Compressed sensing image restoration based on data-driven multi-scale tight frame

It has been shown that redundant signal representation, e.g. tight frame, plays important role in compressed sensing image restoration. In order to get a good sparse representation, one has made enduring efforts to pursue tight frames. Although there are some tight frames under which a type of images has a good sparse approximation, another type of images may not have sparse approximation because of the images’ great difference in structure. This paper presents a novel compressed sensing image restoration method based on data-driven multi-scale tight frame. This method derives a discrete multi-scale tight frame system adaptive to the original image from the input compressed sensing image. Such an adaptive tight frame construction scheme is applied to compressed sensing image restoration. The experimental results show our approach’s efficiency.