基于非线性空间及空间约束的遥感图像配准

Improved SIFT Multispectral Remote Sensing Image Registration

针对遥感图像对之间存在显著的非线性强度差异,鲁棒性不强,且缺少足够多的关键点对应关系等问题,提出了一种遥感图像配准算法。通过非线性扩散方程和快速显式扩散方法构建空间,避免高斯尺度空间的边界模糊;为克服遥感图像之间强度差异,改进了梯度幅值和方向的计算方法;为进一步改善配准结果,建立空间约束,包括点和线约束;利用快速抽样一致性算法消除误匹配,并求得相似变换模型参数,完成图像配准。实验结果表明,上述算法克服非线性强度差异问题,鲁棒性增强,与其它几种先进的算法相比,具有更高的精度和更多正确的对应关系,配准性能得到明显改善。

In view of the significant non-linear intensity difference between remote sensing image pairs,weak robustness,and insufficient key point correspondence,a remote sensing image registration algorithm is proposed.In order to overcome the intensity difference between remote sensing images,the calculation method of gradient magnitude and direction was improved.To further improve the registration,spatial constraints including point and line constraints were established.The fast sampling consistency algorithm was used to eliminate mismatches,and the parameters of similar transformation model were obtained to complete image registration.The experimental results show that the algorithm can overcome the problem of non-linear strength difference and enhance the robustness.Compared with other advanced algorithms,the algorithm has higher accuracy and more correct correspondence relations,and the registration performance has been significantly improved.