针对存在平移、旋转和尺度关系的多传感器、多谱段遥感图像配准问题,提出了一种基于边缘特征提取和伪对数极坐标傅里叶变换(EPLPFFT)的频域配准技术。采用边缘检测等图像预处理技术。提取了图像的大部分共有显著特征,有效消除了图像间的灰度差异;利用伪对数极坐标离散傅里叶变换(DFT),改善了对数极坐标DFT的计算精度,从而可以通过对数极坐标频域配准技术配准经边缘提取后的特征图像。遥感图像的配准实验证实了这一方法的稳健性和配准精度。
Aiming at automatic registration of multisensor images and multispectral images, we present an Fourier-based registration algorithm to estimate translations, rotations, and scalings in these images. First,with edge detection,salient and distinctive features can be extracted. Next, interpolated by pseudo-log-polar discrete Fourier transform(DFT), the logpolar DFT values of images can be approximated more accurately than by conventional techniques. Finally,we can use the log-polar frequency registration algorithm to align the features of images. Experimental results with various kinds of remote sensing images have verified robustness and high accuracy of this algorithm.