US 12,170,785 B2
Method and apparatus for adaptive image preprocessing and reconstruction
Sang-Woon Kwak, Daejeon (KR); and Joung-Il Yun, Daejeon (KR)
Assigned to ELECTRONICS AND TELECOMMUNICATIONS RESEARCH INSTITUTE, Daejeon (KR)
Filed by ELECTRONICS AND TELECOMMUNICATIONS RESEARCH INSTITUTE, Daejeon (KR)
Filed on Nov. 3, 2022, as Appl. No. 17/980,135.
Claims priority of application No. 10-2021-0164317 (KR), filed on Nov. 25, 2021; and application No. 10-2022-0040527 (KR), filed on Mar. 31, 2022.
Prior Publication US 2023/0164337 A1, May 25, 2023
Int. Cl. H04N 19/42 (2014.01); H04N 19/103 (2014.01); H04N 19/132 (2014.01)
CPC H04N 19/42 (2014.11) [H04N 19/103 (2014.11); H04N 19/132 (2014.11)] 10 Claims
OG exemplary drawing
 
1. A method for adaptive image preprocessing and reconstruction, comprising:
preprocessing an input image;
encoding and decoding the preprocessed image; and
reconstructing the encoded and decoded image, wherein
preprocessing the input image comprises using a preprocessing kernel generated based on a control parameter indicating a weight for human vision and machine vision,
reconstructing the encoded and decoded image is performed using a reconstruction network corresponding to a value of the control parameter,
reconstructing the encoded and decoded image is performed using multiple models respectively trained using a loss function corresponding to the control parameter, and
the reconstruction network is trained by giving priority to a loss function for performing a task of the machine vision as the value of the control parameter is larger.