GRABAC NN model,which means Gradient Radical Basis Cell Neural Network,is a kind of generalized neural *** with Hopfield-type models,it has much more dynamic qualitative behavior,that is all tracks converge one of the...
详细信息
GRABAC NN model,which means Gradient Radical Basis Cell Neural Network,is a kind of generalized neural *** with Hopfield-type models,it has much more dynamic qualitative behavior,that is all tracks converge one of the fixpoints,its complexity of time and space are O(mn) other than O(nn),and unlimited memory capacity can be obtained,in
＜正＞Normal neural networks have tendency to "forget" previously learned pattrens when taught new *** a characteristic would be ruinous if expected to adapt to a continuously changing practical *** paper in...
＜正＞Normal neural networks have tendency to "forget" previously learned pattrens when taught new *** a characteristic would be ruinous if expected to adapt to a continuously changing practical *** paper introduces a definition of neural nets structure and use two structure models:series-structure and tree-structure,to deal with the problem of *** practical examples are character recognition and parity *** the application, tag set shows excellent characteristic of enlarging distance between *** results prove that the structure method and tag set are valuable and feasible in solving forgetfulness problem.
＜正＞An Automatical Engineering Drawings recognition System（AEDRS）, which combines artificial neural networks（ANN） methods and conventional methods,is proposed in this *** ANN methods can correctly recognize char...
＜正＞An Automatical Engineering Drawings recognition System（AEDRS）, which combines artificial neural networks（ANN） methods and conventional methods,is proposed in this *** ANN methods can correctly recognize characters and marks in engineering drawings unaffected by shifting and rotating,find important parts of drawings,provide heuristic information to conventional *** methods,which exchange information with ANN methods,work more effectively and *** system can automatically recognize the scanned raster engineering drawings,give the recognition result including vectors,representatons and positions of characters and marks,circles and arcs *** we get the standard description of engineering drawings in computer.
An improved H.261（I-H.261） scheme for videoconference image coding is presented in this *** the case of lower bitrate（p=l,2） and higher error code rate(10～10),p×64kb/s H.261（CIF） has inferior picture quali...
详细信息
An improved H.261（I-H.261） scheme for videoconference image coding is presented in this *** the case of lower bitrate（p=l,2） and higher error code rate(10～10),p×64kb/s H.261（CIF） has inferior picture quality and poor noise *** view of this situation,we proposed I-H.261 which is compatable with H.261 in different levels. There are two improvements:the first,improved the control approach for the quantizer stepsize,a new control strategy for quantization is proposed;the second,improved the code table for DCT coefficients,a new code table is set,in which codeword length is quadruple and odd-even check function is *** with H.261,I-H.261 has three notable advantages:（a） It reduced the fuzziness distrotion and mosquito effect at lower bitrate（p=1,2）(b) A powerful noise immunity is achieved,the range of bit error diffusion is narrowed,（c） In the main,it's compatable with ***'s easy for implementation in resl time to use existing ASIC chips designed for H.261 and general purpose DSP chips. The excellent simualtion result is obtained at bitrate of 96kb/*** code rate of 10.
A group of threshold update algorithms with respect to simultaneous, partially simultaneous and sequential update modes is given. The network with these algorithms is asymptotically stable. The upper bounds of residua...
详细信息
ISBN:
(纸本)0780308034
A group of threshold update algorithms with respect to simultaneous, partially simultaneous and sequential update modes is given. The network with these algorithms is asymptotically stable. The upper bounds of residuals of the convergent solutions are presented. Simulation results are shown.< >
A full domain optimum neural network is proposed. A method of devising stable points firstly and basins of attraction latterly increases speed and correctness of imagerecognition. This paper proves the quality and sp...
详细信息
A full domain optimum neural network is proposed. A method of devising stable points firstly and basins of attraction latterly increases speed and correctness of imagerecognition. This paper proves the quality and speed of convergence and invariance of mapping. Several computer simulation examples involving trained and recognized targets illustrate the usefulness of the method.< >
In this approach to bilevel image restoration the autoconnections of the network generally weight more heavily than interconnections. This characteristic exists in general degradation models of image restoration and c...
详细信息
ISBN:
(纸本)0780308034
In this approach to bilevel image restoration the autoconnections of the network generally weight more heavily than interconnections. This characteristic exists in general degradation models of image restoration and can be utilized to guide the network to be updated more efficiently. A criterion for choice of the neurons to be updated at each step is proposed. An algorithm using the criterion converges to more precise solutions with fewer updates as shown by simulation.< >
The scheme presented here finds the boundary of moving front scene by subtracting the image by the background saving in *** the boundary at hand,a new kind of motion compensation algorithm called region matching algor...
详细信息
The scheme presented here finds the boundary of moving front scene by subtracting the image by the background saving in *** the boundary at hand,a new kind of motion compensation algorithm called region matching algorithm is used to improve motion compensation *** based on the boundary information,an interframe interpolation method for reconstruction image omitted by subsampling in time domain is developed. Computer simulation showed that no conspicuous noise is observed at the bit rate about 64Kbit/s.
The authors present a novel neural network model for visual information processing. The model uses a hierarchical network with local connectivity as a stem network. This network generates hypotheses about the expected...
详细信息
The authors present a novel neural network model for visual information processing. The model uses a hierarchical network with local connectivity as a stem network. This network generates hypotheses about the expected image content, and then selectively uses small neural network modules on parts of the image to check these hypotheses. The resulting neural network is able to use different spatial resolutions, and is both modular and hierarchical. Applying this model to remotely sensed image classification (Landsat TM) is described. A slightly better classification accuracy was achieved at reduced computational cost, compared to classification without the model.< >
Dynamic neural networks are able to alter the topology of the underlying graph in order to provide better performance. A lot of interesting algorithms for topology learning have been published, but no formal descripti...
详细信息
暂无评论