US 12,272,027 B2
Image restoration method and apparatus
Jaeseob Shin, Seoul (KR); Sungul Ryoo, Seoul (KR); Sehoon Son, Seoul (KR); Hyeongduck Kim, Gyeonggi-do (KR); and Hyosong Kim, Seoul (KR)
Assigned to PIXTREE Inc., (KR)
Appl. No. 17/771,007
Filed by PIXTREE Inc., Seoul (KR)
PCT Filed Oct. 13, 2020, PCT No. PCT/KR2020/013937
§ 371(c)(1), (2) Date Apr. 21, 2022,
PCT Pub. No. WO2021/080233, PCT Pub. Date Apr. 29, 2021.
Claims priority of application No. 10-2019-0130542 (KR), filed on Oct. 21, 2019.
Prior Publication US 2023/0005114 A1, Jan. 5, 2023
Int. Cl. G06T 5/50 (2006.01); G06T 3/40 (2024.01); G06T 3/4053 (2024.01); G06T 5/00 (2024.01); G06T 5/20 (2006.01); G06T 5/70 (2024.01); G06V 10/77 (2022.01)
CPC G06T 5/50 (2013.01) [G06T 3/4053 (2013.01); G06T 5/20 (2013.01); G06T 5/70 (2024.01); G06V 10/7715 (2022.01); G06T 2207/20076 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/20212 (2013.01); G06T 2207/30168 (2013.01); G06T 2207/30201 (2013.01)] 15 Claims
OG exemplary drawing
 
1. An image restoration apparatus characterized by including:
a receiver for receiving an image; and
a processor for processing the image through convolution operation and non-linearization,
wherein the processor includes:
a learning module configured to receive a plurality of images having different resolutions and performs learning for each of the different resolutions to learn an independent restoration model for each resolution;
a selection module configured to, when receiving a distorted observed image y, analyze the resolution of the distorted observed image y to select a suitable restoration model; and
an image restoration module configured to generate a restored image x  in which the observed image is restored by using a restoration model corresponding to the resolution of the observed image y among independent restoration models that are different for each resolution,
wherein the learning module includes:
a generator configured to generate a fake image that increases the resolution of the input image by a preset resolution;
a discriminator configured to receive the generated fake image and the original image x, determine whether the generated fake image is an original (real) image or a fake image to calculate a score, and generate the restoration model for each resolution based on the calculated score.