基于改进MSER与SIDs特征的遥感图像配准算法

Remote Sensing Image Registration Algorithm Based on Improved Maximally Stable Extremal Regions and Salient Image Disks

由于遥感图像存在仿射变形且纹理结构丰富,基于单一特征往往难以实现高效、高精度的配准利用显著图像圆盘(SIDs)易于自动提取的特点,并结合改进的极大稳定极值区域(MS-ER)实现仿射归一化,提出了一种新的仿射不变的遥感图像配准算法.该算法利用多尺度各向同性相匹配滤波(Multiscale Isotropic Matched Filtering,MIMF)检测稳定SIDs特征,结合高斯函数来精确定位SIDs极值点,再利用匹配好的MSER实现仿射归一化.实验结果表明,所提算法能较好适应较大尺度、光照及视角变化,具有较高匹配效率和配准精度,尤其是大尺寸遥感图像,算法效率提升更加明显.

Owing to affine deformation and abundant texture structure in remote sensing image,it is often more difficult to achieve high-efficiency and high-accuracy registration based on single feature.Considering that salient image disks(SIDs)is easy to automatically extract,for which the improved Maximally Stable Extremal Regions(MSER)can be used to realize affine normalization.To propose a new remote sensing image registration algorithm of affine invariant.Stable SIDs characteristics is detected by multiscale isotropic matched filtering(MIMF),while using the Gaussian function to accurately locate SIDs extreme value points.Then affine normalization can be achieved by the matched MSER.It is experimentally shown that the proposed algorithm can better adapt to the larger scale,illumination and the angle of view changes with a higher accuracy matching efficiency and registration accuracy.Especially,algorithm efficiency is more obvious in large remote sensing image registration.