论文部分内容阅读
针对目前立体影像匹配时普遍存在的配准稳定性与配准精度不能同时兼顾问题,该文采用金字塔和分块策略,首先在多尺度空间下进行粗匹配,获得大量的初始特征点集;然后通过限制金字塔影像的阶数对初始特征点进行坏点剔除和滤点均匀化来获得良好的初始特征点;最后通过进行分块影像并行化SIFT算法匹配来实现精确匹配,提高效率。实验结果证明,此遥感立体影像并行稳定匹配方法能够提高匹配的效率,对影像的旋转、缩放和视角变换具有更强的抵抗性,其配准精度较高,适用于大范围的遥感立体影像的匹配。
Aiming at the problem that the common registration stability and registration accuracy can not be considered at the same time in the current stereo image matching, this paper uses pyramid and block strategy, first, rough matching is performed in multi-scale space to obtain a large number of initial feature point sets; then The initial feature points were filtered out and the initial initial feature points were obtained by limiting the order of the pyramid image. Finally, the matching algorithm of block image parallelization SIFT was used to achieve the exact matching and improve the efficiency. The experimental results show that the remote sensing stereo image parallel stable matching method can improve the matching efficiency, resist the rotation, scaling and viewing angle transformation of the image with high registration precision and is suitable for a wide range of remote sensing stereo images match.