基于梯度互信息与均匀设计PSO的航拍图像配准

Aerial Image Registration Based on Gradient Mutual Information and Uniform Designed PSO

基于互信息的图像配准算法精度较高,且易于实现自动配准,适合于无人机航拍图像配准.但在利用粒子群算法搜索最佳配准参数时,随机分布的初始粒子群容易集中在参数空间的某一区域,使算法陷入局部极值而得不到准确的配准结果;针对这一问题,提出一种新的图像配准算法,算法采用均匀设计思想改进粒子群算法,使初始粒子群能够均匀分布在搜索空间中,从而获得最佳配准参数;同时采用梯度互信息作为相似性测度,降低算法陷入局部极值的可能性.实验表明,新的算法能够更快更优地得到理想的航拍图像配准结果,验证了算法的有效性.

Image registration based on mutual information has high accuracy,and easy to realize automatic image registration,so it is fit for UAV aerial image registration.However,while using PSO to find the optimal registration parameters,the randomized initial particles easily gather in a certain spatial area,leading to the algorithm trapped in local extreme,and can't get accurate registration result.To solve the problem,the paper proposes a new image registration algorithm.The new algorithm merges uniform design idea into PSO,making the initial particles distributed uniformly in search space so as to search the global optimal registration parameters better.Meanwhile the paper adopts gradient mutual information as the similarity metric to reduce the possibility of local extremum.Experimental results demonstrate that the new algorithm can get fast and accurate registration results,and verifies its effectiveness.