Most Read articles

    Published in last 1 year |  In last 2 years |  In last 3 years |  All

    Please wait a minute...
    For Selected: Toggle Thumbnails
    Improved communication algorithm of mobile agent
    HE Min-wei 1,3,FU Qing-ni 2,LI Gui-hai 2,LIN Jian 3
    Computer Engineering and Applications    2008, 44 (29): 119-121.   DOI: 10.3778/j.issn.1002-8331.2008.29.033
    Abstract11624)      PDF(pc) (510KB)(711)       Save
    The reliability and real-time capabilities are the hot spots of the mobile agent communication.A high-efficient and reliable multi-mailbox’s mobile agent communication algorithm is realized based on the former’s algorithms.The new algorithm guarantees communication reliability,real-time capabilities and makes the movement of mobile agent simple and free.
    Related Articles | Metrics
    Analysis of users and users’ Weibo information in Sina Weibo
    LIANG Hong, XU Nanshan, LU Gang
    Computer Engineering and Applications    2015, 51 (7): 141-148.  
    Abstract9915)      PDF(pc) (586KB)(3051)       Save
    Based on the relationship network of Weibo users, the number of fans, User PR values and users’ activities are considered as measurements of users’ influence on Weibo with the distributions of the three factors. Results show that both the distributions of the number of fans and User PR values follow power-law distribution. It is found that there are much more verified users in top User PR ranking list than in fans ranking list and it is suggested that top activity users are much more popular in advertisement campaign after analyzes the top users and their posts in fans ranking, User PR ranking and activity ranking. It is also found that Sina Weibo users prefer to repost and comment on other users’ Weibo. There are a large number of images, videos and links on Sina Weibo, and most of them are reposted from another user.
    Related Articles | Metrics
    Survey on active learning algorithms
    LIU Kang, QIAN Xu, WANG Ziqiang
    Computer Engineering and Applications    2012, 48 (34): 1-4.  
    Abstract9735)      PDF(pc) (572KB)(3894)       Save
    As a method of constructing an effective training set, the goal of active learning algorithm is to find informative sample which can enhance the classification results of the model during the iteration, thereby reducing the size of the training set and improving the efficiency of the model within the limited time and resources. At present, active learning has become a hot issue in the field of pattern recognition, machine learning and data mining. The fundamental ideas, some latest research results and algorithm analysis of active learning are introduced. Some problems for further research are presented and analyzed.
    Related Articles | Metrics
    Human-machine interaction network robot system based on smart phone
    LI Zongwei1,SUN Lei2
    Computer Engineering and Applications    2011, 47 (15): 64-68.  
    Abstract8908)      PDF(pc) (736KB)(614)       Save
    To break through the constraints of the complexity and exclusivity of conventional method,a novel way of controlling robots is proposed in this paper.By combining the advanced interactive technology of iPhone,such as multi-touch technique,with robotic technology,a promising and effective HMI(Human Machine Interface) based on HMM is developed that can analyze human motion.The latest 3G technology is utilized to realize the video feedback based on the Internet.And finally an interaction system is achieved that is universal,portable,and without limitation of distance.Moreover,this system might become the milestone of service robot who will change our life style.In the experiment,this system has shown perfect real-time ability,highly friendliness,and extreme portability.
    Related Articles | Metrics
    Two dimensional heat conduction equation finite volume method to achieve on MATLAB
    XUE Qiong1, XIAO Xiaofeng2
    Computer Engineering and Applications    2012, 48 (24): 197-200.  
    Abstract8269)      PDF(pc) (544KB)(6737)       Save
    Two dimensional diffusionless heat conduction phenomenon has been described on partial differential equation. Based on finite volume method, discretized algebraic equation of partial differential equation has been deduced. Different coefficients and source terms have been discussed under different boundary conditions, which include prescribed heat flux, prescribed temperature, convection and insulation. Transient heat conduction analysis of infinite plate with uniform thickness and two dimensional rectangle region is realized by programming using MATLAB. It is useful to make the heat conduction equation more understandable by its solution with graphical expression. Feasibility and stability of numerical method have been demonstrated by running result.
    Related Articles | Metrics
    Application of improved CSFCM clustering in red tide monitoring
    WANG Xingqiang1, LIU Changxing1, LIU Guowei1, ZHANG Xudong2
    Computer Engineering and Applications    2012, 48 (8): 233-235.  
    Abstract7931)      PDF(pc) (535KB)(307)       Save
    In order to satisfy data mining requirements of red tide monitoring data in marine observation system, a novel CSFCM clustering with subsets of attribute’s weights, based on thorough discussion of algorithms for data mining, is introduced. In the proposed method, a classic COSA(Clustering Object on Subsets of Attributes) algorithm and the fuzzy C-means algorithm are adopted with pretreatment of similary relation and then the criterion function and clustering model are improved. The experimental results demonstrate the applicability, feasibility and efficiency of this new algorithm on data mining of red tide monitoring, and it provides necessary decision-
    making basis for red tide forecast.
    Related Articles | Metrics
    Real-time 3-D reconstruction algorithm based on Kinect and MonoSLAM
    XIA Wenling1, GU Zhaopeng2, YANG Tangsheng2
    Computer Engineering and Applications    2014, 50 (24): 199-203.  
    Abstract7616)      PDF(pc) (586KB)(2480)       Save
    As an important branch of the computer vision technology, the 3-D reconstruction techniques based on the Monocular vision get more and more attention, for it’s simple, low cost and easy to implement. This paper launches a study on the algorithm of the SLAM(Simultaneous Localization and Mapping) by introducing the RGB-D camera Kinect to obtain the depth information of the 3D scene. An algorithm of the 3-D reconstruction based on the Kinect and monocular vision SLAM is achieved.
    Related Articles | Metrics
    Image quality assessment based on phase congruency
    YANG Diwei, YU Shaoquan
    Computer Engineering and Applications    2015, 51 (2): 16-20.  
    Abstract7608)      PDF(pc) (625KB)(1003)       Save
    In practical image processing, some post-processing operations largely depend on the availability of image features. Operations, like classification, segmentation, and image quality assessment, are often carried out in a feature space. The availability of image features plays an important role for further analysis. Rather than defining features directly at points with sharp change in intensity, the Phase Congruency(PC), which is a dimensionless measure of the significant of a local structure, postulates that features are perceived at points where the Fourier components are maximal in phase. The PC model is in accordance with the human visual system that demonstrates good invariance to light conditions and can achieve effective feature information and well feature location accuracy. Based on those properties, it proposes a new image quality metric based on the PC model, which utilizes the local similarities of phase congruency between the reference and distorted image to quantify the image distortion. Moreover, considering that human visual system is sensitive to phase congruent structures and that the PC value at a location can reflect how likely it is a perceptibly significant structure point, PC values at a location are employed to combine the similarities within local regions into a single quality score. The experimental result shows that the proposed algorithm is correlated well with the judgment of human observers. More importantly, the metric is invariant to changes in image brightness or contrast.
    Related Articles | Metrics
    Face tracking and recognition system for video surveillance
    SANG Haifeng, WU Danyang, WANG Hui
    Computer Engineering and Applications    2014, 50 (12): 175-179.  
    Abstract7560)      PDF(pc) (608KB)(641)       Save
    For requirements of intelligent video surveillance system, this paper presents an automatic multiple face tracking recognition system based on video surveillance, which can track multiple faces real-timely and recognize the identity. Aiming at the influence of complex background and similar to the face region, it puts forward a face detection algorithm based on both Adaboost face detection algorithm and Active Shape Model, and realizes the face detection accurately; a multiple faces tracking algorithm combining CamShift with Kalman filter is proposed for many faces deflection, interlaced and the number changes in the video surveillance scene. Meanwhile, the algorithm also can identify the faces which have been tracked. The experiment results show that in the video surveillance, the system is capable of improving the accurate rate of faces detection and recognition, and it also can track the real-time faces effectively. It is a practical method for developing visual surveillance system.
    Related Articles | Metrics
    High-speed digital textile printing system based on multi-core processor and FPGA
    LUO Cheng, CHEN Jiayang, CHEN Yaowu
    Computer Engineering and Applications    2015, 51 (17): 265-270.  
    Abstract7331)      PDF(pc) (600KB)(563)       Save
    Traditional digital textile printing machine uses PC to complete the image data processing and transmission. Its data bandwidth is narrow and the print speed is slow. To this end, a solution of high-speed digital textile printing system based on multi-core processer is proposed. The system receives the image date by two gigabit fiber ethernet interface. The multi-core processer is applied to complete the decoding of data and the transposition of image. The algorithm of shift transposition is applied to promote the transposition efficiency significantly in place of the algorithm of traversal transposition. The processor achieves the data sharing and synchronization between different cores by sharing memory and applying mailbox technique, as a result, the efficiency of the system is significantly improved. FPGA is applied to achieve the data format conversion and printing output. Test results show that, the system’s total output bandwidth reaches 1.9 Gb/s with 8 sprinkler head, and the printing speed can get to 720 m2/h with 600 dot/inch resolution.
    Related Articles | Metrics
    Performance analysis of MPTCP and CMT-SCTP multi-path transport protocols
    FU Fa1, ZHOU Xing1, YANG Xiong1, Hakim Adhari2, Erwin P.Rathgeb2
    Computer Engineering and Applications    2013, 49 (21): 79-82.  
    Abstract7300)      PDF(pc) (529KB)(1638)       Save
    Achieving multipath transport by terminal’s multi-homed characteristics is a hot topic of Internet protocol application. MPTCP and CMT-SCTP are both more mature and key protocols for multipath transmission, but there is still a lot of performance analysis needed before large-scale application. This paper presents the behaviors of MPTCP and CMT-SCTP multipath transmission and compares the difference between both protocols on the local test-bed environment. Results show that both MPTCP and CMT-SCTP protocols can significantly increase throughput. Meanwhile the scheduling strategy for transmission is still not perfect and needs further improvement.
    Related Articles | Metrics
    Segmentation method for merged characters in CAPTCHA based on drop fall algorithm
    LI Xingguo1,2, GAO Wei1
    Computer Engineering and Applications    2014, 50 (1): 163-166.  
    Abstract7183)      PDF(pc) (466KB)(4360)       Save
    Many researches demonstrate that good result can be gained by existing machine learning algorithms in the recognization of CAPTCHA(Completely Automated Public Turing test to tell Computers and Humans Apart) if single characters can be split. A method is presented to segment the merged characters in the recognization of CAPTCHA with touching characters. It seeks division points by combining the statistics of character width and the vertical histogram projection minimums, and then uses these points as the starting points of the drop fall algorithm to segment merged characters in CAPTCHA. The experiments show that it is a general method and can improve the recognization rate.
    Related Articles | Metrics
    Novel interpretation for Levenberg-Marquardt algorithm
    ZHANG Hong-yan,GENG Zheng
    Computer Engineering and Applications    2009, 45 (19): 5-8.   DOI: 10.3778/j.issn.1002-8331.2009.19.002
    Abstract7032)      PDF(pc) (731KB)(1301)       Save
    The Levenberg-Marquardt(LM) algorithm is closely related with the Least Squares(LS) approach.The Scaled Total Least Squares(STLS) approach is a unification and generalization of the LS,Data Least Squares(DLS) and Total Least Squares(TLS) approaches,but its relation with the LM algorithm is not clear.In this paper,a STLS algorithm and its interpretations via subspace and topology are proposed.The relation of the STLS approach and the LM algorithm are explored by matrix decomposition and the results show that:LS solutions are converted to STLS solutions when the damp factors are introduced.The robustness and convergence performance of the LM algorithm are reached by eliminating the noise subspace and controlling the condition number of the coefficient matrix.The capabilities in solving the over-parameterized problems of the LM algorithm are determined by the robustness of the STLS approach.
    Related Articles | Metrics
    Personalized E-Commerce real-time recommendation based on information flow
    SHEN Ai-guo
    Computer Engineering and Applications    2008, 44 (28): 91-94.   DOI: 10.3778/j.issn.1002-8331.2008.28.032
    Abstract7010)      PDF(pc) (883KB)(747)       Save
    In dealing with the needs of real-time personalized recommendation,personalized recommendation method for E-Commerce system is studied through deep mining of customer’s browse record based on information flow representation,generation and analysis.In order to manage information flow,this paper proposes a personalized recommendation algorithm,which consists of four phases:information flow definition,information flow evaluation,knowledge matching and website dynamic organization.A model including client,server and storage system is designed based on J2EE.To demonstrate the practical usefulness of this methodology,a real-life case is illustrated,which shows recommendation accuracy increases according to the increase of training number,and superior to association rule.
    Related Articles | Metrics
    Collaborative filtering algorithm based on matrix decomposition
    LI Gai1,2,3,LI Lei2,3
    Computer Engineering and Applications    2011, 47 (30): 4-7.  
    Abstract6922)      PDF(pc) (588KB)(2957)       Save
    Collaborative filtering recommendation algorithm is one of the most successful technologies in the e-commerce recommendation system.Aiming at the problem that traditional collaborative filtering algorithms generally exist sparseness resistance and extendibility,in this paper,a CF algorithm,alternating-least-squares with weighted-[λ]-regularization(ALS-WR) is described.That is,by using regularization constraint to the traditional matrix decomposition model to prevent model overfitting training data and using alternating-least-squares method to train the decomposition model.The experimental evaluation using two real-world datasets shows that ALS-WR achieves better results in comparison with several classical collaborative filtering recommendation algorithms not only in extendibility but also in sparseness resistance.
    Related Articles | Metrics
    Simulation research of force control system for manipulator based on Matlab
    QIN Haiqiang1, XIONG Qingyu1,2, SHI Xin1, WANG Kai1
    Computer Engineering and Applications    2014, 50 (12): 242-246.  
    Abstract6812)      PDF(pc) (587KB)(1558)       Save
    The position control of manipulator can’t meet the application requirements of some complex tasks(assembly, polishing, deburring). The control of contact force between manipulator and environment is by far one of the topics of most active and continuing interest in the field of Robotics. A simulation method on force control system for 2-R planar manipulator based on Matlab/SimMechanics is proposed. By simulating the contact surface of manipulator with environment in the plane, this paper designs the suppression controller, realizes the control of the contact force when manipulator interacts with environment, and the oscillation suppression in the phase of impact oscillation when manipulator interacts with rigid environment, generates the expected motion trajectory of manipulator. The simulation results show that the method can realize the control of contact force between manipulator and environment under a specific type of job.
    Related Articles | Metrics
    Improving CNM algorithm to detect community structures of weighted network
    HAN Hua,WANG Juan,WANG Hui
    Computer Engineering and Applications    2010, 46 (35): 86-89.   DOI: 10.3778/j.issn.1002-8331.2010.35.025
    Abstract6798)      PDF(pc) (585KB)(997)       Save
    For detecting community structures on weighted network that can reflect the important properties of the network structure,this paper chooses the hierarchical clustering methods that have been widely used in community structure,and improves CNM algorithm.The new algorithm introduces the link weight and vertex weight,defines a new Q-function to calculate the community modularity.The type of communities are classified by finding the Q peak.When being tested on the weighted network with the stock price fluctuation of correlation for link weigh,the community division results show that the improved CNM algorithm is effective.And a comparative analysis is maken with the improved GN algorithms,global optimization algorithms on the same network of detecting community structures.The improved CNM algorithm demonstrates excellent detection results,the accuracy of classification and very fast process performance.
    Related Articles | Metrics
    Development of vehicle master section parameter measure tool based on CATIA\CAA
    HU Sibo1,2, DING Yi1, ZHU Dexia1, HU Ping2
    Computer Engineering and Applications    2014, 50 (20): 249-253.  
    Abstract6610)      PDF(pc) (924KB)(1947)       Save
    During the period of vehicle concept design, the evaluation of master section performance mostly depends on reference vehicle data and the experience of structure design engineers. In this paper, it studies the influence of master section parameters on the vehicle body performance and sets up parameterized model of master section. A Master Section Measure Tool(MSMT) is developed based on CATIA/CAA, which improves design efficiency by measuring the master section parameters according to CAD surface data. As an example, six master sections of A pillar are respectively measured by MSMT and Altair/Hypermesh. The measure results of master section measure tool are the same as the Altair/Hypermesh’s, which approve the practicability and the veracity of this master section measure tools.
    Related Articles | Metrics
    People detection based on Gentleboost algorithm
    CHENG Li-fang,ZHAO Ming-chang,ZHANG Xiang-wen
    Computer Engineering and Applications    2008, 44 (23): 187-190.   DOI: 10.3778/j.issn.1002-8331.2008.23.056
    Abstract6541)      PDF(pc) (921KB)(836)       Save
    The traditional person detection method is more about small number of samples.Using this method,the detection rate of the image with complex scenes is low.But the reality scene is complex and the real-time detection system need to handle a large number of images.In order to remedy the deficiency of the traditional test methods in the human detection,a person detection approach based on the integrated study—Gentleboost algorithm is suggested.The method uses human patch and its relative position with the person’s center as the feature and uses Gentleboost algorithm train the classifier which is used to classify the person.In order to improve the learning efficiency of the classifier and solve the detecton problem of the image with complicate scenes,a approach which use the linear regression stump as the weak classifier is advanced.The approach weights to the pretiction model from the positive and negative classification,but not to weight it only from the positive aspect.This article compares Gentleboost algorithm with the detecton method based on YCbCr surface filter and the body part feature(for short YCbCr algorithm)and also compares classification with different iteration.The result of the experiment shows that the classification of Gentleboost is better than that of YCbCr algorithm and the method is simple to implement and the numberical is also stable and the positive detection rate is high.This method can handle a large number of images and solve some key issues of human detection.
    Related Articles | Metrics
    Parallel fuzzy C-means algorithm based on MapReduce
    YU Qianqian, DAI Yueming
    Computer Engineering and Applications    2013, 49 (14): 133-137.  
    Abstract6485)      PDF(pc) (562KB)(779)       Save
    Fuzzy C-means?is an important?soft-clustering algorithm, but with the increased amount of data the time complexity will be increased. In this paper, a parallel?fuzzy?C-means?algorithm based on?the MapReduce is proposed. The fuzzy?C-means?algorithm is redesigned to meet the MapReduce programming model. The membership degree of data set to the center is computed in parallel, and the new cluster center is re-calculated, so that the higher calculating efficiency of processing large amount of data can be got. The experimental results show that the parallel?fuzzy?C-means?algorithm based on?the MapReduce has the advantages of both high speedup and good scalability.
    Related Articles | Metrics
    Fast and accurate volume calculation method for arbitrary triangular meshes
    WANG Quan-de
    Computer Engineering and Applications    2009, 45 (18): 32-34.   DOI: 10.3778/j.issn.1002-8331.2009.18.009
    Abstract6399)      PDF(pc) (597KB)(1783)       Save
    Triangular meshes are widely used in many field.In this paper,a new volume calculation method for arbitrary triangular mesh is presented based on the triangles set of the mesh only.Give a projecting plane,sign volume of pentahedron for each triangle in mesh,which is constructed by the triangle and it projection in projecting plane,is computed.Volume of the mesh is algebraic sum of all pentahedrons’ signed volume.Triangles will be partition into several smaller triangles when necessary.Volume calculation method presented in this paper can computed volume of meshes rapidly and accurately.Applications of the method to compute volume for constructing moulding unit in simulating large water and hydropower project constructing,and test result of computing volume for triangular meshes with different number of triangles,it is proved the effectiveness of the volume calculation method presented in this paper.
    Related Articles | Metrics
    Fast algorithm for modular operation
    XU Xin, LI Shundong
    Computer Engineering and Applications    2014, 50 (22): 136-140.  
    Abstract6211)      PDF(pc) (500KB)(1299)       Save
    Related Articles | Metrics
    Non-recursive programming algorithm for quickselect
    QIU Yonghong1,2, ZENG Yongnian2, ZOU Bin2
    Computer Engineering and Applications    2012, 48 (29): 13-15.  
    Abstract6098)      PDF(pc) (465KB)(863)       Save
    After analyzing the basic idea of the quickselect algorithm, a non-recursive programming algorithm, cyclically iterative algorithm, for quick selection is presented. The performance of the algorithm is validated by both the traditional robust recursive algorithm and “nth_element” which is the VC++ standard library function. The result shows that the proposed cyclically iterative algorithm performs relatively better than the recursive algorithm in terms of time efficiency and reliability, while it is also more efficient compared to the “nth_element”.
    Related Articles | Metrics
    Identifying opinion leaders by improved algorithm based on LeaderRank
    XU Junming, ZHU Fuxi, LIU Shichao, ZHU Biying
    Computer Engineering and Applications    2015, 51 (1): 110-114.  
    Abstract6091)      PDF(pc) (589KB)(1363)       Save
    Opinion leader mining is an important topic of social network research which is of great significance in terms of internet public opinion control and information dissemination. LeaderRank is an effective opinion leader mining algorithm. This paper proposes a new algorithm based on LeaderRank with sentiment analysis and user’s liveness to identify the opinion leaders in comment network to achieve a better performance. Experiments based on SIR model show that both accuracy and anti-jamming capability of the proposed algorithm have been effectively improved.
    Related Articles | Metrics
    Real-time method for people counting based on RGB-D cameras
    ZHANG Hua, LIU Hongli, XU Kun, MA Ziji
    Computer Engineering and Applications    2014, 50 (23): 156-162.  
    Abstract5922)      PDF(pc) (752KB)(1143)       Save
    To solve problems such as occlusion and illumination changes that lead to low accuracy on conventional methods, a novel method D-RWS (Depth map based Rainfalling Watershed Segmentation) is proposed. Depth map is inpainted and foreground is extracted with the help of mixture of Gaussian background model. D-RWS algorithm is used to segment head area as Region Of Interest(ROI). People are tracked and counted by analyzing trajectories, which associated by minimal Euclidean distance?between the centers. Experimental results show that proposed people counting system can, on average, count people with an accuracy of 98% and operate at approximately 25 milliseconds per frame(40 f/s). The accuracy and real-time performance fully meet the requirements of practical application.
    Related Articles | Metrics
    Application of SIFT and advanced RANSAC algorithm on image registration
    LUO Wenchao, LIU Guodong, YANG Haiyan
    Computer Engineering and Applications    2013, 49 (15): 147-149.  
    Abstract5897)      PDF(pc) (538KB)(1285)       Save
    There have been great advances in object recognition and image registration, through the match of invariant local image feature. With the feature of invariance to affine, 3D projection, scaling and rotation, illumination changes, image translation, the Scale Invariant Features Transform(SIFT) is commonly used in object recognition. The RANdom Sample Consensus(RANSAC) is widely used as robust estimator, as a standard in the field of computer vision. In the process of R-RANSAC, the Randomized(hypothesis evaluation) RANSAC, a modification is made to RANSAC by checking data points sequentially, while the standard RANSAC check all the data points in the model verification step. Own to this improvement, hypotheses with low support can be rejected before all points are considered. In addition, the Sequential Probability Ratios Test(SPRT) is used to minimize R-RANSAC runtime.
    Related Articles | Metrics
    HVACSIM+ and MATLAB-based simulation of air-conditioning system
    YAN Xiuying1, REN Qingchang1, MENG Qinglong2
    Computer Engineering and Applications    2012, 48 (15): 227-232.  
    Abstract5854)      PDF(pc) (1396KB)(869)       Save
    The simulation of the air-conditioning system and the optimized control of minimizing the energy consumption under the constraint of temperature are mainly studied. The simulation software HVACSIM+ for air-conditioning systems is introduced. The air handling unit and the rooms of the first floor of one building are taken as the simulation target; a simulation system is set up based on the platform of HVACSIM+. The function describing the relation between the consumption of cooling coil and the chilled water velocity is found and taken as the objective function, and with improved cyclic variable method the controller’s parameters are optimized under MATLAB environment. The HVACSIM+ simulator is run again to calculate the energy consumption supplied by the cooling source. Results show that the system runs steadily while the controller works with the optimal parameters and the energy consumption of the cooling coil is greatly reduced compared with working in non-optimized condition.
    Related Articles | Metrics
    User similarity function considering weight of items similarity
    LUO Jun, ZHU Wenqi
    Computer Engineering and Applications    2015, 51 (8): 123-127.  
    Abstract5848)      PDF(pc) (552KB)(1156)       Save
    In traditional user similarity function, the weight for each item is the same. However, analyzing traditional collaborative filtering algorithms and practical case, the weight of user’s jointly high scoring item should be higher than the weight of user’s jointly low scoring item. And traditional user similarity function does not take taxa relationship between the items. To address the problem, it proposes a method to weight project and finally obtains a user similarity function which considers the similarity weight of items. The experimental results conducted on the movieLens data sets show that compared with the collaborative filtering algorithm which is based on the traditional user similarity function, the collaborative filtering algorithm which considers the similarity weight of items can significantly improve the ratings prediction accuracy and the quality of the recommendation system.
    Related Articles | Metrics
    Classifier and feature selection algorithm by kernel-weighted Logistic regression model
    CHI Guanghui1, LIU Jianwei1, LI Weimin2, LUO Xionglin1
    Computer Engineering and Applications    2013, 49 (9): 41-44.  
    Abstract5797)      PDF(pc) (518KB)(816)       Save
    Under supervised learning settings, problems that the dimension of the samples is typically larger than the number of samples are often encountered, i.e. many irrelevant features exist. In such case, the approaches that simultaneously achieve sparsely variable selection and better accuracy of classification are more preferable. In this paper, classification and feature slection algorithm based on kernel-weighted nonlinear logistic regression model is proposed. Each diagonal element of the weight diagonal matrix has a value between 0 and 1, which is as a learning parameter determined by optimization procedure, and fast alternative optimization methods are discussed. The proposed methods are tested on ten real-world datasets. The experimental results indicate that the proposed methods show high classification accuracies on these data sets than L1, L2, Lp norm regularization classifier algorithm of logistic regression model.
    Related Articles | Metrics
    Recommender algorithm combined with neighborhood model and LFM
    LU Quan, WANG Rulong, ZHANG Jin, DING Yi
    Computer Engineering and Applications    2013, 49 (19): 100-103.  
    Abstract5772)      PDF(pc) (596KB)(1238)       Save
    As one of the most successful approaches to building recommender systems, Collaborative Filtering(CF) uses the known preferences of a group of users to make recommendations or predictions of the unknown preferences for other users. The two successful approaches to CF are latent factor models, which directly profile both users and products, and neighborhood models, which analyze similarities between products or users. This paper introduces some innovations to both approaches. The factor and neighborhood models can now be smoothly merged, thereby building a more accurate combined model. Further accuracy improvements are achieved by extending the models to exploit both explicit and implicit feedback by the users. The methods are tested on the Netflix data, and the results are better than those previously published on that dataset.
    Related Articles | Metrics
    Semi-auto stitching of scrapped paper based on character characteristic
    LUO Zhizhong
    Computer Engineering and Applications    2012, 48 (5): 207-210.  
    Abstract5731)      PDF(pc) (544KB)(1041)       Save
    This paper analyzes demerits of classical stitching method based on edge geometry characteristic of scrapped paper, studies characteristic of character row and table grid in scrapped paper, investigates their acquisition method, and proposes a semi auto stitching method for scrapped paper based on character row and table grid. In the end, C language computer programs for semi-auto stitching of scrapped paper are developed according to this algorithm. They are applied in a stitching experiment for pieces of scrapped paper. The result shows that this semi-auto stitching method is effective.
    Related Articles | Metrics
    Model-free calibration method for omni-directional vision system
    YANG Shao-wu,LU Hui-min,ZHANG Hui,ZHENG Zhi-qiang
    Computer Engineering and Applications    2010, 46 (25): 203-206.   DOI: 10.3778/j.issn.1002-8331.2010.25.059
    Abstract5708)      PDF(pc) (548KB)(543)       Save
    The traditional calibration methods for omni-directional vision system make the assumptions that the vision system is a single view point system and the shape of the omni-directional mirror is symmetric in every direction.As the assumptions are hard to meet in applications,the calibration precision may be affected greatly.A new model-free calibration method is designed according to the characteristic of the omni-directional vision.The new method can be applied in the calibration of all kinds of catadioptric omni-directional vision systems with good precision without making the above assumptions.After applying the new method in the calibration of the omni-directional vision system of NuBot,the accuracy of robot’s self-localization based on omni-directional vision is obviously improved,and this verifies the efficiency and practicability of the new method.
    Related Articles | Metrics
    FPGA-based design for convolution neural network
    FANG Rui, LIU Jiahe, XUE Zhihui, YANG Guangwen
    Computer Engineering and Applications    2015, 51 (8): 32-36.  
    Abstract5691)      PDF(pc) (567KB)(3253)       Save
    According to the characteristics of the Convolution Neural Network(CNN), a FPGA-based acceleration program which uses deep-pipeline architecture is proposed for the MNIST data set. In this program, theoretically 28×28 clock cycles can finish the whole calculation and get the output of the CNN. For the propagation stage of the training process, and in the same network structure and the same data set, this FPGA program with 50 MHz frequency can achieve nearly five times speedup compared to GPU version(Caffe), achieve eight times speedup compared to 12 CPU cores. While the FPGA program just costs 26.7% power which GPU version costs.
    Related Articles | Metrics
    Algorithm of parameter estimation of HMM via Gibbs sampling
    ZHU Chengwen1, LI Bing2, HU Kui3
    Computer Engineering and Applications    2012, 48 (18): 57-60.  
    Abstract5687)      PDF(pc) (575KB)(1020)       Save
    The parameter estimation of Hidden Markov Model(HMM) is critical to all its applications. The classic Baum-Welch algorithm is not flexible with the initial parameters and is easy to fall into the local optimal solution. The great computational power of MCMC is employed. The algorithm of parameter estimation of HMM via Gibbs sampling avoids the local optimal solution and can be more effective.
    Related Articles | Metrics
    Using cartoon texture image decomposition and GAPG for total-variation-based image restoration
    JIANG Zhengjin, DUANMU Chunjiang
    Computer Engineering and Applications    2014, 50 (2): 162-169.  
    Abstract5504)      PDF(pc) (678KB)(677)       Save
    In order to research the image restoration technology, this paper proposes a kind of approach by using cartoon texture image decomposition and a generalized accelerated proximal gradient algorithm to realize image restoration. To decompose the original fuzzy image into cartoon part and texture part, cartoon part is the low frequency component of the image, noise interference is small, the texture of the image is the high frequency component of the image, noise interference is sharp, and it deblurs and denoises by using the generalized accelerated proximal gradient algorithm, selects small regularization parameters for the cartoon parts, selects big regularization parameters for the texture parts, composes the restoration of the cartoon part and the restoration of the texture part to achieve restoration image. Through simulation of two standard test images by MATLAB, it shows that this method not only converges faster but also is better than the effect of generalized accelerated proximal gradient algorithm, especially is suitable for recovering the image of small fuzzy degrees.
    Related Articles | Metrics
    Design of IOT embedded gateway based on STM32
    LI Xiaodan
    Computer Engineering and Applications    2015, 51 (4): 61-65.  
    Abstract5487)      PDF(pc) (537KB)(1185)       Save
    In view of wide application in intelligent sensor control of internet of things, an embedded gateway serviced in the IOT intelligence collection system, by using the embedded technology, and the modern high-speed development of the network platform is proposed. The embedded gateway system is composed of the embedded STM32F host device, router, GPRS module and wireless coordination device. The host device based on μC/OS-II operating system, connects to the router by 100 Mbit/s Ethernet interface, connects to the GPRS module and wireless coordination device by using serial port. It is observed that the function of real-time display, the previous data storage, the gateway control parameters setting and inquiry, the Internet network remote access and control, and GPRS network remote access and control.
    Related Articles | Metrics
    Moving objects detection based on HSV colorspace and Vibe algorithm
    ZHANG Lei, FU Zhizhong, ZHOU Yueping
    Computer Engineering and Applications    2014, 50 (4): 181-185.  
    Abstract5475)      PDF(pc) (554KB)(1235)       Save
    Vibe algorithm runs very quickly and can reduce the impact of shadow and the shaking of a camera effectively, and has a good performance for foreground detection. But for those dynamic background outdoor videos, it’s inevitable to avoid the background disturbance, and the algorithm can’t achieve precise foreground detection. To solve the problem, a novel strategy combined Vibe algorithm with color information is proposed. Pixels are converted from RGB space to HSV space to compute pixel color distortion. And the amount of calculation is obviously reduced due to the construction of Dual model. Experiments show that the algorithm can obtain more accurate detection property for moving objects, and has showed well robustness against disturbance.
    Related Articles | Metrics
    Imbalanced data resampling based on oversampling and under-sampling
    WU Lei1, FANG Bin1, DIAO Liping2, CHEN Jing1, XIE Nana1
    Computer Engineering and Applications    2013, 49 (21): 172-176.  
    Abstract5458)      PDF(pc) (499KB)(1124)       Save
    There are?several?aspects that might influence the performance achieved by?existing learning systems in the area of machine learning. It has been reported that one of these aspects is related to class?imbalance in which examples in training data belonging to one class heavily outnumber the examples in the other class. Though there are several kinds of methods to get rid of this problem, this paper only discusses using resampling method to balance data in the period of preprocessing to improve the effect of classification. There are two kinds of resampling methods:over resampling and under resampling. In this paper, four methods which combine oversampling and under-sampling method are proposed for binary classification:BSM+Tomek, BSM+ENN, CBOS+Tomek and CBOS+ENN, and present very good results for data sets with a small number of positive examples. Moreover, ten other resampling methods are also taken to make comparative experiments with the four methods proposed by this paper, and the four methods also present very good results.
    Related Articles | Metrics
    Semantic similarity computation between sentences
    LV Qiang 1,DENG Wei 2,SONG Ling 3
    Computer Engineering and Applications    2010, 46 (36): 150-153.   DOI: 10.3778/j.issn.1002-8331.2010.36.041
    Abstract5458)      PDF(pc) (534KB)(740)       Save
    The problem of measuring the similarity between two sentences or two text segments has become increasingly important for many Web-related tasks.A Semantic Similarity between Sentences(SSBS) is devised based on semantic similarity between words.Relevant experiments are carried.Compared with other methods,SSBS considers not only semantic similarity and string edit distance between vocabularies but also part of speech tagging.
    Related Articles | Metrics
    Decryption method of Baidu map’s coordinates based on artificial neural network
    PAN Weizhou, CHEN Zhenzhou, LI Xingmin
    Computer Engineering and Applications    2014, 50 (17): 110-113.  
    Abstract5420)      PDF(pc) (487KB)(878)       Save
    This paper presents a decryption method of Baidu map’s coordinates. A converting interface from GPS coordinates to Baidu map’s coordinates given by Baidu is utilized to get pairs of GPS-Baidu coordinates. The inverse converting function is fitted based on BP neural network. The trained BP network is used to predict GPS coordinates corresponding to some new Baidu coordinates. Experimental result demonstrates that the converted GPS coordinate given by this method is closed to the real GPS coordinate. Hence the decrypted result is ideal.
    Related Articles | Metrics