US 11,783,184 B2
Kernel prediction with kernel dictionary in image denoising
Federico Perazzi, San Francisco, CA (US); Zhihao Xia, St. Louis, MO (US); Michael Gharbi, San Francisco, CA (US); and Kalyan Sunkavalli, San Jose, CA (US)
Assigned to Adobe Inc., San Jose, CA (US)
Filed by Adobe Inc., San Jose, CA (US)
Filed on Feb. 2, 2022, as Appl. No. 17/590,995.
Application 17/590,995 is a continuation of application No. 16/686,978, filed on Nov. 18, 2019, granted, now 11,281,970.
Prior Publication US 2022/0156588 A1, May 19, 2022
This patent is subject to a terminal disclaimer.
Int. Cl. G06N 3/08 (2023.01); G06T 15/50 (2011.01); G06T 5/00 (2006.01); G06N 20/10 (2019.01)
CPC G06N 3/08 (2013.01) [G06N 20/10 (2019.01); G06T 5/002 (2013.01); G06T 15/50 (2013.01)] 20 Claims
OG exemplary drawing
 
9. A system, comprising:
one or more processors; and
a memory having stored thereon instructions that, when executed by the one or more processors, cause the one or more processors to:
receive, as input to a neural network, a reference image including noise, the reference image comprising a plurality of pixels;
identify at least one feature of the reference image;
generate, based on the at least one feature of the reference image, a base kernel; and
include the base kernel in a kernel dictionary of the neural network, wherein the kernel dictionary is usable by the neural network to denoise an input image,
wherein the neural network is trained for generating a denoising kernel for the input image based on a combination of the base kernel with a coefficient vector.