摘要
文章提出了一种平移、旋转及尺度不变光学神经网络识别系统方案。该系统可分为两层，第一层为预处理层，利用圆-傅里叶-径-梅林变换提取输入模式不变性特征，并采用修正格雷码对其进行二值编码；第二层为异联想记忆层，利用模式间异联想（IHA）网络对输入模式的二值编码进行联想记忆。最后给出了计算机模拟结果。
A shift, rotation and scale invariant neural network is proposed in this paper. The network can be divided into two levels-preprocess level and hetero-association memory level. In the first level, the input patterns are sampled and binary-coded by using Circular- Fourier- Radial - Mellin transformation and modified Gray one respec tively. In the second level, the recall of the input pattern are processed by using internpattern hetero-association (IHA) model. The computer simulation results are also provided.
出处
《红外与激光工程》
EI
CSCD
1997年第5期17-21,共5页
Infrared and Laser Engineering
基金
南京理工大学科研发展基金
关键词
模式识别
光学神经网络
二值编码
Pattern recognition
Optical neural network
Binary-coding
