Super-resolution (SR) image reconstruction from multiple low-
resolution (LR) frames have many applications, such as in remote sensing, surveillance,
and medical imaging.
In this paper, we show that the low rank property can in fact be constructed under MFSR framework. The idea is to consider each LR image as a downsampled instance of a different blurred and shifted HR image. Then when all these different HR images are properly aligned, they should give a low rank matrix; and therefore we can use a low rank prior to obtain a better solution.
||W1f1,W2f2,...Wpfp||*+∑i||gi-DKCifi||22