US 12,340,439 B1
Method and apparatus for adaptively extrapolating image through graph-based patch rearrangement, and computer program
Jong Won Choi, Gyeongsangnam-do (KR); Hyoung Jun Lim, Gyeonggi-do (KR); Joo Young Lee, Seoul (KR); and Soo Hyun Park, Gangwon-do (KR)
Assigned to CHUNG ANG UNIVERSITY INDUSTRY ACADEMIC COOPERATION FOUNDATION, Seoul (KR)
Filed by CHUNG ANG UNIVERSITY INDUSTRY ACADEMIC COOPERATION FOUNDATION, Seoul (KR)
Filed on Sep. 23, 2024, as Appl. No. 18/892,812.
Claims priority of application No. 10-2023-0189564 (KR), filed on Dec. 22, 2023.
Int. Cl. G06T 11/00 (2006.01); G06T 5/20 (2006.01); G06T 5/70 (2024.01); G06V 10/26 (2022.01); G06V 10/50 (2022.01); G06V 10/82 (2022.01)
CPC G06T 11/00 (2013.01) [G06T 5/20 (2013.01); G06T 5/70 (2024.01); G06V 10/26 (2022.01); G06V 10/50 (2022.01); G06V 10/82 (2022.01); G06T 2207/20021 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01)] 9 Claims
OG exemplary drawing
 
1. A method for adaptively extrapolating an image, which is performed on a computing device including one or more processors and a memory that stores one or more programs to be executed by the one or more processors, the method comprising:
dividing an input image into a grid pattern and constructing patches of at least some regions as a pool of candidate patches;
deriving a graph-based target feature by replacing one patch in the constructed pool of candidate patches with a blank patch;
performing learning so that the derived graph-based target feature is to be closer to a feature of the patch replaced by the blank patch than a feature derived from the regions other than the pool of candidate patches and farther away from the feature derived from regions other than the pool of candidate patches than the feature of the patch replaced by the blank patch through contrastive learning between the derived graph-based target feature, the feature of the patch replaced by the blank patch, and the feature derived from the regions other than the pool of candidate patches;
selecting one of the patches included in the input image as a patch to be inserted into a grid-based pool of extrapolated region based on the learning result; and
repeatedly performing the selecting by inserting the selected patch to be inserted into the pool of extrapolated region and reselecting a pool of extrapolated region, wherein
in the selecting, a patch that has the smallest sum of a distance between pixel values of an edge of the patch and those of other adjacent patches that are encountered when inserted into the pool of extrapolated region and a distance between patch features of the patch and those of other adjacent patches is selected from among patches included in the input image.