Satellite Imagery Superresolution Based on Optimal Frame Accumulation

As actual spatial resolution continues to be a primary bottleneck for satellite imagery interpretation, the aim of resolution enhancement remains very urgent. The resolution enhancement is especially important for medium- and high-resolution satellite systems targeted at small-size objects observation. One of the efficient methods for satellite imagery spatial resolution enhancement is superresolution, when several subpixel-shifted relative to each other low-resolution frames are combined into a single image that accumulates information from all input ones. Our team has previous experience implementing this kind of superresolution in satellite imagery preprocessing. A twofold formal and 48% actual resolution enhancement has been achieved by combining two to four frames into a single image. However, the potential of the formal resolution duplication model is exhausted now. Therefore, we propose scaling up the discretization of the joint subpixel grid over the input imageset from one-half to one-third of the pixel. Accordingly, a ninefold increase in the number of subpixels requires additional information, which is obtained from a larger number of input frames. At the moment, some remote sensing satellite systems can generate many highly-overlapping frames, such as SkySat-1/2, Gaofen-1/6, Gaofen-4/13, Jilin-1, etc. As a result, too many frames are available, which leads to the problem of selecting their optimal subset. We solve this problem by optimizing the joint subpixel grid position and achieving the maximum signal-to-noise ratio in the resulting image using the matrix of mutual subpixel shifts of the input frameset.