US 11,748,853 B2
Method and architecture for blind image deconvolution
Marios Savvides, Wexford, PA (US); Raied Aljadaany, Pittsburgh, PA (US); and Dipan Kumar Pal, Pittsburgh, PA (US)
Assigned to Carnegie Mellon University, Pittsburgh, PA (US)
Filed by CARNEGIE MELLON UNIVERSITY, Pittsburgh, PA (US)
Filed on Apr. 27, 2021, as Appl. No. 17/241,379.
Claims priority of provisional application 63/016,734, filed on Apr. 28, 2020.
Prior Publication US 2021/0334939 A1, Oct. 28, 2021
Int. Cl. G06T 5/00 (2006.01); G06T 5/50 (2006.01); G06T 11/00 (2006.01); G06T 1/20 (2006.01)
CPC G06T 5/002 (2013.01) [G06T 1/20 (2013.01); G06T 5/50 (2013.01); G06T 11/005 (2013.01); G06T 11/006 (2013.01); G06T 2207/20024 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/20182 (2013.01); G06T 2211/424 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A computer-implemented method for performing blind deconvolution of corrupted images comprising:
iteratively performing a minimization of a sum of a data fidelity term and an image prior term using a Douglas-Rachford algorithm;
wherein a proximal operator of the—data fidelity term is represented in each iteration by a deep neural network trained to approximate the data fidelity proximal operator;
wherein a proximal operator of the image prior term is represented in each iteration by a deep neural network trained to approximate the image prior proximal operator; and
wherein the deep neural network trained to approximate the data fidelity proximal operator is different for each iteration and further wherein the deep neural network trained to approximate the image prior proximal operator is different for each iteration.