• 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

J4 ›› 2013, Vol. 35 ›› Issue (1): 119-123.

• 论文 • 上一篇    下一篇

基于归一化边缘互信息与自适应加速粒子群的图像配准方法

冯雪芳,吴锡生   

  1. (江南大学物联网工程学院,江苏 无锡 214122)
  • 收稿日期:2011-11-21 修回日期:2012-02-25 出版日期:2013-01-25 发布日期:2013-01-25
  • 作者简介:冯雪芳(1986),女,山西太原人,硕士生,研究方向为人工智能与模式识别。
  • 基金资助:

    国家自然科学基金资助项目(60973094)

Adaptive accelerated particle swarm algorithm for image registration based on normalized edge mutual information

FENG Xuefang,WU Xisheng   

  1. (School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China)
  • Received:2011-11-21 Revised:2012-02-25 Online:2013-01-25 Published:2013-01-25

摘要:

传统的归一化互信息配准方法未利用图像的空间信息,当图像中混有一定噪声时,会出现误配准。边缘是图像最基本的特征之一,为了改进归一化互信息方法,提高图像配准的精度,加快收敛速度,将图像的边缘信息与灰度信息自适应地结合,形成归一化边缘互信息测度(NCMI),提出一种基于加速因子的自适应加速粒子群优化算法(AAPSO)来优化基于NCMI测度的图像配准。AAPSO算法通过对解排序,将指定数量的劣解进行进化加速来引导粒子的飞行,并对自适应惯性权重公式加以改进,提高了算法的收敛性,防止早熟收敛并增加优化解的多样性,同时加入加速因子来提高收敛速度。实验结果表明,该方法配准精度高,速度快,具有较强的实用性。

关键词: 图像配准, 归一化互信息, 归一化边缘互信息, 粒子群优化, AAPSO算法

Abstract:

The traditional normalized mutual information neglects the spatial information, so the registration result will be incorrect when the image mixes with noises. Edge is the basic character of image, in order to solve the drawback of the normalized mutual information method, improve precision and speed up the convergence, we combines the image edge information with the gray information adaptively to form the normalized edge mutual information measure (NCMI) and propose an adaptive accelerate particle swarm optimization algorithm (AAPSO) based on the accelerated factor. The AAPSO is used for image registration based on the NCMI. By sorting the solutions, a specified number of worst solutions will be forced to accelerate in order to determine the direction of global solution, and we also improve the adaptive inertia weight formula, thereby it improves convergence, prevents premature convergence and increases the diversity of the optimal solution. Meanwhile, the AAPSO algorithm adds the accelerated factor to improve convergence speed. Result shows that the method has a high registration precision, fast registration speed, and it has strong applicability.

Key words: image registration;normalized mutual information;normalized edge mutual information;particle swarm algorithm;AAPSO algorithm