG Jin-tao1, WANG Xiang2*
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  12-16. 
    Abstract ( 275 )   PDF (909KB) ( 2034 )   Save
    Related Articles | Metrics

    A multiobjective optimal grid workflow scheduling algorithm with QoS constraints, named ISPEA2 is proposed. The proposed algorithm, based on the rich-construct abstract grid workflow language (AGWL) grid workflow model, is introduced constraints detection into strength pareto evolutionary algorithm 2 (SPEA2) to optimize the grid workflow scheduling problem. The algorithm overcomes the following drawbacks: only considering DAG structure of grid workflow model, fewer multidimensional QoS parameters, and aggregating the multi-dimensional QoS parameters into a single objective function for optimal scheduling. Decision makers can choose a satisfied solution according to user’s preferences from the produced Pareto optimal solutions. Compared with a grid workflow scheduling algorithm OSPEA2 based on the original SPEA2, the experimental results show that all of Pareto optimal solutions are obtained by ISPEA2 which are non-dominated solutions of satisfying the QoS constraints and better mean result of solutions than OSPEA2.
     

    An AR parameters-based source selection method in general nonlinear blind source extraction
    CAI Ying 1, WANG Gang 2*
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  17-23. 
    Abstract ( 286 )   PDF (1371KB) ( 1499 )   Save
    Related Articles | Metrics

    In the nonlinear blind source separation (BSS) case,an AR parameters method was introduced as a new selection procedure. Compared with previous methods, the proposed algorithm not only can do the separation, but also can extract any desired signal with the corresponding AR parameter. It can deal with nonlinear blind source extraction (BSE) at the cost of more prior information and its performance is demonstrated on nonlinearly mixed speech data.
     

    A method based on FFD B-spline registration of the iris image fusion
    WU Guo-yao1, MA Li-yong2
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  24-27. 
    Abstract ( 331 )   PDF (765KB) ( 1640 )   Save
    Related Articles | Metrics

    Images collected by camera are usually blur and can not be used directly for iris recognition. However, as the iris texture may occur as the pupil of scaling of non-rigid deformation, direct image fusion does not work well. Firstly the B-spline free-form deformation model was used in this paper to register images, then the wavelet transform was employed in image fusion. Entropy was performed to judge the results of fusion. It was showed that the method was efficient and the image after fusion was intuitively detailed.

    Generalized fuzzy subinclines(ideals) with norm T
    RUI Ming-li, LIAO Zu-hua, HU Miao-han, LU Jin-hua
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  28-33. 
    Abstract ( 308 )   PDF (326KB) ( 1197 )   Save
    Related Articles | Metrics

    The concepts of generalized fuzzy subinclines(ideals) and generalized fuzzy subinclines(ideals) with norm T are given. Then some of their properties are discussed. Based on this, their equivalent characterizations are also obtained.

    Evolutionary algorithm based on idea of particle swarm optimization
    LIU Jianhua1,2, HUANG Tiangqiang2, YAN Xiaoming2
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  34-40. 
    Abstract ( 289 )   PDF (1148KB) ( 1045 )   Save
    Related Articles | Metrics

    Particle swarm optimization (PSO) is an intelligence algorithm simulated the social behavior of bird swarm or fish group. It is difficult for original formula of PSO to show mathematical essence and principle. Using the simplified modal of PSO, the current theoretical analysis of PSO has  constructed a mathematic modal that give a clear essence of PSO from mathematic view. Which has  illustrated that the PSO is an iteration evolutionary system. Using the mathematic modal of PSO, this paper develops a new evolutionary algorithm that velocity and location updating equation of PSO are replaced by the mathematic equations. And the some parameters of new algorithm are discussed and selected properly. With selection of appropriate parameters, the performance of new evolutionary algorithm is not inferior to standard PSO by simulation on benchmark functions. The new evolutionary algorithm is easy to understand and has mathematical meaning. Its parameters are fewer and easier to be analyzed than standard PSO.
     

    An examination of classification model with partial least square based dimension reduction
    ZENG Xue-qiang1, LI Guo-zheng2
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  41-47. 
    Abstract ( 256 )   PDF (2419KB) ( 2213 )   Save
    Related Articles | Metrics

    Among various methods, partial least square based dimension reduction (PLSDR) is one of the most effective one, which has been applied in many fields such as the analysis of microarray data. But the problem of choosing classification model with PLSDR has often been neglected, different classification models are applied arbitrary. Aim to this problem, an examination of different classification model with PLSDR by intensive experiments was gived.Furthermore,by using paired twotailed ttest, artificial neural network, logistic discrimination and linear support vector machine were suggested to be well performance classification models used with PLSDR.

    OPHCLUS:An order-preserving based hierarchical clustering algorithm
    LEI Xiao-feng1, ZHUANG Wei1, CHENG Yu1, DING Shi-fei1, XIE Kun-qing2
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  48-55. 
    Abstract ( 274 )   PDF (2152KB) ( 1632 )   Save
    Related Articles | Metrics

    The idea of maintaining order relation was proposed, i.e.,the original order of distance between samples should be preserved by the inter-cluster measurement of hierarchical clustering as far as possible. Based on this idea, we defined the notion of order relation of sample’s pair and the loss measurement of order relation, which could be used as the objective criteria function of clustering and the validity standard of consequent clusters. Furthermore, we extended two kinds of distance measurement from the loss of order relation, i.e.,inter-cluster adjusted distance and inter-cluster 0-1 weighted distance; implemented an order-preserving based hierarchical clustering algorithm by using these two measurements. The experiment simulation demonstrated the improvement in the clustering quality.
     

    Sampled peculiarity factor and its application in anomaly detection
    SUN Jing-yu, YU Xue-li, CHEN Jun-jie, LI Xian-hua
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  56-59. 
    Abstract ( 293 )   PDF (362KB) ( 1179 )   Save
    Related Articles | Metrics

    The peculiarity factor (PF), an important feature of data and obtained by accumulating differences between data, is a core concept of peculiarity-oriented mining (POM). But it meets a higher computational time complexity for any algorithm.A sampled approach firstly was suggested to define PF through analyzing current versions of PF and computational complexities of algorithms to compute it. A sampled PF (SPF) was proposed to meet realtime requirement and a SPFoutlier detection algorithm was given. Experiments using real datasets show that the SPF-outlier detection algorithm is efficient with losing a few of precisions through contrasting with two baseline algorithms and it is a feasible and right approach to define PF by sampling. Furthermore, some right sampling methods could be used to compute SPFs in order to meet real-time requirement.

    A fast SVM-based feature selection method
    DAI Ping, LI Ning*
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  60-65. 
    Abstract ( 418 )   PDF (354KB) ( 4278 )   Save
    Related Articles | Metrics

    Aiming at the large computation and slow convergence speed of the traditional feature selection methods, a fast SVMbased feature selection method is proposed to overcome.Support vecor machine is employed as the classifier and particle swarm optimization method is employed as searching strategy.The proposed method reduces the iterations of training classifiers by taking advantage of the characteristics of linear and polynomial kernel functions so that it reduces the complexity of calculation. Experimental results show that the method accelerates feature selection in the case of no loss of classification performances.

    Application of ReliefF feature evaluation in un-supervised manifold learning
    TAN Tai-zhe, LIANG Ying-yi, LIU Fu-chun
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  66-71. 
    Abstract ( 264 )   PDF (535KB) ( 1553 )   Save
    Related Articles | Metrics

    As regards to the noise-sensitive, vulnerable to the missing values problem, the complexity and the large sparseness of real world data, and so on, propose to introduce ReliefF feature evaluation, that is to apply it into manifold learning. The experiments are divided into four cases: one is not to use any feature selection algorithm; one is to use only ReliefF feature evaluation; one is to use only the representative Locally Linear Embedding algorithm; and the last one is to use both. Results show that the classifying accuracy rate obtained by using the improved algorithm is higher than by ReliefF or Locally Linear Embedding respectively.

    The AIM model of pattern expression based on  an agent
    CHENG Xian-Yi1,2 ZHU Qian2, GUAN Zhi-jin1
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  72-76. 
    Abstract ( 335 )   PDF (1033KB) ( 1590 )   Save
    Related Articles | Metrics

    To solve the existing problems of semantic losing in pattern expression, a new model of pattern expression is proposed, which called AIM(agent influence map) that based on agent and memory principle. AIM reflects the whole character of pattern, and it provides an effective soft computing tool to support adaptive behavior that based on prior knowledge. AIM shows the hierarchy of memory model by integrating the character in multi-stages, while the pattern information is stored in the entire network, and high-level features is manifested by collaboration, demonstrate memory’s semantic features.
     

    Music structure analysis based on lyrics and content
    LIANG Shuang, XU Jie-ping*, LI Xin
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  77-81. 
    Abstract ( 249 )   PDF (889KB) ( 2507 )   Save
    Related Articles | Metrics

    To solve the problem of high cost and redundant pieces generation of content-based analysis, a lyrics-based analysis to detect the music structure was proposed, combing SVM and beat to detect the boundaries of vocal and nonvocal, to correct the result of lyrics-based analysis. The experiment on the same database proves that our musical structure analysis algorithm that combines the lyrics and content has improved the accuracy of Bridge and Outro by 9% and 11% respectively.

    An immune network based unsupervised classifier
    LIANG Chun-lin1, PENG Ling-xi2*
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  82-86. 
    Abstract ( 267 )   PDF (583KB) ( 1214 )   Save
    Related Articles | Metrics

    A novel unsupervised classification algorithm based immune network was presented. First of all, the formal definitions of antibodies, antigens and immune network were given according to shape space theory, respectively. Afterward, the mathematical models and corresponding equations were established, such that the clonal selection and highfrequency mutation of antibodies, the immunological memory, and etc. Finally, the process of unsupervised classification was presented. The experimental results showed that the algorithm achieves the higher classification accuracy than other traditional clustering algorithms, and has some better characters such that continuous learning, dynamic adjustment, features remembering, and etc. If the antibody is regarded as a given model, and rationalizes the antigens collection, then the model has a wide range of applications.

    An empirical study on tongue image detection
    LI Guo-zheng1, SHI Miao-jing1, LI Fu-feng2, WANG Yi-qin2
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  87-95. 
    Abstract ( 351 )   PDF (2160KB) ( 1827 )   Save
    Related Articles | Metrics

    The extraction of tongue area from digital image is essential to an automatic tongue diagnostic system in Traditional Chinese Medicine. Classic image segmentation methods couldn’t be effective due to its feature diversity. First, it was  tried to categorize existing research in this field for recent 5 years, especially for Snake mold,practical test will be given at the same time so as to summarizing advantages and disadvantages of representative methods. Three innovated approaches are proposed afterwards, which improve the performances at certain degree.

    NURBS curve approximation based on annealing genetic algorithm
    LIU Bin, ZHANG Ren-jin
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  96-100. 
    Abstract ( 348 )   PDF (898KB) ( 1491 )   Save
    Related Articles | Metrics

    The annealing genetic algorithm is presented to approximate a sequence of characteristic points by NURBS curve with any order and any number of knots. First, the control vertices, weights of NURBS curves, knots sequence and t parameters approximating characteristic points were encoded as genes. Then the cross operator, mutation operator and annealing selection operator were executed cyclically to search the global optimum or the suboptimal. In the end, four NURBS curves with different number of control vertices and degree were used to approximate the same sequence of characteristic points. Four groups of numerical values and four graphics in different condition were presented. The example proves that the annealing genetic algorithm can stably approximate the NURBS curves with different degree and number of control vertices.

    Wood CT image registration by Harris corner detector
    ZHANG Xun-hua1, YE Ning2, WANG Hou-li3
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  101-104. 
    Abstract ( 240 )   PDF (766KB) ( 1930 )   Save
    Related Articles | Metrics

    Proposed a triangle method to describe Harris feature points, firstly, the algorithm uses Harris operator to detect the corner points of the two images, then sorts the corner points in accordance with the size of the weights, and uses the triangle method to describe the feature of the feature points in order to identify the feature points’ corresponding relationship between the two images, and has also obtained the image registration parameters. Experimental results show that the algorithm with high accuracy, speed, and has a strong robustness.

    The inter-sentence semantic relevancy degree calculation using the quantified correlation of words
    ZHONG Maosheng 1, LIU Hui2, ZOU Jian3
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  105-111. 
    Abstract ( 259 )   PDF (874KB) ( 2440 )   Save
    Related Articles | Metrics

    The coherence in form and relevancy in meaning between sentences in a context are the important grounds in text reasoning and text-structure analysis. There are two ways of analyzing the inter-sentence correlativity in a context—qualitatively and quantitatively. According to the analysis of quantified correlation between words, assuming that the inter-sentence correlativity in a context was the outcome of the quantified correlation between word-pairs, which consist of the adjacent sentences, this paper measured quantitatively the degree of semantic relevancy between sentences. The experiment results show that this method of relevancy measurement can avoid the window-length constraint which exists in similarity measurement; in addition, the calculation for the correlation coefficient of correlativity between sentences, done by hand and done by computer, indicates that the method can well-simulate the cognition of the human brain to measure the semantic relevancy between sentences in a context.

    iCome: image retrieval system based on ambiguity
    CHEN Hu, LI Ming*, JIANG Yuan, ZHOU Zhi-hua
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  112-116. 
    Abstract ( 286 )   PDF (1598KB) ( 1134 )   Save
    Related Articles | Metrics

     Recently, the advances in multi-media techniques greatly increase the number of images. Image retrieval becomes a hot topic. It is well-known that semantic gap is the major problem in contentbased image retrieval (CBIR), while the existing image retrieval systems cannot tackle this problem well. Since the ambiguity is one of the important reasons leading to semantic gap, we build iCome, an image retrieval system based on learning for ambiguity objects. Concerning the ambiguity in output space, iCome implements text-based image retrieval. Concerning the ambiguity in input space besides relevance feedback, iCome implements content-based image retrieval.
     

    Word-position-based tagging for Chinese word segmentation
    YU Jiang-de1, SUI Dan1, FAN Xiao-zhong2
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  117-122. 
    Abstract ( 295 )   PDF (478KB) ( 2177 )   Save
    Related Articles | Metrics

     The performance of Chinese word segmentation has been greatly improved by word-position-based approaches in recent years. This approach treats Chinese word segmentation as a wordposition tagging problem. With the help of powerful sequence tagging model, word-position-based method quickly rose as a mainstream technique in this field. Feature template selection is crucial in this method. We further studied this technique via using four wordpositions and conditional random fields. Closed evaluations are performed on corpus from the third and the fourth international Chinese word segmentation Bakeoff, and comparative experiments are performed on different feature templates. Experimental results show that the feature template set: TMPT-10′  is much better performance than the traditional template set.
     

    Im-IG: A novel feature selection method for imbalanced problems
    YOU Ming-yu, CHEN Yan, LI Guo-zheng
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  123-128. 
    Abstract ( 304 )   PDF (1115KB) ( 2346 )   Save
    Related Articles | Metrics

    Imbalanced data set is a ubiquitous problem in machine learning field, which attracts much attention from related scientists. Information Gain (IG) method is widely used in feature selection, but it is seldom researched in imbalanced problem. Based on the performance discussion of IG on imbalanced data sets, a new method Im-IG was proposed for imbalanced problem in feature selection. Im-IG increased the weight of minor class in the entropy calculation, in order to select features which were better for minor class. Im-IG focused on improving the classification accuracy of minor class, based on the performance improvement of the whole data set. Experimental results on several imbalanced data sets showed that Im-IG can solve the imbalanced predicament IG met and it was an effective feature selection method for imbalanced problem.

    Image retrieval algorithms based on manifold learning
    HE Guang-nan, YANG Yu-bin*
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  129-136. 
    Abstract ( 213 )   PDF (1026KB) ( 1644 )   Save
    Related Articles | Metrics

    The purpose of the manifold learning is to discover the intrinsic dimensions of nonlinear high-dimensional data, which makes it more suitable for data analysis and dimensional reduction. The gap between high-dimensional data space and low-dimensional semantic subspace forms the “semantic gap” problem in image retrieval. Although using relevance feedback mechanism can narrow down the gap and increase the retrieval accuracy, the limitations of relevance feedback and the high dimensionality of image features make it prone to the course of dimensionality. Manifold learning has brought promise for settling these problems. Using the learned intrinsic dimensions of highdimensional image feature data by manifold learning can considerably enhance retrieval performance. The image retrieval algorithms based on manifold learning all take semisupervised learning strategy. It makes the most of the feedback information to learn the semantic subspace of image, and reduces the high dimensionality effectively.

    Wood defect recognition based on BIRCH cluster algorithm
    WU Dong-yang, YE Ning
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  137-140. 
    Abstract ( 259 )   PDF (938KB) ( 1377 )   Save
    Related Articles | Metrics

    A new method for wood defect recognition based on BIRCH algorithm is been proposed. The problems about branch factor (B, L)、the selection of threshold T and the discrimination of non-defect class are been discussed. To produce the initial clustering, distinguish non-defect class for the initial clustering, automatically identify the location of the wood’s defects and mark it, a CFtree within a certain threshold is been built. The experimental results show that this algorithm can identify the wood’s defects efficiently, the average defecting precision ratio is about 86.3%, and the average defecting recall ratio is about 90.1%.
     

    Domain of automatic entity relation extraction based on seed self-expansion and maximum entropy machine learning
    LEI Chun-ya1, GUO Jian-yi1,2, YU Zheng-tao1,2, MAO Cun-li1,2, ZHANG Shao-min1, HUANG Pu1
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  141-145. 
    Abstract ( 228 )   PDF (672KB) ( 1725 )   Save
    Related Articles | Metrics

    Entity relation extraction is one of difficulties in information extraction’s field.In this paper, a method that seed selfexpansion and maximum entropy machine learning was proposed to extract entity relation in the filed of tourism. Firstly, used seed self-expansion to get words semantic that express the big types relation between entity pairs, and this words semantic as a characteristic was added to the set of characteristics, meanwhile designed threshold to tag studying corpus automatically; then used maximum entropy machine learning algorithm to learn corpus tagged and built the classifier of entity relation extraction. Experiments based on artificial collection of 600 corpuses obtained a better result for four big types of entity relation extraction, the F values reached 82.56% and 81.17% in which the two big types relation of geographical location and date-season, it showed in the condition of less manual participation, adding the word semantic of entity pairs could effectively improve the performance of classifier.

    Study on validity of hierarchical clustering
    HU Xiao-qing1, MA Ru-ning1*, ZHONG Bao-jiang2
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  146-149. 
    Abstract ( 253 )   PDF (1247KB) ( 1155 )   Save
    Related Articles | Metrics

    In allusion to choose the most satisfactory classification from several partitioning results of the dataset obteined by hierarchical clustering, after deeply studying clustering validity indices, a new clustervalidity index was established via describing compactness and separation using the fuzzy similarity matrix of the dataset. The experimental results on both synthetic and real-world datasets have demonstrated the effectiveness of the new cluster-validity index.

    Semi-supervised image retrieval based on diversity and invariant features
    SU Hong-lu, LI Fan-zhang*
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  150-153. 
    Abstract ( 251 )   PDF (683KB) ( 1049 )   Save
    Related Articles | Metrics

    Based on the idea of isomorphism, the image translation and rotation invariant feature could be formalized through the bispectrum. In order to expand the semantic scope of the results, a method called image retrieval based on diversity and invariant features (IRDIF) which can expand the diversity in search result has been applied in the semisupervised image retrieval. The item which has been visited will be set to be absorbing state, then the other items which are similar to the item of absorbing state will have smaller visiting probability. With this method, an experiment upon Corel database is conducted, and the final effect turns out to be quite satisfactory.

    Quality evaluation for fingerprint image based on orientation field
    LI Tie-jun, LIU Qian, ZHANG Yu
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  154-158. 
    Abstract ( 288 )   PDF (1156KB) ( 1084 )   Save
    Related Articles | Metrics

    Fingerprint image quality seriously affects the performance of fingerprint identification system, quality evaluation for fingerprint image has important applications in fingerprint identification system such as the fingerprint segmentation, matching etc, and is also of great significance on the research of fingerprint recognition algorithm. In this paper, using fingerprint image orientation field information, a quality evaluation method for fingerprint image based on the continuity of the orientation field information is presented. For each fingerprint image subblock, the original orientation field information and the new orientation field information smoothed through Lowpass filtering are calculated in this method, and the quality of fingerprint image is evaluated based on the statistics of the rate of orientation change after Lowpass filtering. Experimental results show that the method proposed can more effectively improve the quality classification accuracy on fingerprint image with low quality.

    A novel users’ interests prediction approach based on concept lattice
    MAO Qin-jiao1, FENG Bo-qin1, LI Yan1,2, PAN Shan-liang3
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  159-163. 
    Abstract ( 254 )   PDF (674KB) ( 1618 )   Save
    Related Articles | Metrics

    Traditional collaborative filtering methods calculate users’ similarity to find the nearest neighbors without distinguishing their attributions, and the recommendation seems to be inefficient. A concept lattice based users’ interests prediction algorithm was presented as follows: first, formal context about the user-navigation was extracted from the user access logs, and the concept lattice was built from it; second, an appropriate sliding window was used to limit the user's current access content, thus could identify the user's current independent preferences; at last,the recommendation utilities of documents were calculated according to the independently preferences, and the weighted sum was used to get the personal preferences reflect by all the current interests to predict users’ interests. This method analyzed the issue of classification between documents in the traditional methods,thus users’ preferences could be effectively identified and divided, which coincide in characteristics that users were similar only in certain aspects, but not all the features of interests. Experiment on real log data proved the effectiveness in resource recommendation, and the cold start problem in the traditional collaborative filtering methods could be smoothed.
     

    An impulse noise filtering algorithm based on a robust neuro-fuzzy network
    LI Yue-Yang, WANG Shi-Tong
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  164-170. 
    Abstract ( 229 )   PDF (1285KB) ( 1202 )   Save
    Related Articles | Metrics

     Based on an integration of a simple impulse detector and a robust neurofuzzy (RNF) network, an effective impulse noise filtering algorithm for color images is presented. It consists of two modes of operation, namely, training and testing (filtering). During training, the impulse detector is used to locate the noisy pixels in the color images for optimizing the RNF network. During testing, if a pixel is detected as a corrupted one according to the impulse detector, the trained RNF network will be triggered to output a new pixel to replace it. The proposed impulse noise filtering algorithm is distinguished by two properties. The first is the use of a simple impulse noise detector, which is efficient and yet effective in detecting the noise pixels in color images. The other is the use of a novel membership function in the design of the adaptive RNF network, making the network robust to impulse noise. As demonstrated by the experimental results, the proposed filter not only has the abilities of noise attenuation and details preservation but also possesses desirable robustness and adaptive capabilities. It outperforms other conventional multichannel filers.

    CAN2:component-assembled neural network
    WU He-sheng1,2, WANG Chong-jun1,2, XIE Jun-yuan1,2
    JOURNAL OF SHANDONG UNIVERSITY (ENGINEERING SCIENCE). 2010, 40(5):  171-178. 
    Abstract ( 251 )   PDF (1528KB) ( 1123 )   Save
    Related Articles | Metrics

    Engineering neurocomputing, as an effective approach to boost intelligent computing technology, focus a puzzle: the “black box” property of neural network. It means that knowledge learning from neural network implicate the vast connected weights. User can’t understand what the neural network learn and what task the neural network can deal with. And what’s more, user can’t know how the neural network predicts and why the neural network reasons these or those conclusions. In order to solve effectively this puzzle, componentassembled neural network (CAN2) is proposed. Based CAN2 technology, We construct comprehensible and reused digital logic neurocomponent library(DLNL). Complex digital logic function is implemented and random classification problems is solved by applying DLNL. Experiment indicates that CAN2 can reduce the “black box” property of neural network effectively and has powerful reusability. It is an effective attempt in engineering neurocomputing, can improve user confidence for constructing intelligent system by applying neural network.