Evaluating super-resolution reconstruction of satellite images

Super-resolution reconstruction (SRR) consists in enhancing image spatial resolution given a single image or a bunch of images presenting the same scene at lower resolution. Potential benefits of applying SRR to satellite imagery are evident, as it may enhance the capacities of images characterized with lower resolution, shorter revisit times, and of higher availability. Unfortunately, the existing SRR approaches are not sufficiently mature to allow for broad real-life applications. One of the main obstacles we identify is that the evaluation procedure commonly adopted while developing new SRR techniques does not reflect the operational conditions. In this paper, we present our validation framework based on real satellite images acquired at different native resolutions, and we elaborate on measuring the reconstruction quality. We argue that this is critical to developing new and tuning the existing SRR methods to adapt them to real-world conditions. We investigate a number of well-established measures, and we also introduce our new metrics that allow for robust evaluation of the SRR outcome given a reference image of higher resolution. In addition to the quantitative tests, we also report qualitative results in which the reconstruction quality measures are used as an objective function in our evolutionary algorithms applied to adapt the SRR methods to a specific imaging model. Overall, the reported study proposes a new way to evaluate the SRR methods, which is an important step towards deploying them in practice.