| DCT编码的一种神经网络方法 |
| |
| 引用本文： | 王卫,蔡德钧,万发贯.DCT编码的一种神经网络方法[J].华中科技大学学报(自然科学版),1993(3). |
| |
| 作者姓名： | 王卫 蔡德钧 万发贯 |
| |
| 作者单位： | 华中理工大学电子与信息工程系
(王卫,蔡德钧),华中理工大学电子与信息工程系(万发贯) |
| |
| 基金项目： | 国家自然科学基金资助,中科院自动化所国家模式识别实验室资助项目 |
| |
| 摘 要： | 提出了一种求非完全DCT逆矩阵的神经网络方法,并将它用于编码图像的重建.文中对Hopfield神经网络的选代规则进行了修改,保证了网络在宽松的条件下仍能稳定地收敛.另外,针对Hopfield神经网络的局部极小问题,提出了一种扰动算法,使网络的迭代避开局部极小点,接近全局最小点.最后,给出了计算机模拟实验结果,并与传统的神经网络方法进行了比较.
|
| 关 键 词： | 神经网络 编码 图像数据压缩 能量函数 随机扰动 |
A Neural Network Approach to DCT Coding |
| |
| Wang Wei Cai Dejun Wan Faguan.A Neural Network Approach to DCT Coding[J].JOURNAL OF HUAZHONG UNIVERSITY OF SCIENCE AND TECHNOLOGY.NATURE SCIENCE,1993(3). |
| |
| Authors: | Wang Wei Cai Dejun Wan Faguan |
| |
| Institution: | Wang Wei Cai Dejun Wan Faguan |
| |
| Abstract: | A neural network approach to calculating an incomplete discrete cosine transform (DCT) inverse matrix is presented. The incomplete DCT inverse matrix has been applied to the reconstruction of an encoded image. The neural network used is a Hopfield one, which requires some prerequisites for the convergence. Iteration rules for the Hopfield neural network has been modified in order to guarantee a stable convergence of network under "loose" conditions. A perturbation technique for enabling the network to avoid local minima and reach global minimum during iteration is proposed. Computer simulation results and a comparison with other neural network approaches show that the method proposed is much better than earlier ones in performance. |
| |
| Keywords: | neural network coding image data compression energy function random perturbation |
| 本文献已被 CNKI 等数据库收录！ |
