Blind restoration of atmospherically degraded images by automatic best step-edge detection

Blind Restoration of Remote Sensing Images by a Combination of Automatic Knife-Edge Detection and Alternating Minimization

“…f and g represent the unknown sharp image and the observed image, which are the raw data without pre-focusing. h is the total shift-invariant atmospheric PSF of the remote sensing images, and is usually approximated by a Gaussian function [3][4][5]. n is the acquisition noise.…”

“…A great deal of effort has been made to improve the MTF measurement methods, which include the knife-edge method [7], the trap-based target method [7], and the pulse method [8]. Once the MTF is obtained, the remote sensing images can be restored by applying an MTF compensation (MTFC) technique [2,9] or other non-blind restoration methods [5,[10][11][12]. Because knife edges can usually be found in remote sensing images near features like bridges, glaciers, buildings, and even shadows, the knife-edge method can be used for many remote sensing applications.…”

“…Therefore, knife edges usually need to be manually selected and may not be ideal, or, in some cases, they may not exist. Previous studies have mainly focused on the extraction procedure of the MTF (or PSF), but the estimated MTF (or PSF) is often not very precise, and does not result in satisfactory results in the restoration procedure [5,7].…”

“…According to the blur identification stages, the blind restoration methods can be classified into two categories: (1) methods that estimate the PSF before the image is identified, as separate procedures [5,[16][17][18]; and (2) methods that estimate the PSF and the image simultaneously in a joint procedure [19][20][21][22][23][24][25][26][27][28][29][30][31][32][33]. In our work, we mainly focus on the second type of method.…”

“…(1) The AM framework is set up first to make the restoration process blind, and an automatic knife-edge detection method [5] is then used to extract the best knife edge. We then prove that the PSF estimated from the knife edges, although not very precise, is actually good enough to initialize the AM framework.…”