Loading...
  • 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

Current Issue

    • A task scheduling method for
      network-on-chip temperature optimization
      JI Hui,ZHOU Lei
      2018, 40(09): 1527-1533. doi:
      Abstract ( 131 )   PDF (861KB) ( 182 )     
      As the scale of network-on-chip (NoC) expands and related research goes further, task scheduling becomes a key problem for optimizing system temperature. For the task scheduling problem of NoC, we present a task scheduling method based on the shortest Manhattan distance. The method fully considers the shortest Manhattan path between the pairs of communication nodes in a core communication graph, uses a search algorithm to find the destination node for scheduling tasks, and determines the task scheduling pair with simulated annealing algorithm. SMDS experimental results indicate that, compared with the traditional distributed task migration (DTM) strategy, for 6*6, 8*8 and 10*10 topologies, the optimization for the times of migration is reduced by 22.08%, 21.74% and 23.02%, respectively, and the average optimization rate for average hop count is 24.04%, 29.18% and 23.04%, which achieves system temperature optimization.

       

       

       

       

       
      An aggregated I/O method of sampled data
      for parallel computing applications
      CAO Liqiang,LUO Hongbing
      2018, 40(09): 1534-1539. doi:
      Abstract ( 94 )   PDF (685KB) ( 154 )     

      Parallel I/O of sampled data constrains the operational efficiency of some parallel applications. We design and implement a parallel aggregation I/O method of sampled data. The method first uses the sampled data cache deployed on the client to reduce the number of I/O, and then collects the data to the output process by aggregating the traffic and stores it in the file. To guarantee the storage consistency of sampled data during the longrunning process of parallel programs, we monitor the running state of the application in the JASMIN framework and refresh or restore the data when parallel programs load or restart. During the output process, we use HDF5's chunk I/O to improve I/O efficiency. Test results show that the new method not only has good scalability, but also improves the parallel IO efficiency of sampled data by more than 7.5 times in parallel applications with complex functions such as load balancing or restart.

      CRSHE: A novel ciphertext retrieval scheme
      based on homomorphic encryption
      FU Wei1,LI Moci2,ZHAO Huarong2,WU Yong2
      2018, 40(09): 1540-1545. doi:
      Abstract ( 227 )   PDF (508KB) ( 233 )      Review attachment
      Aiming at the storage and retrieval requirements of ciphertext, we propose a new cloud storage model with separate retrieval and sharing functions. We first design a homomorphic encryption algorithm and propose a novel ciphertext retrieval scheme, namely CRSHE based on the model. It solves the problem of keywords privacy leakage and provides support for homomorphic encryption. The retrieval results can be sorted to reflect the correlation degree between documents and keywords, which greatly improves the retrieval performance in multikeyword retrieval. Experimental results show that the scheme is more efficient and accurate than the traditional linear ciphertext retrieval scheme.
       
       
      Review:Key techniques of write endurance
      improvement for phase change memory
      ZHANG Zhen1,2,FU Yinjin1,HU Guyu1
      2018, 40(09): 1546-1555. doi:
      Abstract ( 121 )   PDF (1098KB) ( 202 )      Review attachment

      With the increasing demand for timeliness in big data analytics and the prominent “memory wall” problem, the storage system becomes the bottleneck of performance improvement of the overall computer systems. The novel nonvolatile memory (NVM), especially phase change memory (PCM), has the advantages of high storage density, low power consumption, high read/write access speed, nonvolatility, small size and quakeproof ability. All these make PCM the most promising candidate for the next generation storage medium. Because of the limited write endurance of PCM, the researches on enhancing PCM lifetime by reducing write operations and performing wearleveling on storage cell attract great attention from academia and industry. We introduce key techniques of write endurance improvement from three aspects of reducing PCM write operations, uniform distribution of write count, and page migration based on hybrid memory. Furthermore, we discuss their advantages and disadvantages. Finally, future research directions of further improvement on PCM lifetime are pointed out and discussed.

      An embedded data recording equipment
      based on ARM
      ZHOU Mingjin1,LU Zaiqi2
      2018, 40(09): 1556-1561. doi:
      Abstract ( 113 )   PDF (936KB) ( 194 )      Review attachment
      In the process of information transformation of a certain equipment’s file recording equipment, by using the framework of FPGA + ARM and the integrated design philosophy for software and hardware, we develop an embedded processing platform, which can realize reliable realtime recording, transmission and storage of multiequipment output data. Problems such as the complexity of interface signal sequential logic, signal distortion, severe noise effect, easy data loss and online function checking are solved. Tests on real equipment and application demonstrate that the embedded system can reach design requirements.
       
      Performance analysis of virtualized congestion control
      with consideration of bandwidth characteristics
      of tenants in datacenter networks
      LI Shunan1,2,3,ZHAN Nanjie1,2,3,ZHANG Yue1,2
      2018, 40(09): 1562-1571. doi:
      Abstract ( 95 )   PDF (1720KB) ( 161 )     
      In datacenter networks, not all the devices are explicit congestion notification (ECN) enabled, especially for legacy devices of tenants. Thus the unfairness between ECN and nonECN flows is inevitable. Virtual congestion control (vCC) solves ECN unfairness by throttling the receiving window of TCP header. We analyze the performance of vCC with consideration of the bandwidth characteristics of tenants in datacenter networks. We design simulation scenarios according to the number of tenant hosts, bandwidth requirement, congestion level of networks, and congestion control algorithms for highspeed networks. Simulation results show that:1) the performance of vCC is good and not affected by the number of tenant hosts, the difference of tenant bandwidth, and the congestion level of networks; 2) the performance of vCC is different under different congestion control algorithms. The ECN unfairness is not well improved by vCC when using YeAHTCP.

       
      Autocorrelation values of
      a binary sequences with even length
      XIONG Zhen,YUE Qin
      2018, 40(09): 1572-1578. doi:
      Abstract ( 81 )   PDF (416KB) ( 138 )     
      Let N be the odd number and ZN be the residual class ring of modular N, the paper obtains the main results as follows:
      (1) If D is a N,N-12,N-34 difference set over ZN, =ZN\D,the binary sequence {si} of length 2N with a characteristic set C1={0}×D∪{1}×,then its autocorrelation value is fourvalued.If two special points are removed,the sequence is optimal.
      (2) If D is a N,N-12,N-54,N-12 almost difference set over ZN, =ZN\D,the binary sequence {si} of length 2N with a characteristic set C1={0}×D∪{1}×,then its autocorrelation value is sixvalued.Finally,this paper also gives the autocorrelation values of a sequence of periodic 4N.
       
      Multi-touch authentication based on
      minimum risk Bayes decision
      SUN Ziwen,LI Fu,PANG Yongchun
      2018, 40(09): 1579-1584. doi:
      Abstract ( 119 )   PDF (468KB) ( 149 )     
      We propose a multitouch authentication algorithm based on minimum risk Bayes decision to improve the recognition rate of multitouch authentication on the smart phone. Firstly, authentication gesture sequences of each finger are classified by logistic regression of the mean dynamic time warping classifier. Then, based on the Bayes rules, we introduce a loss function to calculate the risk of local decisionmaking under the minimum risk Bayes decision fusion rules. Finally, we achieve global decisionmaking results according to the minimum risk Bayes decision fusion rules. Experimental results show that the leak alarm rate maintains satisfactory while the false alarm rate is effectively reduced by introducing minimum risk Bayes decision fusion rules into multitouch authentication, and the overall classification result is good.
       
      Optimization of multidimensional eigenvalue algorithm for
      distributed signal detection in wireless sensor networks
      LIU Yun,CHEN Qian
      2018, 40(09): 1585-1590. doi:
      Abstract ( 78 )   PDF (495KB) ( 225 )     
      In the distributed signal detection of largescale wireless sensor networks, data sets feature high correlation and some redundancy, so when ensuring data acquisition is trusted, it is an important research direction to improve accuracy of high efficiency algorithms. We propose a decentralized power algorithm for the distributed calculation of the maximum eigenvalue of the sample covariance matrix. By combining the average consensus and the iterative power methods, the fast convergence rate and the higher accuracy estimation of the maximum eigenvalue of the covariance matrix are realized under the condition of relatively small sample and a finite number of iterations. Compared with the MECD algorithm and the DST algorithm, simulation results show that the proposed algorithm can effectively reduce the number of signal samples and the number of iterations, the convergence speed is faster, and the detection accuracy can be improved.

       

       
      An improved speed dial gesture authentication method
      GENG Bo1,2,GE Lina1,2,WANG Qiuyue1,2,WANG Lijuan1,2
      2018, 40(09): 1591-1597. doi:
      Abstract ( 115 )   PDF (853KB) ( 260 )     
      Given the singularity problem of the gesture password in speed dial gesture authentication, we can improve the program by adding a 1bit random number and some frequency variables, thus realizing the dynamic change of the authentication password. By improving the authentication process, the number of user authentication points is reduced and the convenience of the authentication mechanism is enhanced. The authentication process is improved as following: the number of authentication points is determined according to the random number and the starting point of the authentication is determined according to the last authentication point of the latest successful authentication (i.e., the authentication point adjacent to the last authentication point of the latest successful authentication is taken as the starting point of the authentication point). If the authentication fails for three times, the random number will be automatically updated. If the random number is updated twice due to the authentication failures, the mobile phone automatically sends its location to the specified mailbox. Finally, the convenience and security of the improved authentication method are analyzed through theory and experiment. Experiments show that the improved authentication method has better security and is more convenient.
       
      Agent communication language based on selfconsciousness
       
      LU Wenhua,LUO Junmin,LI Junwei,GAO Wuqi
      2018, 40(09): 1598-1605. doi:
      Abstract ( 102 )   PDF (460KB) ( 169 )     

      We argue that the purpose of communication between agents is to improve their selfconsciousness, the communication contents between agents is the explanation to agent’s selfconsciousness, and the communication process between agents is the reasoning in agent’s selfconsciousness. Based on the study on classical agent communication language, combining with the research results such as knowledge category, selfconsciousness and mood, we design an agent communication language based on selfconsciousness (ACLBSC), and use the dynamic ontology description language which can describe mutualdescription, fuzziness, dynamics and selfconsciousness of semantics as the representation tool of the communication content in ACLBSC. We explain the message structure and fundamental words in ACLBSC, and demonstrate that the communication process between agents is the reasoning in agent's self consciousness with an example.

      Design of ground testing system and data preprocessing
      software for space-bornesearch coil magnetometer
      #br#  
      CHEN Yu,WANG Miao,REN Haiyan,ZENG Li
      2018, 40(09): 1606-1610. doi:
      Abstract ( 84 )   PDF (725KB) ( 154 )     

      Space-borne search Coil Magnetometer(SCM)is able to detect space low frequency alternating magnetic field.Ground testing system and data preprocessing are important tools for design and development.Firstly,the working principle, data product and research significance of SCM are introduced. Secondly, aground testing system is designed based on theexternal interface of the payload, which implements the communication function between the payload and the computer.By using Visual C++ to call Matlab, a data preprocessing software for SCM is designed, with realtime scientific display function. The system can not only completely receive scientific data but also display the waveforms of different directions and frequency bands in dynamic refresh mode. The method is simple in programming and has comprehensive integrated functions. It has undergone multi-stage satellite testing and experiments. The software running results indicate that the system can monitor and control the working status of the payload in real-time. It has the characteristics of easy maintenance and expandability, and provides test conditions for the development and application of the onboard instrument in the follow-up stage.

      An interval fuzzy spectral clustering
      algorithm for image segmentation
      LIU Hanqiang,ZHANG Qing
      2018, 40(09): 1611-1616. doi:
      Abstract ( 99 )   PDF (1093KB) ( 187 )     
      In recent years, spectral clustering algorithms have been widely used in the field of pattern recognition and computer vision, and the construction of the similarity matrix is the key issue of spectral clustering algorithms. Due to the high computational complexity, it is hard to apply spectral clustering algorithms to large scale image segmentation. Aiming at this problem, an interval fuzzy spectral clustering algorithm for image segmentation is proposed. The method firstly uses the grayscale histogram and the interval fuzzy theory to obtain the interval fuzzy membership degree between image grayscales, then uses this membership degree to construct the grayscalebased interval fuzzy similarity measure. Finally, the similarity measure is used to construct the similarity matrix and the image are grouped by normalized cut criterion so as to obtain the final image segmentation results. Due to the introduction of interval fuzzy theory, the segmentation performance of traditional spectral clustering algorithms is improved, and the comparative experiments also show that the algorithm greatly improves segmentation effect and computational complexity.
       
      A human action behavior recognition method
      based on new projection strategy
      ZHAO Xiaoye,WANG Haocong,JI Xunsheng,PENG Li
      2018, 40(09): 1617-1623. doi:
      Abstract ( 133 )   PDF (866KB) ( 182 )     
      To solve the problem of low recognition rate of micro motion, this paper proposes a multilayer depth motion maps human action recognition method based on new projection strategy and energy homogeneous video segmentation. Firstly, the paper proposes a new projection strategy that projects the depth image into three orthogonal Cartesian planes, so as to retain more behavioral information. Secondly, considering that the image of multilayer depth motion maps based on the whole video can reflect the whole motion information of the video but ignores a lot of detail information, the paper adopts a video segmentation method based on energy homogenization to divide an action video into multiple subvideo sequences, which can more sufficiently depict the detail information. Lastly, this paper uses a local binary pattern feature descriptor to describe the detail texture features of depth motion maps and adopts kernel extreme learning machine classifier to recognize actions. Experimental results on MSRAction3D and MSRGesture3D show that the proposed algorithm can achieve the accuracy of 94.14% and 95.67%, respectively, and has higher accuracy than the existing algorithms.
       
      A Retinex image enhancement algorithm
      based on image fusion technology
      CHANG Jian,LIU Wang,BAI Jiahong
      2018, 40(09): 1624-1635. doi:
      Abstract ( 147 )   PDF (1652KB) ( 230 )     
      We propose a Retinex image enhancement algorithm based on image fusion technology to overcome the disadvantages of halo phenomenon and grey phenomenon of the singlescale Retinex algorithm. For the halo phenomenon, the illumination image of an original image is estimated by Gaussian weighted bilateral filtering instead of Gaussian kernel function, which can remove the halo phenomenon effectively. For the problem of grey, the image fusion technology is introduced into the process of image enhancement. Here are the several steps of the proposed algorithm. In the first place, the reflection image is stretched by nonlinear transformation and the bright and dark areas of the stretched image are determined by the Otsu threshold segmentation algorithm. Then the optimal image for bright areas and the optimal image for dark areas are obtained by adjusting the parameters of nonlinear transformation according to information entropy, and the original image, the optimal image for bright areas and the optimal image for dark areas are fused by the optimal fusion algorithm. Finally, a consistency verification method is introduced to remove the blocking effect caused by the optimal fusion algorithm. Experimental results show that the new algorithm can get the details of an image, remove the halo phenomenon and overcome the grey problem effectively. It also has a better ability to enhance images compared with the singlescale Retinex algorithm, Retinex algorithm based on bilateral filtering, histogram equalization and unshaped mask algorithm.
       
      Color texture features extraction based on quaternion Gabor
      MENG Bo1,WANG Xiaolin1,LI Dongwei2
      2018, 40(09): 1636-1645. doi:
      Abstract ( 177 )   PDF (1482KB) ( 183 )     
      Current color texture feature extraction methods transform color images into gray images or process color image by channel separation, which can lead to color information loss of the original image or correlation loss between channels, so that the texture feature of the feature image differs greatly from that of the original image. To solve the above mentioned problems, we propose a quaternion Gabor method to extract color texture features. Firstly, quaternion Gabor filter is deduced according to traditional Gabor filter and quaternion Euler’s formula, and the color image is described by Quaternion field. Secondly, we propose a quaternion Gabor convolution algorithm to process the color image, and obtain a multiscale, multidirection color texture image. Finally, Tamura statistical feature is extracted from the color texture image. Experimental results show that the proposed method can maintain text features of the original image such as coarseness, contrast and directionality to a great extent and meanwhile obtain color information, which outperforms the traditional Gabor feature image and LBP method.
       
      A high similar image recognition and classification algorithm
      fusing wavelet transform and convolution neural network
      JIANG Wenchao1,2,LIU Haibo1,YANG Yujie1,CHEN Jiafeng1,SUN Aobing2
      2018, 40(09): 1646-1652. doi:
      Abstract ( 206 )   PDF (798KB) ( 511 )     

      A high similar image recognition and classification algorithm fusing wavelet transform and convolution neural network is proposed for high similar image recognition and classification in specific fields with small color and texture feature differences. Firstly, image texture featuresare extracted by wavelet transform, and the optimal texture difference parameter threshold is determined by different categories and different resolution image sets. Secondly, the wavelet decomposition method is used to segment the image,extract each subgraph’s energy features, and normalize them. Then, a convolution neural network with 5 convolution layers and 3 pool layers are used to transform the input image texture feature vector into onedimensional vector. Finally, by increasing the training number and the data amount, the network parameters are continuously optimized and the classification accuracy in the training set is improved. The actual accuracy of the weights is verified in the test set, and the convolutional neural network model with the highest classification accuracy is obtained.Eggs and apples are chosen as the experimental data. Whether the eggs are freerange or captive and where are the original places of the apples are identified in the experiments. The experimental results show that the average accuracy rate of the algorithm is above 90%.

      A spatio-temporal context based visual tracking algorithm
      with anti-occlusion and adaptive target change
      ZHANG Jing,WANG Xu,FAN Hongbo
      2018, 40(09): 1653-1661. doi:
      Abstract ( 135 )   PDF (1166KB) ( 186 )     
      In order to adapt to the target scale changes and resolve the unrecoverable problem of target tracking failure in the tracking process in traditional spatiotemporal context, we propose an antiocclusion and adaptive target change visual tracking algorithm based on spatiotemporal context learning, called STCALD. Firstly, we employ  the TLD median flow algorithm to initialize the tracking point and the forwardbackward (FB) error algorithm to predict the location of the next frame. Secondly, we use the STC algorithm to determine the output box and calculate its conservative similarity. When the threshold is exceeded the tracking is valid, and the movement similarity between the tracking point and the target frame is calculated. On the contrary, if the tracking fails, we use the TLD detector for detection. As for a single cluster box, it is taken as the output directly; but for multiple detection clusters, its spacetime context model is learned, confidence graphs are calculated one by one by the current spatial model, and  the maximum confidence map is taken as the output. Finally, we update classifier related parameters for online learning and conduct experiments on different test video sequences. Experimental results show that the STCALD algorithm can be applied to visual target tracking in complex conditions, such as scale change, occlusion, and so on with a certain degree of robustness.
       
      Dynamic 3D hand gesture interaction
      based on single camera
       
      WANG Yanquan,SUN Bowen
      2018, 40(09): 1662-1669. doi:
      Abstract ( 131 )   PDF (899KB) ( 204 )     
      Traditional gesture interaction requires special interaction equipments such as Leap Motion or Kinect. In this paper, image channel conversion, binarization and other image processing methods are used to extract gestures. The planar motion information of gestures is obtained by the change of the coordinate value of the hand gestures. The problem of gesture depth is solved by the change of gesture area.The matching rate of different images is calculated by drawing the hand contour and using the image matching algorithm, and the highest matching rate is used to selecting the corresponding hand gesture motion information. The transformation matrix is calculated by the transformation from the camera coordinate system to the 3D scene coordinate system and the geometric transformation of 3D graphics, thus achieving the spatial movement and rotation of hand. Finally, the dynamic hand gesture interaction is realized without any specific gesture interaction devices.

       
      A Canny edge detection algorithm based on
      truncated singular value lowrank matrix recovery
       
      GUO Wei,DONG Hongliang,ZHAO Deji
      2018, 40(09): 1670-1678. doi:
      Abstract ( 83 )   PDF (1556KB) ( 569 )     
      In order to improve the accuracy and robustness of the Canny algorithm in processing noise images, we propose a low rank matrix recovery method based on truncated singular value, and present a more accurate dual noiseconvex optimization model and a method of solving the optimization model. We use the classical Canny edge detection method on the decomposed principal component without redundant information, and thus transform the edge detection of the image into the edge detection of the principal component. This method can effectively eliminate the impulse noise and Gaussian noise while preserving the edge information better. To verify its effectiveness, we conduct experiments under different noise concentrations and mixed noises, and the results show that the edge detection algorithm based on lowrank matrix recovery can better preserve the complete edge information and improve the accuracy and robustness of edge detection methods.