Article Preview
Top2. Background
ISR reconstruction algorithms include difference-based (Bätz, Eichenseer et al., 2015; Wei, 2016), reconstruction-based (Papyan & Elad, 2015; Xu et al., 2013), and learning-based (Lai et al., 2017; Wang et al., 2022) methods. While simple and effective, the classical difference-based method relies heavily on prior knowledge of natural images, resulting in obvious jagged edges of reconstructed images and limited restoration of detailed information. The reconstruction-based approach utilizes powerful image priori, such as non-local self-similar (Dong et al., 2013), sparse (Yang et al., 2010) and denoising (Zhang et al., 2017) priori, which can flexibly rebuild an HR image of relatively high quality. However, due to limited prior knowledge and time consumption of the optimization process, the restoration effect of high-frequency details of images is poor. Most methods to reconstruct remote sensing images are based on this method, and their performance is limited.