电子学报 ›› 2000, Vol. 28 ›› Issue (1): 48-52.

• 论文 • 上一篇    下一篇

基于Gibbs图像模型的细胞神经网络模板设计法

赵建业, 王 新, 余道衡   

  1. 北京大学电子学系,视觉与听觉信息处理国家重点实验室,北京 100871
  • 收稿日期:1998-05-12 修回日期:1999-04-15 出版日期:2000-01-25 发布日期:2000-01-25

A Template Design Method for Cellular Neural Networks Based on Gibbs Image Model

ZHAO Jian-ye, WANG Xin, YU Dao-heng   

  1. Dept.of Electronics,Peking University,Beijing 100871,China
  • Received:1998-05-12 Revised:1999-04-15 Online:2000-01-25 Published:2000-01-25

摘要: 在本文中我们提出了将细胞神经网的模板设计与对Gibbs统计图像模型的分析结合起来,形成一个用于解决诸如图像恢复、分割、边缘检测、重建等问题的细胞神经网络,新的通用设计方法——能量函数法.这种方法将权的学习、硬件退火等方法有机地结合起来,从而合理地得到细胞神经网络模板形式,较好地解决一些图像处理问题.

关键词: Gibbs图像模型, 细胞神经网, 模板, 学习, 硬件退火

Abstract: In this paper,we note that there is a kind of relation between CNN and Gibbs image model.So we proposed a new approach for CNN's template design based on Gibbs image model,this approach is valid to such questions as segmentation,edge detection and restoration.We also discuss the learning algorithm and hardware annealing joined with the new approach.Simulations of an example (binary segmentation) are shown in order to validate the approach.

Key words: Gibbs image model, cellular neural network, template, learning, hardware annealing

中图分类号: