Super-resolution Image Metrics and Model-based Distortion Inversion for Turbulence

SSCI and AER propose a novel model-based approach that aims to exploit the best concepts from current super-resolution methodologies to achieve real-time super-resolved video imaging: model-based inversion image reconstruction, image quality metrics for "lucky image" selection, and (an-)isoplanatic distortion models. Our aim will be to select models and methods in Phase 1 that are most likely to be transitionable to inexpensive, lightweight, highly-parallelized Graphics Processing Units (GPUs) in Phase 2 for real-time processing. High resolution imaging of terrestrial targets and scenes at very long ranges from aerial or ground-based imaging platforms can be hampered by distortions of the optical paths from heat- and wind-induced turbulence and haze. Such distortions impact the fundamental diffraction limit of the overall optical system from scene to focal plane array and can make it challenging to resolve targets. However, most existing model-based methods for super-resolution are highly dependent on known or easily estimable blur models and do not extend easily to distortions from anisoplanatic atmospheric turbulence. Also, "lucky imaging" methods, which select particularly high-quality images or subimages from very high-data rate streams, can require specialized high-data rate cameras, significant real-time computation rates, and are not likely to be suitable for low-light applications where signal-to-noise ratios for short duration pixel integrations can be very low.