论文部分内容阅读
利用超分辨率重建技术来提高遥感图像的空间分辨率,成本低并且可充分利用现有的数据。但必须要解决好三个问题:1)如何实现高精度的自动配准;2)如何仅用2～3张图像进行快速准确重建;3)如何快速鲁棒的处理大尺寸图像。针对上述问题,该文在以下两个方面对传统的图像超分辨率方法凸集投影法(POCS)作了改进:首先,采用SIFT特征进行图像配准,并利用层次结构kd-tree加速特征匹配;其次,提出了一种基于外存的策略,实现了大尺寸遥感图像的超分辨率重建。通过实验可以看出,使用该文方法,仅用2帧原始图像就可高效的重建出一幅具有更多细节特征的图像,图像质量比原始图像有了明显提高。
The use of super-resolution reconstruction technology to improve the spatial resolution of remote sensing images, low cost and make full use of existing data. However, we must solve three problems: 1) how to achieve high-precision automatic registration; 2) how to use 2 to 3 images for fast and accurate reconstruction; 3) how to deal with large-size images quickly and robustly. In order to solve the above problems, this paper improves the traditional image super-resolution convex projection method (POCS) in the following two aspects: Firstly, the SIFT feature is used for image registration, and the hierarchical kd-tree is used to accelerate the feature matching Secondly, a memory-based strategy is proposed to realize the super-resolution reconstruction of large-scale remote sensing images. It can be seen from the experiment that using this method, only two frames of original image can efficiently reconstruct an image with more detailed features, and the image quality is obviously improved compared with the original image.