Zero-Shot Super Resolution for Satellite Remote Sensing Images
Junzhi Yang · 2019 IEEE International Conference on Signal, Information and Data Processing (ICSIDP) · 2019
Due to the great distance between the satellite platform and the ground targets of interests, various factors such as weather and light conditions degrade the image quality captured by the sensors. Single image super resolution (SISR) has always been an important research field in satellite remote sensing image process. With large amounts of low resolution to high resolution (LR-HR) pairs generated by predefined downscaling process, usually noise-free bicubic interpolation with anti-alias, recent deep learning models have shown great improvements on natural image under such ideal conditions. However the real-world degrading process of remote sensing (RS) images differs greatly from the ideal configuration, the state-of-the-art models lose their power when faced with the practical problems. This letter adapted the recently proposed zero shot super resolution scheme to accommodate satellite RS images, both quantitative and qualitative results show that our method outperforms the previous DL-based models by a clear margin.