Started in January,1974(Monthly)
Supervised and Sponsored by Chongqing Southwest Information Co., Ltd.
ISSN 1002-137X
CN 50-1075/TP
Current Issue
Volume 45 Issue 3, 15 March 2018
Unique Curriculums for Data Science and Big Data Technology
CHAO Le-men, XING Chun-xiao and WANG Yu-qing
Computer Science. 2018, 45 (3): 1-8.  doi:10.11896/j.issn.1002-137X.2018.03.001
Abstract PDF(1656KB) ( 1316 )   
References | Related Articles | Metrics
How to construct a novel major,called Data Science and Big Data Technology,is one of the hot topics in China.An in-depth analysis on typical universities was conducted including University of California,Berkeley,Johns Hopkins,Washington University,New York University,Stanford University,Carnegie Mellon University,Columbia University,and City University of London.And then,ten core courses of Data Science and Big Data Technology were identified and described.Finally,eight common misunderstandings in existing data science curriculum were proposed,and the solutions were provided respectively.
Survey of Graph Modification Problems Related to Specific Graphs
KE Yu-ping and WANG Jian-xin
Computer Science. 2018, 45 (3): 9-15.  doi:10.11896/j.issn.1002-137X.2018.03.002
Abstract PDF(1310KB) ( 611 )   
References | Related Articles | Metrics
Graph modification problems refer to deleting or adding edges or vertices in a graph to make a graph transform to another graph with a certain property.Graph modification problems have been widely studied for many years,especially on chordal graphs,interval graphs and unit interval graphs.Chordal graphs are the most important perfect graph class and supersets of (unit) interval graphs.There are many NP-hard problems which can be solved in polynomialtime on chordal graphs.Interval graphs and unit interval graphs have momentous application on computational biology.Research on graph modification problems of these graphs make a great contribution to both computer theory and applications.This survey first summarized important results for the graph modification problems on chordal graph,interval graph and unit interval graphs,then analyzed these problems,and provided some open problems to be worth studying.
Survey on Cognitive Domain Feature Prediction of Social Network Users
ZHENG Jing-hua, GUO Shi-ze, GAO Liang and ZHONG Xiao-feng
Computer Science. 2018, 45 (3): 16-22.  doi:10.11896/j.issn.1002-137X.2018.03.003
Abstract PDF(1304KB) ( 1399 )   
References | Related Articles | Metrics
Cognitive domain security is of primary importance in cyber space security,and cognitive domain characteristic prediction is the basis of researching on cognitive domain security.This paper showed a kind of cyber space cognitive domain safety model.Then,this paper is clear that cognitive domain characteristic prediction of social network users has a vital role in cyber space cognitive domain security.It summarized home and abroad researches about the prediction of social network users from three aspects:prediction process,characteristics choosing and model building.Aiming at characteristics of domestic typical social network users samples,this paper pointed out the problems existing in researches,and showed some possible thoughts and methods.At last,this paper summed up current challenges and shortages in this field including some related problems to be solved as stress.
Survey on Converting Image to Sentence Based on Depth Neural Networks
MAO Dian-hui, XUE Zi-yu, LI Zi-qin and WANG Fan
Computer Science. 2018, 45 (3): 23-28.  doi:10.11896/j.issn.1002-137X.2018.03.004
Abstract PDF(8555KB) ( 606 )   
References | Related Articles | Metrics
In the context of big data,the number of images increases rapidly,and knowledge acquisition is of great significance to the use and analysis of images.Image semantic analysis method based on deep model is a technique which can convert image content into intuitive understandable semantic knowledge through deep model,attracting wide attention at home and abroad.The target of image semantic analysis method can be divided into phrases,multiple tags,and statements.This paper introduced the research status of the above methods and their advantages,and analyzed the features of the image during the process of knowledge acquisition and the existing problems,including the structural features of convolutional neural network and the recurrent neural network.From the aspects such as model structure and connection,this paper analyzed the research hotspot and the cases,then analyzed the differences between academia and industry,and adopted image sentence conversion to excute a discriminant comparison.Finally,this paper drew a conclusion and gave its hope for the images semantic analysis method with deep model.
Interval Gradient Based Joint Bilateral Filtering for Image Texture Removal
WEI Ming-qiang, FENG Yi-dan, WANG Wei-ming, XIE Hao-ran and WANG Fu-li
Computer Science. 2018, 45 (3): 29-34.  doi:10.11896/j.issn.1002-137X.2018.03.005
Abstract PDF(10032KB) ( 882 )   
References | Related Articles | Metrics
Image texture removal is a fundamental problem in image processing.It aims to decompose an image into texture patterns and structure features.Many filters have been proposed for removing image textures.However,these techniques suffer from some problems in balancing the performance among texture distinction,texture removal and time efficiency.In this paper,an interval gradient-based filter was proposed to remove image textures.First,to simplify the prior model,interval gradient is employed to extract the structure features.And the binary map that separates the structures from texture is achieved,which will be used as a guidance image when the filtering texture regions are processed.Then,to deal with complicated texture patterns, the shift smoothing technique is incorporated into joint bilateral filtering,and the pixel possessing the maximal color difference with the target pixel is selected as the center point for color weight distribution,so that it can dominate the filtering process.Experiments show that the proposed method can be applied to various types of texture images,achieving both effective texture removal and high time efficiency.Moreover,this method can better preserve the edge features attributing to fewer iteration times required for obtaining similar results of texture removal. 〖BHDWG1,WK32,WK44,WK42W〗第3期 魏明强 ,等:基于区间梯度的联合双边滤波图像纹理去除方法
Rational Fractal Surface Modeling and Its Application in Image Super-resolution
LIU Tian-tian, BAO Fang-xun, ZHANG Yun-feng, FAN Qing-lan and YANG Xiao-mei
Computer Science. 2018, 45 (3): 35-45.  doi:10.11896/j.issn.1002-137X.2018.03.006
Abstract PDF(2731KB) ( 679 )   
References | Related Articles | Metrics
The construction of surfaces is a key issue in computer aided geometric design.In order to make the constructed surfaces more flexible and effective in practical application,this paper proposed a constructive method of rational fractal surfaces and a single image super-resolution reconstruction algorithm based on this model.Firstly,a bivariate rational spline iterated function system was presented in which the fractal interpolation functions are regarded as the fractal perturbation of height functions,generating rational fractal surfaces.Secondly,some analytical properties of the ratio-nal fractal functions were investigated,and the box-counting dimension of fractal surfaces was obtained.Finally,a super-resolution reconstruction algorithm of single image was proposed based on the model and its theoretical results.In this algorithm,the image is divided into edge region and non-edge region by using the Non-subsampled Contourlet Transform (NSCT).And then,the scaling factors are accurately calculated by the dimension formula,the shape parameters are determined based on maintaining the structure similarity of the image,and different models are selected in different regions to interpolate the image data.Rational fractal interpolation and rational interpolation are used in edge region and non-edge region respectively.Next,the target image is obtained by a proper transformation.The experimental results show the effectiveness of the model and algorithm.The presented method is better in maintaining texture details and edge information than the compared algorithms,especially,it achieves competitive performance for preserving the structure information of image,and obtains good objective evaluation data and subjective visual effects. 〖BHDWG1,WK32,WK44,WK42W〗第3期 刘甜甜 ,等:有理分形曲面造型及其在图像超分辨中的应用
A New Kind of Parametric Curves by Special Basis Function
LI Jing-gai, CHEN Qiu-yang, HAN Jia-qi, HUANG Qi-li and ZHU Chun-gang
Computer Science. 2018, 45 (3): 46-50.  doi:10.11896/j.issn.1002-137X.2018.03.007
Abstract PDF(1295KB) ( 820 )   
References | Related Articles | Metrics
The construction of parametric curves and surfaces is very important in computer aided geometric design.It’s well known that Bézier curve,which is defined by Bernstein basis functions is a basic method in curve design,and the B-spline curve and NURBS curve are generalizations of the Bézier curve .This paper defined a new kind of basis functions by a given real knot points set,which is a generalization of Bernstein basis functions,and defined a new parametric curve by these basis functions,called T-Bézier curve ,which preserves some properties of Bézier curve.What’s more,this paper presented the limit property of T-Bézier curve while some knots move and gave some examples to verify the properties of the curve.
Interpolatory Subdivision Schemes for Mixed Higher-order Exponential Polynomials Reproduction
LI Zhao-hong, ZHENG Hong-chan, LIAN Hui-fen and JIN Ming-ya
Computer Science. 2018, 45 (3): 51-57.  doi:10.11896/j.issn.1002-137X.2018.03.008
Abstract PDF(1359KB) ( 638 )   
References | Related Articles | Metrics
By introducing new shape control parameter,this paper presented a family of unified interpolatory subdivision schemes which can accurately interpolate mixed and high-order exponential polynomials.The basic idea is to obtain new interpolatory subdivision schemes with the same spare reproducing through generating exponential B-spline subdivision schemes exponential polynomial space.These schemes have smaller support and greater freedom degree than other schemes with the same reproduction.This paper analyzed the reproduction property of the interpolatory schemes in theo-ry.Finally,the influence of initial shape control parameters and free parameters on the limit curve was analyzed.For specific initial control parameter,the presented schemes can be used to reproduce some special curves which are represented by mixed and high-order exponential polynomials.This paper further showed that variety of schemes will be obtained by choosing different free parameters.
Illumination Parameter Estimation of Outdoor Scene Using Chromaticity Consistency
ZHANG Rui, HAN Hui-jian, LIANG Xiu-xia, FANG Jing and ZHANG Cai-ming
Computer Science. 2018, 45 (3): 58-62.  doi:10.11896/j.issn.1002-137X.2018.03.009
Abstract PDF(1348KB) ( 910 )   
References | Related Articles | Metrics
For the outdoor scene images shot in the same solar azimuth under different weather conditions,this paper proposed an algorithm to estimate the illumination parameters using the chromaticity consistency.In this algorithm,based on the basis image decomposition,the chromaticity consistency is used to solve the illumination parameters of outdoor scenes.And then,according to the illumination chromaticity correction model,the illumination parameters are optimized.The experimental results show that the algorithm is effective and correct,and can accurately reconstruct the origi-nal image according to the base images and the illumination parameters,so as to realize the seamless integration between the virtual object and the real scene.
Reparameterization-based Clipping Method for Root-finding Problem of Non-linear Equations
JIN Jia-pei, CHEN Xiao-diao, SHI Jia-er and CHEN Li-geng
Computer Science. 2018, 45 (3): 63-66.  doi:10.11896/j.issn.1002-137X.2018.03.010
Abstract PDF(1233KB) ( 615 )   
References | Related Articles | Metrics
The root-finding problem of non-linear equations has wide applications in computer aided geometric design,computer graphics,robotics,etc.This paper presented a reparameterization-based cubic clipping method for finding the roots of a non-linear equation. Firstly,it computes a cubic polynomial interpolating the given smooth function f(t) at four points.Then,it searches two reparameterization functions so that the reparameterized functions have the same derivatives,which leads to higher approximation order and convergence rate.Compared with the prevailing cubic clipping methods,the new method can achieve the convergence rate 9 or higher for single root cases,and directly bound the root without computing the bounding polynomials.Numerical examples show that it can converge to the proper solution even in some cases that Newton’s methods fail.
Birational Trilinear Mapping on Convex Hexahedrons
YE Jin-yun, WANG Xu-hui and QIAN Yi-jia
Computer Science. 2018, 45 (3): 67-68.  doi:10.11896/j.issn.1002-137X.2018.03.011
Abstract PDF(1203KB) ( 728 )   
References | Related Articles | Metrics
Based on the knowledge of three dimensional generalized barycentric coordinates,the results of birational trilinear mapping on planar quadrilateral were generalized to three dimensional convex hexahedron,namely,by assigning a suitable weight to every vertex of convex hexahedron,a three dimensional birational mapping was achieved.In addition,an example was given to illustrate the correctness of this method.
Discriminative Visual Tracking by Collaborative Structural Sparse Reconstruction
YOU Si-si, YING Long, GUO Wen, DING Xin-miao and HUA Zhen
Computer Science. 2018, 45 (3): 69-75.  doi:10.11896/j.issn.1002-137X.2018.03.012
Abstract PDF(11690KB) ( 667 )   
References | Related Articles | Metrics
Though the appearance likelihood model based on sparse representation has been widely applied in visual tracking,the single generation object representation model can easily be interrupted by background clutter due to not considering the full discriminative structural information.In order to alleviate the drift problem of the visual tracking,this paper presented a novel tracking method based on collaborative sparse reconstruction of object appearance dictionary and background dictionary.This paper achieved a more accurate description of the target appearance model by constructing a discriminative appearance likelihood model based on sparse representation.Then,it embedded discriminative information into the appearance likelihood model by a reasonable method of selecting the sparse coefficients of candidate target region and candidate background region.By that way,it can explore the potential correlation of candidate target region and the structure relation of candidate background region,so as to learn the appearance model of candidate target area more accurately.Many experimental results in challenging sequence verify the robustness of this method.The proposed tracker outperforms excellent performance in comparison with other state-of-the-art trackers.
Selection of Control Points of Quadratic-trigonometric Hermite Interpolation Splines
LIU Cheng-zhi, HAN Xu-li and LI Jun-cheng
Computer Science. 2018, 45 (3): 76-82.  doi:10.11896/j.issn.1002-137X.2018.03.013
Abstract PDF(1341KB) ( 739 )   
References | Related Articles | Metrics
This paper studied the selection of the free control points of the C1 continuous quadratic-trigonometric Hermite interpolation curves.Firstly,this paper discussed the selection of the free control points when the conditions of midpoint were given.In order to obtain the most smooth or the shortest arc length interpolation curves,an optimization model for solving the optimal control points was established based on the energy optimization method.By solving the optimization model,the optimal control points were obtained to minimize energy value of the curve.Then,an optimization model was also established for solving the shortest arc length.Numerical examples show that the optimal control points can make the curves smooth or have the shortest arc length.
Bivariate Non-tensor-product-typed Continued Fraction Interpolation
QIAN Jiang, WANG Fan and GUO Qing-jie
Computer Science. 2018, 45 (3): 83-91.  doi:10.11896/j.issn.1002-137X.2018.03.014
Abstract PDF(3845KB) ( 584 )   
References | Related Articles | Metrics
Based on the new recursive algorithms of bivariate non-tensor-product-typed inverse divided differences,the scattered data interpolating schemes via bivariate continued fractions were established in the case of odd and even interpolating nodes,respectively.Then two equivalent identities of the interpolated function were obtained.Moreover,by means of the three-term recurrence relations,the degrees of the numerators and denominators were determined,i.e.,the characterization theorem,so do the corresponding recursive algorithms.Meanwhile,compared with the degrees of the numerators and denominators of the well-known bivariate Thiele-typed interpolating continued fractions,those of the presented bivariate rational interpolating functions are much lower respectively,due to the reduction of redundant interpolating nodes.Furthermore,the operation count for the rational function interpolation is smaller than that of radial basis function interpolation from the aspect of complexity of the operations.Finally,some numerical examples show that it’s valid for the recursive continued fraction interpolation,and imply that these interpolating continued fractions change as the order of the interpolating nodes change,although the node collection is invariant.
Study on WSN Node Localization Technology for Environment Monitoring
YANG Pei-ru and XUE Shan-liang
Computer Science. 2018, 45 (3): 92-97.  doi:10.11896/j.issn.1002-137X.2018.03.015
Abstract PDF(6393KB) ( 683 )   
References | Related Articles | Metrics
WSN node localization is of great significance in wireless sensor network research.The data collected by sensor nodes with node location is more meaningful.Combining environmentent monitoring features and application requirements ,DV-Hop (Distance Vector-Hop) algorithm is suitable in environmentent monitoring scene because of its small environmentent impact and little hardware overhead.Aiming at the shortage of traditional DV-Hop algorithm,this paper proposed a hybrid DV-Hop algorithm based on weighted factor,called HDV-Hopw,which is improved by two methods.Firstly,the average per-hop distance of beacon node is weighted to reduce the error caused by the average distance per hop.Then,position estimation of unknown nodes is transformed into objective optimization,and GA-PSO algorithm is used to optimize the coordinates of unknown nodes.The feasible region of the initial population is restricted and initial population is improved to improve the position accuracy of the algorithm.The simulation results show that compared with the DV-Hop algorithm,the localization error of HDV-Hopw is reduced by about 11% without increasing the hardware cost.
Tradeoff Optimization of Spectrum Opportunity Discovery and Licensed User Protection in Dynamic Spectrum Management
TIAN Jia-qiang, CHEN Yong and ZHANG Jian-zhao
Computer Science. 2018, 45 (3): 98-101.  doi:10.11896/j.issn.1002-137X.2018.03.016
Abstract PDF(5869KB) ( 633 )   
References | Related Articles | Metrics
In dynamic spectrum management based on cognitive radio,spectrum sensing is desired to explore more spectrum opportunity while incurring less interference to licensed users.This paper investigated the tradeoff optimization of the two performance metrics,and constructed a joint optimization model in which the weighted sum of two metrics is regarded as objective function,and sensing duration and sensing threshold are regarded as variables. This problem is proved to be in the form of biconcave optimization problem (BOP).An optimization algorithm based on alternative convex search was proposed,which can quickly find the near optimal solutions without relying on the predefined sensing parameters.Simulation results demonstrate that the joint parameters optimization scheme generates 32.0% and 85.9% promotion over single parameter optimization schemes on average.
Optimization of Co-resident Inter-VM Communication Accelerator XenVMC Based on Multi-core
YOU Zi-qi, REN Yi, LIU Ren-shi, GUAN Jian-bo and LIU Li-peng
Computer Science. 2018, 45 (3): 102-107.  doi:10.11896/j.issn.1002-137X.2018.03.017
Abstract PDF(2515KB) ( 656 )   
References | Related Articles | Metrics
Nowadays,on virtualized platform,using shared memory channels to accelerate communication between virtualmachines(VMs) located on the same physical machine is a wildly used solution.XenVMC is such a project,which is implemented with high efficiency,full transparency and VM live migration supported.With the development of multi-core technology,XenVMC can be improved further.This paper proposed a multi-core optimization solution on XenVMC based on its individual communication model.By scheduling other CPUs in receiving VM,and updating the design of shared memory channels,XenVMC can receive data concurrently with mutli-cores.Experiment results illustrate that connection transaction throughout is improved obviously and communication throughout is also improved in some cases with multi-core optimization.
Wireless Charging Scheduling Algorithm of Single Mobile Vehicle with Limited Energy
XU Xin-li, CHEN Chen, HUANGFU Xiao-jie and CUI Yong-ting
Computer Science. 2018, 45 (3): 108-114.  doi:10.11896/j.issn.1002-137X.2018.03.018
Abstract PDF(6209KB) ( 691 )   
References | Related Articles | Metrics
Multi-node charging technology based on magnetic resonance coupling provides a potential solution for the robustness of wireless sensor networks.Considering the charging efficiency of magnetic resonance coupling,cellularstructure was adopted to divide the network into several charging areas,and a charging scheduling algorithm in wireless sensor network based on mobile charger was proposed,so as to reduce the energy consumption of charging devices and guarantee schedulability of charging plans.A self-adaptive dynamic algorithm for automatically selecting k charging areaswas raised to comprehensively consider the energy of mobile charger,residual energy of nodes and other factors du-ring each charging period.When planning the route,elastic network algorithm with good timeliness was adopted to meet demands.The simulation results show that the energy of mobile charger has direct impact on total energy and minimum residual energy of networks.When the energy of mobile charger is limited,the proposed algorithm can maximize minimum network energy and extend the life cycle of the network.
Non-uniform Hierarchical Routing Protocol Based on New Clustering for Wireless Sensor Network
TAO Zhi-yong and WANG He-zhang
Computer Science. 2018, 45 (3): 115-123.  doi:10.11896/j.issn.1002-137X.2018.03.019
Abstract PDF(1378KB) ( 675 )   
References | Related Articles | Metrics
Aiming at the problem of energy hole caused by uneven load energy in clustering routing protocol,a non-uniform hierarchical routing protocol based on new clustering for load balancing(NHRPNC) was proposed.Firstly,the algorithm utilizes the improved threshold function of LEACH protocol to select the region-heads and make the reasonable non-uniform partitions of the network.Secondly,the region-heads use the new clustering algorithm to achieve the non-uniform clustering in each region.Then,the four-step selection mechanism of cluster heads is adopted to select the cluster head in each cluster periodically.At last,for inter-cluster multi-hop communication,the multi-hop path is optimized by dynamic weight.Simulation results reveal that compared with LEACH (Low Energy Adaptive Clustering Hierarchy) protocol,DEBUC (Distributed Energy Balanced Unequal Clustering routing) protocol and UCDP (Uneven Clustering based on Dynamic Partition) protocol,NHRPNC can promote the percentage point of 257.5,33.74 and 12.83 in the life cycle of the network respectively and has favorable performance in balancing the energy consumption.
Study on Malicious URL Detection Based on Threat Intelligence Platform
WANG Xin, WU Yang and LU Zhi-gang
Computer Science. 2018, 45 (3): 124-130.  doi:10.11896/j.issn.1002-137X.2018.03.020
Abstract PDF(3804KB) ( 1669 )   
References | Related Articles | Metrics
With Internet penetrating into daily life,it is hard to prevent ubiquitous malicious URLs,threatening the properties and privacies of people seriously.Traditional method to detect malicious URL relies on blacklist mechanism,but it can do nothing with the malicious URLs which are not in the list.Therefore,one of the fundamental directions is bringing in machine learning to optimize the malicious URL detection.However,the results of most existing solutions are not satisfying,as the characteristics of URL short text make it extract a single feature.To address those problems above,this paper designed a novel system to detect malicious URLs based on threat intelligence platform.The system extracts structural features,intelligence features and sensitive lexical features to train classifiers.Next,the voting me-chanism with results of multiple classifiers is exploited to determine the type of URLs.Finally,the threat intelligence can be updated automatically.The experimental results show that the method for detecting malicious URL has good de-tection effect,and is capable of achieving classification accuracy up to 96%.
0-1 Code Based Privacy-preserving Data Value Matching in Participatory Sensing
LIU Meng-jun, LIU Shu-bo and DING Yong-gang
Computer Science. 2018, 45 (3): 131-137.  doi:10.11896/j.issn.1002-137X.2018.03.021
Abstract PDF(1954KB) ( 553 )   
References | Related Articles | Metrics
In participatory sensing,protecting both the privacy of requestor and provider while satisfying the special requirement of data types and data value of data requesters at the same time,is a crucial problem before the widespread of participatory sensing application.This paper put forward a 0-1 encode based privacy-preserving data value matching scheme.It first converts two users’ data value into two 0-1 code sets,and then matches the two sets with a spatial-timing efficient data structure-bloom filter,thus preserving the privacy of data value while completing efficient data value matching.Theoretical analysis and simulation experiment prove the correctness,safety and effectiveness of the proposed scheme.
Research and Implementation of Light-weight Mandatory Access Control Technology for RTOS
YANG Xia, YANG Shan, GUO Wen-sheng, SUN Hai-yong, ZHAO Xiao-yan and ZHANG Yang
Computer Science. 2018, 45 (3): 138-143.  doi:10.11896/j.issn.1002-137X.2018.03.022
Abstract PDF(1340KB) ( 731 )   
References | Related Articles | Metrics
Mandatory access control(MAC) technology can control the accesses of all subjects to objects in the system,which is an effective method to enhance the system security.Implementing the mandatory access control mechanism in a real-time operating system(RTOS) can effectively improve the system security,and make the RTOS pass the high-level certification. Aiming at the problem that the real-time operating system has the characteristics of less resources,low overhead and hard real-time,this paper presented a light-weight mandatory access control (L-MAC) mechanism.The L-MAC technology consists of an L-MAC model,a configurable access monitor and a light-weight security policy model with task permission set based on DTE.Finally,this paper implemented a prototype system based on RTEMS system and a security policy tool that can conveniently add,modify or delete a security policy according to user’s requirements.The results of multiple tests about function and time overload show that L-MAC mechanism is effective and feasible.
BTDA:Dynamic Cloud Data Updating Audit Scheme Based on Semi-trusted Third Party
JIN Yu, CAI Chao, HE Heng and LI Peng
Computer Science. 2018, 45 (3): 144-150.  doi:10.11896/j.issn.1002-137X.2018.03.023
Abstract PDF(1565KB) ( 537 )   
References | Related Articles | Metrics
Cloud storage has been widely used since its birth because of its convenience and low price.But compared with the traditional system,the users in the cloud storage system lost the direct control of the data,so users are most concerned about whether the data stored in the cloud are security,where integrity is one of the security needs.Public audit is an effective way to verify the integrity of cloud data.Existing research work can not only achieve cloud data integrity verification,but also support dynamic data update audit.However,such schemes also suffer from some drawbacks,for example,when multiple second-level file block update tasks are implemented,users need to be online for the update audit of each task,and in this process the communication cost and the computational cost on user side are larger. On this basis,this paper proposed BTDA,namely a semi-trusted third party dynamic cloud data update audit program.In BTDA,semi-trusted third party deals with update audit instead of user,so during the update audit process,the user can be off-line,thereby reducing the communication cost and the computational cost on user side.In addition,BTDA uses data blind and proxy re-signature technology to prevent semi-trusted third party and cloud server to obtain user sensitive data,thus protecting user privacy.Experiments show that compared with the current scheme about second-level file block update,BTDA has a large reduction in both computation time and communication cost on user side.
Internal Consistency Preserving for Component Dynamic Evolution
ZHENG Ming, LI Tong, MO Qi, ZHOU Xiao-xuan, XIANG Wen-kun and HE Yun
Computer Science. 2018, 45 (3): 151-157.  doi:10.11896/j.issn.1002-137X.2018.03.024
Abstract PDF(4863KB) ( 577 )   
References | Related Articles | Metrics
Component-based software development has become the mainstream of the software development method.Aiming at the problem of consistency preserving after component-based software dynamic evolution,most scholars analyze and study the component-based software dynamic evolution mainly from the perspective of external consistency at pre-sent.For this reason,a new method was proposed to preserve the internal consistency of component-based software dynamic evolution from internal perspective.Firstly,the component and its correlation were modeled,and an algorithm was put forward for judging the homomorphism mapping relation of a class directed graphs.Secondly,the criterion of internal consistency of component-based software dynamic evolution was given,based on strong simulation theory in process algebra and homomorphism mapping in graph theory,the sufficient condition and necessary condition for the internal consistency criterion of component in the component-based software before and after dynamic evolution were defined from global and local perspective,and proof was carried out respectively lastly.Thirdly,on the basis of the above work,the process of internal consistency preserve way of component dynamic evolution was give.Finally,case study shows that the proposed approach is feasible and effective.
ORC Metadata Based Reducer Load Balancing Method for Hive Join Queries
WANG Hua-jin, LI Jian-hui, SHEN Zhi-hong and ZHOU Yuan-chun
Computer Science. 2018, 45 (3): 158-164.  doi:10.11896/j.issn.1002-137X.2018.03.025
Abstract PDF(1433KB) ( 659 )   
References | Related Articles | Metrics
The load imbalance problem ranks first among the performance issues in large-scale MapReduce cluster,and it’s very prone to be triggered by Hive join queries.An effective solution is to design reducer load balancing partitioning algorithms by consulting the key’s frequency distribution histogram estimated from intermediate key-value pairs.The existing works of key histogram estimation rely on monitoring and sampling the output of map in a distributed way,which triggers huge network traffic load and notably delays the start of the shuffle.A novel key histogram estimation method based on ORC metadata and the corresponding load balancing partitioning strategy was proposed for Hive join queries.The proposals only need some light-weight computation before the start of the job,thus imposing no extra loads on network traffics and the shuffle.Benchmarking test proves the proposal’s significant improvement on both the key histogram estimation and the reducer load balancing.
Location-awareness Publication Subscription System Based on Topic Model
XIAN Xue-feng, CUI Zhi-ming, ZHAO Peng-peng, LIU Zhao-bin and GU Cai-dong
Computer Science. 2018, 45 (3): 165-170.  doi:10.11896/j.issn.1002-137X.2018.03.026
Abstract PDF(1317KB) ( 612 )   
References | Related Articles | Metrics
Location-awareness publication subscription system has drawn extensive academic and industrial attention with the booming development of mobile Internet and the popularity of smart-phones.The existing systems on location-awareness publication/subscription mainly focus on handling the query and matching problem of events among massive spatial data,whose matching model is mainly based upon the similarities of spatial keywords,while the semantic aspect is ignored.In order to explore how to realize the semantic query and matching in subscription/publication system,this paper proposed a location-awareness publication/subscription system based upon theme model.Firstly,the system makes use of theme model algorithm and realizes the thematic reflection of keywords in location-awareness publication/subscription system.Secondly,it designs a two-step partition index structure RPTM-trees and utilizes RPTM-trees to createan index between thematic aggregation and spatial information.As RPTM-trees conducts a two-step partitioning and indexing of the subscription information based on the topic numbers of thematic aggregation and key topics,a stronger subscription partitioning ability is achieved,and the efficiency of query and matching is significantly improved.Finally,an experiment on high-speed event stream and millions and millions subscription data aggregation was conducted,indicating the effectiveness and the efficiency of the proposed solution.
In-memory B+tree Construction Methodology for Big Data Stream
YANG Liang-huai, XIANG Jun-jian, XU Wei and FAN Yu-lei
Computer Science. 2018, 45 (3): 171-177.  doi:10.11896/j.issn.1002-137X.2018.03.027
Abstract PDF(3017KB) ( 734 )   
References | Related Articles | Metrics
This paper investigated into the issues of indexing on data stream with time dimension in near real-time.By resorting to 2-tier B+tree index,this paper invented a highly effective in-memory B+tree construction method for scenarios with real-time query requirements,which separates as many parallelizing operations as possible.This paper parallelized the operations of sorting and data receiving by dividing the time-window into equal-duration slice,and parallelized the construction of B+tree skeleton with sorting.This paper avoided unnecessary locking and synchronizing cost by adopting sorting-based bulk loading techniques and optimized constructing sequence.The proposed in-memory B+tree construction algorithm called MBSortSBLoad can build B+tree quickly and accept higher data arriving rates.Extensive experiments demonstrate the effectiveness of the proposed methods.
k-step Reachability Queries Based on Bidirectional Double Interval Labeling Indexes
SONG Ya-qing, WU You-xi, LIU Jing-yu and LI Yan
Computer Science. 2018, 45 (3): 178-181.  doi:10.11896/j.issn.1002-137X.2018.03.028
Abstract PDF(1252KB) ( 559 )   
References | Related Articles | Metrics
Recently,reachability query is one of the main research topics on graph data.GRAIL can deal with k-step reachability queries efficiently,however,it is not suitable for processing the query in which the vertex pairs are located in different branches.This paper further proposed RE-GRAIL algorithm which employs a bidirectional double interval labeling indexes to tackle the problem.At last,five real datasets were employed to validate the performances of the proposed algorithm in terms of different metrics,including indexing time index size,query processing time and scalability.Experimental results show that RE-GRAIL has better performance than other competitive algorithms.
Study on K-line Patterns’ Profitability Based on Similarity Match and Clustering
LV Tao and HAO Yong-tao
Computer Science. 2018, 45 (3): 182-188.  doi:10.11896/j.issn.1002-137X.2018.03.029
Abstract PDF(1313KB) ( 2020 )   
References | Related Articles | Metrics
K-line pattern is the most popular technical analysis method for short term stock investment.However,there are some disputes about the K-line patterns’ profitability in academia.To resolve the debate,this paper used the method of pattern recognition,pattern clustering and pattern knowledge mining to study the profitability of K-line patterns.Therefore,firstly,the similarity match model was proposed for solving the problem of similarity match of K-line pattern.Secondly,the nearest neighbor-clustering algorithm was proposed for solving the problem of clustering of K-line pattern.Finally,the measurement model of K-line pattern’s profitability was defined for measuring the profitability of a K-line pattern’s different shapes.In the experiment,the profitability of three white soldiers pattern and three black crows pattern was analyzed with the testing dataset of the K-line series data of Shanghai 180 index component stocks over the latest 11 years.Experimental results show that the main reason for the debate is that the profitability of one pattern varies a great deal for different shapes and they are even opposite at sometimes.There is a need to further classify each of the existing K-line patterns based on the shape feature and give their strict mathematical definitions for improving the profitability and resolving the disputes.
Ensemble Multi-label Classification Algorithm Based on Tree-Bayesian Network
ZHANG Zhi-dong, WANG Zhi-hai, LIU Hai-yang and SUN Yan-ge
Computer Science. 2018, 45 (3): 189-195.  doi:10.11896/j.issn.1002-137X.2018.03.030
Abstract PDF(549KB) ( 578 )   
References | Related Articles | Metrics
The performance of learning algorithm can be improved by utilizing existing label dependencies in multi-label classification.Based on the strategy of classifier chain and stacking ensemble learning,this paper built a model to explain the dependency of different labels,and extended the linear dependency into tree dependency to deal with much more complicated label relations.Compared with the original Stacking algorithm,the performance of the proposed algorithm is improved in the experiments.
Activity Recommendation Algorithm Based on Latent Friendships in EBSN
YU Ya-xin and ZHANG Hai-jun
Computer Science. 2018, 45 (3): 196-203.  doi:10.11896/j.issn.1002-137X.2018.03.031
Abstract PDF(7135KB) ( 485 )   
References | Related Articles | Metrics
EBSN(Event-based Social Networks) not only contain online social interactions in the conventional online social networks,but also include valuable offline social interactions captured in offline activities,possessing a complicated heterogeneous nature.How to use the coexistence of online and offline social interactions to improve the service quality has become a hot issue in both academic and industry domains.Besides considering basic attributes from events and users, most traditional social activity recommendation approaches recommend interesting activities to users based on explicit friendships.However,there is no explicit friendships in EBSN,which makes these traditional algorithms can’t be applied to EBSN’s event recommendation directly.In this light,a novel concept LF (Latent Friendship) was defined in this paper.LF not only takes into account the online same group relationships,but also considers the offline same activity relationships.Further,ARLF(Activity Recommendation Algorithm based on Latent Friendship)was proposed by synthesizing the influence of group and activity for activity recommendation.Meanwhile,this paper creatively applied the idea of Meta-Path to capture the latent friends,which exploits the heterogeneous information fully in EBSN.Finally,extensive experiments based on real data of Meetup show that ARLF is feasible and effective on recommending desirable and interesting activities for EBSN users.
Service Clustering Approach for Global Social Service Network
LU Jia-wei, MA Jun, ZHANG Yuan-ming and XIAO Gang
Computer Science. 2018, 45 (3): 204-212.  doi:10.11896/j.issn.1002-137X.2018.03.032
Abstract PDF(1434KB) ( 620 )   
References | Related Articles | Metrics
The existing service clustering approaches mainly focus on functionality or QoS attribute,and they are lack of considering the social attribute in services.The growing number of Web services brings about a series problems of reducing efficiency of service discovery.Thus,this paper proposed a new service clustering approach for global social ser-vice network which can connect the isolated service into a social network.First,the similarity of services is calculated according to descriptive information,tag of domain area and QoS attribute in REST and SOAP service.Second,similarity calculations are clustered by combining with social attribute to enhance the services’ sociability on a global scale.At last,service visualization of global social service network is given to show the social relationships among realted servi-ces.The experimental result shows the effectiveness of the proposed method.
Study on Collaborative Filtering Algorithm Based on User Interest Change and Comment
DONG Chen-lu and KE Xin-sheng
Computer Science. 2018, 45 (3): 213-217.  doi:10.11896/j.issn.1002-137X.2018.03.033
Abstract PDF(1283KB) ( 695 )   
References | Related Articles | Metrics
The user-item rating matrix is becoming more and more sparse with the increasing number of users and commodities in the traditional collaborative filtering algorithm.To solve this problem,a collaborative filtering algorithm based on user interest change and comment was proposed.The algorithm introduces user comment and forgetting curve into the traditional collaborative filtering algorithm.The comment text is used as the text of commodity feature description,the topic model is used to calculate the commodity topic features,and Ebbinghaus’s forgetting curve is also introduced for the cooperative computing of user comment distribution and comment similarity.The similarity of user comment and the similarity of user rating are combined to get the final similarity,and then the rating of commodity is predicted.The algorithm was validated by real data crawled over the network.The experimental results show that the proposed algorithm can get better recommendation results in sparse data sets than the traditional collaborative filtering algorithm.
Friend Recommendation Method Based on Users’ Latent Features in Social Networks
XIAO Ying-yuan and ZHANG Hong-yu
Computer Science. 2018, 45 (3): 218-222.  doi:10.11896/j.issn.1002-137X.2018.03.034
Abstract PDF(2309KB) ( 904 )   
References | Related Articles | Metrics
With the popularity of social networks,such as Facebook,Twitter and Microblog,friend recommendation systems have gradually become an important part of social networks.Friend recommendation systems effectively expand the scale of user’s social circle and improve user’s social experience by actively recommending new potential friends for users,thus receiving widespread attention.However,how to personalize the user’s needs and recommend realfriends to users has been one of the difficulties for personalized friend recommendation.This paper presented a social networking friend recommendation method based on users’ latent features,called SNFRLF.SNFRLF first leverages latent factor model to mine users’ latent features,and then calculates the similarity between users by means of users’ latent features.Finally,the similarity is introduced into the random walk model to get a recommended list.The experimental results show that the proposed method significantly outperforms the existing friend recommendation methods.
Spatial-Temporal Co-occurrence Pattern Mining Algorithm Based on Network
ZHANG Yong-mei, GUO Sha, JI Yan, MA Li and ZHANG Rui
Computer Science. 2018, 45 (3): 223-230.  doi:10.11896/j.issn.1002-137X.2018.03.035
Abstract PDF(1378KB) ( 720 )   
References | Related Articles | Metrics
Most databases cannot effectively deal with time dimension of data,the spatial-temporal co-occurrence pattern mining is helpful to extract implicit valuable information from large spatio-temporal dataset,and it has become a hot research topic at present.To overcome lower mining efficiency of current co-occurrence pattern discovery methods,a double-level network model was used to initialize spatio-temporal dataset.In the calculation of spatial-temporal interes-tingness,traditional methods ignore the fact that every object-type has effective lifecycle.Thus,the current computation of interestingness was improved in this paper.We introduced weight eigenvalue and proposed a new spatial-temporal co-occurrence pattern mining algorithm based on network.Experiment results show that the proposed algorithm is more effective to calculate co-occurrence patterns in test sets with different data volumes than the methods without modeling or modeling instance layer only.
Parallel PSO Container Packing Algorithm with Adaptive Weight
LIAO Xing, YUAN Jing-ling and CHEN Min-cheng
Computer Science. 2018, 45 (3): 231-234.  doi:10.11896/j.issn.1002-137X.2018.03.036
Abstract PDF(1261KB) ( 935 )   
References | Related Articles | Metrics
With the arrival of intelligent manufacturing,the intelligent packing of product or container in the late production line has become an important part of industrial production,and how to get the packing results faster is also important for improving the production efficiency.Mainly aiming at the rapid packing,this paper proposed an intelligent packing algorithm for industrial production line.The algorithm uses the adaptive weight method to improve the particle swarm optimization algorithm,which has a faster convergence rate than the traditional heuristic algorithm,such as standard particle swarm optimization algorithm and genetic algorithm.The calculation speed is greatly accelerated by achieving high performance parallel computing with GPU acceleration.Experiments show that the algorithm proposed in this paper can also get very high space utilization rate,and its convergence speed is faster than the traditional algorithm.
Uncertain Vehicle Intersection Trajectory Prediction
MAO Ying-chi and CHEN Yang
Computer Science. 2018, 45 (3): 235-240.  doi:10.11896/j.issn.1002-137X.2018.03.037
Abstract PDF(2569KB) ( 872 )   
References | Related Articles | Metrics
In the city road,real-time,accurate and reliable trajectory prediction of mobile vehicles can bring very high application value,which can not only provide accurate location-based services,but also can help the vehicle to predict the traffic situation.At present,the trajectory prediction method of moving vehicles is mainly based on the precise historical trajectory in Euclidean space,and does not consider the vehicle trajectory prediction with uncertain historical data in restricted road network.A trajectory prediction method based on Markov chain was proposed to solve this problem.Its advantages include redefining the path algorithm of completion,making up for the incompleteness of uncertain historical data,and achieving prediction by using the characteristics of low time complexity and high prediction accuracy with Markov chain.This method avoids the problem of low prediction accuracy caused by too much query time due to the frequent pattern mining and the excess noise.The results show that under the same parameter setting,the prediction accuracy of the method is 18.8% higher than that of the mining frequent trajectory model,and the prediction time is reduced by 80.4% on average.Therefore,the method has high prediction accuracy for the trajectory prediction of the vehicle intersection,and achieves the prediction of a series of vehicle future trajectories.
Change Detection of Multiple Sclerosis in Brain Based on Multi-modal Local Steering Kernel
GUO Yang and QIN Pin-le
Computer Science. 2018, 45 (3): 241-246.  doi:10.11896/j.issn.1002-137X.2018.03.038
Abstract PDF(2053KB) ( 571 )   
References | Related Articles | Metrics
Volume effect and artifact are important influence factors in MR image processing and single-modal methods can be easily affected.This paper proposed an improved method based on multi-modal local steering kernel to detect the multiple sclerosis in the brain.This method utilizes multi-modal brain MR images and the approximate symmetry of the brain for change detection of the brain.Local steering kernel can measure the similarity between pixels and their surroundings.The proposed method takes the local steering kernel as the feature and measures the dissimilarity by cosine similarity.The experimental results show that the introduction of multi-modal reduces the volume effect and artifact in the MRI,improving the detection effect.
Automatic Recognition of Breast Gland Based on Two-step Clustering and Random Forest
WANG Shuai, LIU Juan, BI Yao-yao, CHEN Zhe, ZHENG Qun-hua and DUAN Hui-fang
Computer Science. 2018, 45 (3): 247-252.  doi:10.11896/j.issn.1002-137X.2018.03.039
Abstract PDF(3456KB) ( 834 )   
References | Related Articles | Metrics
Automatic recognition of the glands is critical in the histopathology diagnosis of breast cancer,as glandular density is an important factor in breast cancer grading.The gland is composed of a central lumen filled with cytoplasm and a ring of nuclei around the lumen.The spatial proximity of the lumen,cytoplasm,and nucleus may mean that it is a gland,but this method can lead to false-positive errors due to the presence of fat,bubbles and other lumen-like objects in the breast tissue section.In order to solve the above problems,this paper presented an automatic recognition method of breast gland based on two-step clustering and random forest.First,the images to be segmented are constructed by clustering and two-step clustering.A series of morphological operations are performed on the images to repair the objects.Then the segmentation is performed.After that,the method builds the candidate glands,and utilizes the spatial position relationship between central lumen and the nucleus around the lumen and some other features to describe glands.By using random forest classification algorithm,the experimental results show that more than 86% accuracy can be achieved.The result lays the foundation for breast cancer automatic grading.
Blind Binary Image Deconvolution Based on Sparse Property
XU Ying and LI Qiang-yi
Computer Science. 2018, 45 (3): 253-257.  doi:10.11896/j.issn.1002-137X.2018.03.040
Abstract PDF(1282KB) ( 727 )   
References | Related Articles | Metrics
Analysis of binary image shows that the pixel values of binary image are sparse,hence the L0 gradient deconvolution is combined with the combinatorial property to deal with the blind binary image restoration problem.Common image restoration methods treat binary image as gray-scale image with an optional threshold,when considering the special property of the binary image,they will get better recovery results for this particular type of image.The proposed blind image restoration algorithm is based on the frame of the first-order gradient space L0 minimization program,uses the L0 gradient image smoothing method to obtain distinct image edges to estimate the blurring kernel,and introduces the special binary property of binary image into the objective function as a regularizer.The binary image prior is used in the restoration process to force the latent restored image to be binary.According to the proposed model,the blind binary image deconvolution algorithm based on the sparse property was presented.The experimental results show that compared with conventional blind deconvolution algorithms,the proposed method has more favorable performance,and is more efficient for binary image restoration.
Sparse Representation Classification Model Based on Non-shared Multiple Measurement Vectors
CAI Ti-jian, FAN Xiao-ping, CHEN Zhi-jie and LIAO Zhi-fang
Computer Science. 2018, 45 (3): 258-262.  doi:10.11896/j.issn.1002-137X.2018.03.041
Abstract PDF(1282KB) ( 816 )   
References | Related Articles | Metrics
Simultaneous sparse reconstruction of multiple measurement vectors(MMV) requires that the multiple mea-surement signals share the same sparse structure.However,it is difficult to get the measurement signals exactly sharing same sparse structure in practical applications.In order to reduce the influence of non-shared sparse structure on simultaneous sparse reconstruction of MMV model,this paper proposed a method to improve simultaneous sparse reconstruction algorithms belonging to greedy series.At each iteration,the method does not require that each measurement vector chooses the same representation atoms,but requires selecting representation atoms in the same class.The improved algorithm can be used for sparse representation classification of non-shared multiple measurement vectors.Experiments on simulated data and standard face database show that the improved model can effectively improve the performance of sparse representation classification.
Single Video Super-resolution Algorithm Based on Non-local Means and Total Variation Minimization
CHEN Cheng, CHANG Kan, MO Cai-wang, LI Tian-yi and QIN Tuan-fa
Computer Science. 2018, 45 (3): 263-267.  doi:10.11896/j.issn.1002-137X.2018.03.042
Abstract PDF(3791KB) ( 608 )   
References | Related Articles | Metrics
The traditional reconstruction-based single video super-resolution algorithms are able to solve the video super-resolution problem well.However,the existing algorithms have not fully exploited the correlation in intra-frames and inter-frames,which leaves much space for further improvement.This paper proposed a new single video super-resolution algorithm to solve this problem.When exploiting the spatial correlations,the non-local means model is used to get the non-local structural property and the total variation model is utilized to get the local structural property.In order to exploit inter-frame correlation,optical flow method is applied to perform inter-frame estimation.Finally,to solve the established optimization problem,a split-Bregman method based fast iteration algorithm was proposed.The experimental results demonstrate the effectiveness of the proposed algorithm.Compared with other algorithms,the proposed algorithm is able to achieve better subjective and objective results.
License Plate Detection Based on Principal Component Analysis Network
ZHONG Fei and YANG Bin
Computer Science. 2018, 45 (3): 268-273.  doi:10.11896/j.issn.1002-137X.2018.03.043
Abstract PDF(8072KB) ( 735 )   
References | Related Articles | Metrics
License plate recognition is the core technology of intelligent transportation system (ITS).License plate detection is a crucial step in the license plate recognition technology.Since only low-level artificial features are used to achieve license plates detection in most traditional methods,the detection error rates are usually low in complex scenes.In this paper,a novel license plate detection method based on principal component analysis network (PCANet) was proposed.Firstly,the license plate candidate area is marked with Sobel operator based edge detection and edges symmetry analysis.Secondly,by inputting candidate area into PCANet,the deep feature extraction is peformed for candidate area in PCANet and the support vector mechine is used to confirm the license plate.Finally,an efficient non maximum suppression (NMS) is used to label the best license plate detection area.For performance evaluation,a dataset consisting of images in various scenes was used to test the proposed method,and the results were also compared with those of traditional methods.The experimental results show the robustness of the proposed algorithm,and its performance is also superior to the traditional method of license plate detection.
Single Color Image Dehazing Based on Dark Channel Prior and MTV Model
ZHAO Sheng-nan, WEI Wei-bo, PAN Zhen-kuan and LI Shuai
Computer Science. 2018, 45 (3): 274-276.  doi:10.11896/j.issn.1002-137X.2018.03.044
Abstract PDF(2405KB) ( 739 )   
References | Related Articles | Metrics
Combining variational partial differential equation with the atmospheric attenuation model,a single color ima-ge dehazing algorithm on the basis of dark channel prior and MTV model called H-MTV model was proposed.Then,using auxiliary variables and Bregman iterative parameters to calculate the model,this paper designed dual Bregman algorithm.Finally, H-MTV model was compared with He algorithm and Kimmel Retinex algorithm.Experimental results show that H-MTV model is superior to the traditional methods qualitatively and quantitatively.
Adaptive Moving Wide Line Detection Algorithm
QU Zhi-guo, TAN Xian-si, LI Zhi-huai, WANG Hong and LIN Qiang
Computer Science. 2018, 45 (3): 277-282.  doi:10.11896/j.issn.1002-137X.2018.03.045
Abstract PDF(4603KB) ( 563 )   
References | Related Articles | Metrics
In this paper,a fast implementation of wide line detector was proposed to eliminate its computation redundancy and improve its computational speed,namely adaptive moving wide line detector.Instead of moving the circular mask pixel by pixel as done in the basic implementation,the adaptive moving wide line detector determines its step adaptively according to the current type of pixel under test.In this way,the computational redundancy can be decreased to a large extent so as to accelerate the detector.Simulated and real images were adopted for performance test of the proposed adaptive moving wide line detector.Experimental results demonstrate that the fast implementation accelerates the wide line detector significantly while keeping its detection performance unaffected.
Image Segmentation Method of Level Set Regularization Based on Bessel Filter
LIU Guo-qi and LI Chen-jing
Computer Science. 2018, 45 (3): 283-287.  doi:10.11896/j.issn.1002-137X.2018.03.046
Abstract PDF(8418KB) ( 649 )   
References | Related Articles | Metrics
A new regularization method based on Bessel filter was proposed to solve the problem of numerical stability of the level set function in the evolutionary process.A new energy model was constructed by embedding this method into the classical region-scalable fitting(RSF) model.Firstly,the K-means algorithm is used to generate the initial level set function automatically to solve the problem of the initialization sensitivity of the RSF model.Secondly,the advantages of region-scalable fitting model are used for iterative segmentation.Finally,in the iterative process,the proposed method is used to maintain the stability of the level set function in order to achieve accurate segmentation results.The experimental results show that the proposed regularization method effectively preserves the stability of the level set functions.The new model has higher efficiency and segmentation accuracy compared with other models based on region.
Embedded Neural Network Face Recognition Method Based on Heterogeneous Multicore Parallel Acceleration
GAO Fang and HUANG Zhang-qin
Computer Science. 2018, 45 (3): 288-293.  doi:10.11896/j.issn.1002-137X.2018.03.047
Abstract PDF(1353KB) ( 890 )   
References | Related Articles | Metrics
Computing performance for massive face data is one of the key problems for face recognition on surveillance device.To improve the performance of embedded face recognition systems,a novel parallel feed forward neural network acceleration framework was established based on CPU-multicore accelerator heterogeneous architecture firstly.Secondly,a feature extraction method based on PCA algorithm was used to extract face features for neural network training and classification.Thirdly,the trained neural network parameters can be imported to the parallel neural network framework for face recognition.Finally,the architecture was implemented on hardware platform named Parallella integrating Zynq Soc and Epiphany.The experimental results show that the proposed implementation obtains very consistent accuracy than that of the dual-core ARM,and achieves 8 times speedup than that of the dual-core ARM.Experiment results prove that the proposed system has significant advantages on computing performance.
Low-rank Constrained Extreme Learning Machine for Efficient Face Recognition
LU Tao, GUAN Ying-jie, PAN Lan-lan and ZHANG Yan-duo
Computer Science. 2018, 45 (3): 294-299.  doi:10.11896/j.issn.1002-137X.2018.03.048
Abstract PDF(4129KB) ( 939 )   
References | Related Articles | Metrics
In complex scenarios,illumination change,occlusion and noise make the image intra-variance of recognition algorithm (taking pixel feature as similarity measure) greater than the between-class variance,and reduce the perfor-mance of face recognition.To solve this problem,this paper proposed an low-rank supported extreme learning machine for robust face recognition to improve recognition performance.Firstly,the subspace linear assumption of face image distribution is used to make the image waiting to be recognized cluster to the corresponding sample subspace.Secondly,the pixel domain is resolved into low-rank feature subspace and sparse error subspace,and the forward network of low-rank structure characteristic of face image for training extreme learning machine is extracted,according to the low-rank principal of the image subspace for noise robustness.Finally,the extreme learning machine face recognition algorithm for noise robustness is realized.Experimental results show that,compared with the state-of-the-art face recognition algorithm,the proposed method not only has high recognition accuracy,but also has lower time complexity and better practicability.
Distributed and Unified Authentication Optimization Mechanism Based on Cache
YANG Dong-ju and FENG Kai
Computer Science. 2018, 45 (3): 300-304.  doi:10.11896/j.issn.1002-137X.2018.03.049
Abstract PDF(1306KB) ( 457 )   
References | Related Articles | Metrics
When the enterprise integrates the application system,it is a common practice to use the independent authentication system to exchange and share the identity information of the platform.How to deal with user requests with high concurrency and large user traffic is an important issue to ensure the stable and efficient operation of the authentication system.In view of the overload of single authentication center,the single point failure and the slow response of the system,this paper proposed to cluster the authentication server.The authentication ticket is stored in the cache so that multiple nodes can share authentication information,and the important and frequently used data can be pre-fetched as cache to improve response speed.This paper proposed a multi-factor cache replacement algorithm based on Hybrid combining the complex and diversified user behavior to improve the effectiveness of data replacement.The experimental results show that the optimized distributed authentication architecture can guarantee system stability and improve system response speed,and the multi-factor cache replacement algorithm based on Hybrid can improve cache hit ratio.
Research on Fault Tolerant Technology for Networks-on-Chip
LI Lu-lu, QIU Xue-hong, ZHOU Duan and ZHANG Jian-xian
Computer Science. 2018, 45 (3): 305-310.  doi:10.11896/j.issn.1002-137X.2018.03.050
Abstract PDF(1265KB) ( 646 )   
References | Related Articles | Metrics
Network-on-chip is a new computer architecture on chip,the study of network-on-chip mainly includes topo-logical structure,routing algorithm,service quality,switching mechanism,congestion avoidance,energy consumption,fault tolerant and so on,and the study of fault tolerant methods is the most important research issue.This paper divided fault tolerant methods into two types:tolerant fault by algorithms and tolerant fault by architecture, from the aspects of software improvement and hardware improvement.This paper analyzed the application conditions,implementation principles and implementation methods of the existing falut tolerant routing algorithms,analyzed the performance of communication latency,throughput and power consumption,and advantages and disadvantages of the existing falut tolerant methods,dissected the situation of the existing falut tolerant methods and offered a possible research orientation.
Rapid Decision Method for Repairing Sequence Based on CFDs
WANG Huan, ZHANG Yun-feng and ZHANG Yan
Computer Science. 2018, 45 (3): 311-316.  doi:10.11896/j.issn.1002-137X.2018.03.051
Abstract PDF(4834KB) ( 515 )   
References | Related Articles | Metrics
Data consistency is one central issue of big data quality management research.Conditional functional depen-dencies (CFDs) are effective techniques for maintaining data consistency.In practice,different repairing sequences may affect precision and efficiency of data repairing.It is critical to select an appropriate repairing sequence.To solve the problem,based on CFDs,this paper presented a rapid decision method for repairing sequence.Firstly,a framework is designed for consistency repairing.Then,by analyzing the association between constraints,the concept of repairing sequence graph is presented to determine repairing sequence on CFDs.It contributes to avoiding some incorrect and unnecessary repairs,which can improve the accuracy of repairing.Meanwhile,repairing sequence with rules runs faster than that with real data.Furthermore,in the process of repairing sequence decision,repairing-deadlock detection is implemented to ensure the termination of repairing.Finally,compared with the existing method,this solution is more accurate and efficient evidenced by the empirical evaluation on two real-life datasets.
Interaction Process Model Mining Method Based on Interface Transitions
ZHAI Peng-jun, FANG Xian-wen and LIU Xiang-wei
Computer Science. 2018, 45 (3): 317-321.  doi:10.11896/j.issn.1002-137X.2018.03.052
Abstract PDF(1299KB) ( 526 )   
References | Related Articles | Metrics
Process model mining is a technology based on the event logs recorded by running system to discover process model corresponding to features.At present,most of the mining methods are based on the frequent interaction between different modules which are decomposed by the system,and there are a few features within modules.There are some limitations of the current process mining methods in the aspect of mining process model which includes multiple features and infrequent interaction.This paper provided an interaction models process mining method based on interface transitions.Firstly,the order of features within modules is discovered using existing methods of mining to find the initial module nets.Secondly,the event log is traversed to search the suspect interface transitions.Then,the interface transition is determined by the mining of the feature net,and the interface place is added to it.Finally,based on the view of open Petri net,the interactive modules are synthesized into a complete process model Petri net.The analysis of instance is used to verify the effectiveness of the mining method.