计算机工程 ›› 2012, Vol. 38 ›› Issue (9): 177-179.doi: 10.3969/j.issn.1000-3428.2012.09.053

• 人工智能及识别技术 • 上一篇    下一篇

基于多维联想记忆神经网络的图像回忆

杨家稳1,2,孙合明1   

  1. (1. 河海大学理学院,南京 210098;2. 滁州职业技术学院,安徽 滁州 239000)
  • 收稿日期:2011-07-14 出版日期:2012-05-05 发布日期:2012-05-05
  • 作者简介:杨家稳(1972-),男,副教授、硕士,主研方向:神经网络,图像处理;孙合明,副教授、博士
  • 基金项目:
    安徽省高校省级自然科学基金资助项目(KJ2011B119)

Image Recalling Based on Multidimensional Associative Memory Neural Network

YANG Jia-wen   1,2, SUN He-ming   1   

  1. (1. College of Science, Hohai University, Nanjing 210098, China; 2. Chuzhou Vocational and Technical College, Chuzhou 239000, China)
  • Received:2011-07-14 Online:2012-05-05 Published:2012-05-05

摘要: 多维联想记忆神经网络在高噪声情况下图像回忆效果差。针对该问题,将图像矩阵垂直分成4个同型小矩阵,依次将4个小矩阵垂向聚合成一个新矩阵,以新矩阵的列向量作为库向量。数值实验结果表明,相比2个列向量构成的库向量,以4个列向量构成的库向量进行回忆的灰度图像更清晰且效率更高。

关键词: 多维联想记忆, 神经网络, 投影, 库向量, 图像回忆, 图像矩阵

Abstract: Multidimensional associative memory neural networks can be used for image recalling. When the image is corrupted by high noise, the recalling image is not clear using the traditional method. In order to make the recalling image clearer, one library vector made up of four column vectors is used in the recalling image to take the place of the other traditional library vectors made up of two column vectors. That is, a new matrix is formed by vertically dividing the mage matrix into four small matrices of the same order and vertically concatenating the four matrices in order. A column vector of the new matrix is regarded as a library vector. Numerical examples show that the restored image is clearer and the recalling process spends less time when the former library vector is used.

Key words: multidimensional associative memory, neural network, projection, library vector, image recalling, image matrix

中图分类号: