欢迎访问《兵工学报》官方网站,今天是

兵工学报 ›› 1999, Vol. 20 ›› Issue (1): 39-42.

• 论文 • 上一篇    下一篇

基于三阶相关量的目标图像不变性识别技术

祝奇、郑链、王克勇   

  1. 北京理工大学,北京,100081
  • 收稿日期:1997-11-01 修回日期:1998-09-01 上线日期:2014-12-25
  • 通讯作者: 祝奇

Translation Scale Rotation Invariancy (TSRI) Recognition of Target Image Based on Triple-Correlation

Zhu Qi, Zheng Lian, Wang Keyong   

  1. Beijing Institute of Technology, Beijing, 100081
  • Received:1997-11-01 Revised:1998-09-01 Online:2014-12-25
  • Contact: Zhu Qi

摘要: 重点在于应用基于三阶相关量的人工神经网络方法解决目标图像的平移( Translation)、尺度变换(Scale)、旋转(Rotation in-plane)不变性(Invariant)识别问题(TSRI)。该方法通过计算“零片”三阶相关量,对数极坐标变换和计算Fourier变换幅度谱,得到TSRI特征,并作为模式分类器的输入信号;模式分类器采用单层感知机。理论分析和对3类飞机目标的实验结果表明,该法具有抗噪声,高识别率,处理速度快,结构简单等特点,是一种有效的TSRI识别方法。

关键词: 图像识别 , 人工神经网络 , 自动目标识别 , 不变性模式识别

Abstract: The emphasis of this paper is laid on solving translation, scale and rotation —invariant target image recognition problems (TSRI) using ANN method based on triple-correlation. By calculating ‘0 slice’ triple-correlation, log-polar mapping, calcu?lating Fourier transform .amplitude-module, the TbRI feature is retrieved and used as input signal of pattern classifier, the single layer perception is used as a classifier. Theo?ry analyzing and results from experiment with three airplanes show that this TSRI recognition method is efficient, and has many advantages such as insensitive to noise, high recognition rate, high speed processing, simple in structure etc.

Key words: image recognition , artificial neural network , automatical target recogni?tion , invariant pattern recognition