US 11,941,862 B2
Apparatus, method, and storage medium
Yasuhiro Okuno, Tokyo (JP)
Assigned to CANON KABUSHIKI KAISHA, Tokyo (JP)
Filed by CANON KABUSHIKI KAISHA, Tokyo (JP)
Filed on Mar. 23, 2021, as Appl. No. 17/209,922.
Claims priority of application No. 2020-066048 (JP), filed on Apr. 1, 2020.
Prior Publication US 2021/0312593 A1, Oct. 7, 2021
Int. Cl. G06V 10/44 (2022.01); G06F 18/21 (2023.01); G06F 18/214 (2023.01); G06N 3/04 (2023.01); G06N 3/08 (2023.01); G06T 5/00 (2006.01); G06T 5/50 (2006.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01)
CPC G06V 10/454 (2022.01) [G06F 18/214 (2023.01); G06F 18/2193 (2023.01); G06N 3/04 (2013.01); G06N 3/08 (2013.01); G06T 5/002 (2013.01); G06T 5/50 (2013.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01); G06T 2207/20084 (2013.01)] 19 Claims
OG exemplary drawing
 
1. An apparatus configured to train an image processing model for removing noise from an image containing the noise, the apparatus comprising:
at least one processor;
a memory coupled to the at least one processor storing instructions that, when executed by the at least one processor, cause the at least one processor to function as:
an acquisition unit configured to acquire a low-noise image and a plurality of high-noise images having more noise than the low-noise image, the low-noise image being training data for training the image processing model comprising one neural network, the plurality of high-noise images corresponding to a scene in the low-noise image and each having different noise patterns;
an error calculation unit configured to calculate each of errors between a plurality of output images and the low-noise image, each of the plurality of output images being acquired by inputting a different one of the plurality of high-noise images to the one neural network;
a stability calculation unit configured to calculate a stability against the noise based on an error between the plurality of output images; and
a training unit configured to train the one neutral network using a loss function based on the calculated errors and the calculated stability.