US 12,169,913 B2
Method and system for a high-frequency attention network for efficient single image super-resolution
Ding Liu, Los Angeles, CA (US)
Assigned to LEMON INC., Grand Cayman (KY)
Filed by Lemon Inc., Grand Cayman (KY)
Filed on Feb. 10, 2022, as Appl. No. 17/669,072.
Prior Publication US 2023/0252605 A1, Aug. 10, 2023
Int. Cl. G06K 9/00 (2022.01); G06T 5/20 (2006.01); G06T 5/50 (2006.01); G06T 5/70 (2024.01)
CPC G06T 5/70 (2024.01) [G06T 5/20 (2013.01); G06T 5/50 (2013.01); G06T 2207/20016 (2013.01); G06T 2207/20024 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] 17 Claims
OG exemplary drawing
 
1. A single image reconstruction method comprising:
extracting a plurality of features from an original image input into a CNN to generate a feature map;
restoring one or more high-frequency details of the original image via an efficient residual block (ERB) and a high-frequency attention block (HFAB) configured to assign a scaling factor to one or more high-frequency areas, wherein restoring one or more high-frequency details further comprises:
reducing, by a first 3×3 convolutional layer of the HFAB, the feature map from a first channel dimension to a second channel dimension to generate first layer output information;
detecting, via a Laplacian filter of the HFAB, second layer output information including the one or more high-frequency details of the original image based on the first layer output information;
expanding, by a second 3×3 convolutional layer of the HFAB, the second layer output information from the second channel dimension to the first channel dimension to generate third layer output information;
performing a sigmoid operation on the third layer output information to determine fourth layer output information; and
performing an element-wise multiplication of the feature map and the fourth layer output information to determine HFAB output used to determine reconstruction input information;
generating the reconstruction input information by performing an element-wise operation on the one or more high-frequency details and cross-connection information from the feature map; and
performing, by the CNN, an enhancement operation on the reconstruction input information to generate an enhanced image.