基于参叉像元和非均匀B样条曲面的遥感图像超分辨率重建

Super- resolution reconstruction of remote sensing image based on staggered pixels and non- uniform B- spline curved surface

遥感图像之间的信息互补可以提高图像分辨率,但插值方法易使图像边界模糊、部分细节信息丢失。针对这一问题,提出一种基于参叉像元与非均匀B样条插值相结合的遥感图像超分辨率重建方法。利用经灰度匹配和亚像元级几何配准的2景低分辨率图像,通过参叉交错像元采样到原图像网格2倍的网格中;对于没有值的坐标处用三次B样条插值,插值时选用非均匀的节点参数化方法,曲面图像网格点由邻域36个已知像元组成;在求解待插值点参数值时引入平行线法和黄金分割法迭代寻找最优值,使插值更准确;最后对插值后的图像进行复原处理,重建可视效果更好的"高"分辨率图像。对实验图像的评价表明,用本文方法重建的图像在清晰度、信息量、信噪比和分辨率等方面均有较大的提高。

The adoption of information from two images can improve resolution of images;nevertheless, the method based on interpolation is apt to make fussy boundary or cause partial loss of detailed information. To solve this problem, the authors employed an approach to super-resolution reconstruction based on staggered pixels and non-uniform B -spline surface. Two frames of images with low resolution, which were matched and geometrically registered at sub-pixel level,were adopted,and then these two images were staggered and re-sampled to double times of the origin grid;the positions with non -value were interpolated and filled with tri -B -spline surface interpolation,and non-uniform node parametric method was adopted. And the curved surface was composed of 36 known neighborhood pixels. For solving the value of interpolated points,the authors introduced parallel and golden section methods to iterate searching for the optimal value,which made interpolation more accurate;the last but not unimportant, the interpolated images were restored to reconstruct better visualization high resolution images. The assessment of the results demonstrates that this approach can considerably improve definition, quantity of information, signal-to-noise ratio and resolution.