US 11,694,083 B2
Signal translation system and signal translation method
Ho Yong Jang, Suwon-si (KR); Kwang Min Choi, Seoul (KR); Se Kyung Kim, Seoul (KR); and Da Woon Kim, Seoul (KR)
Assigned to ALETHIO CO., Daejeon (KR)
Filed by ALETHIO CO., Daejeon (KR)
Filed on Oct. 15, 2018, as Appl. No. 16/160,859.
Claims priority of provisional application 62/572,507, filed on Oct. 15, 2017.
Claims priority of application No. 10-2018-0050390 (KR), filed on May 2, 2018; and application No. 10-2018-0120762 (KR), filed on Oct. 10, 2018.
Prior Publication US 2019/0114462 A1, Apr. 18, 2019
Int. Cl. G06N 3/084 (2023.01); G06F 40/40 (2020.01); G06V 40/20 (2022.01); G06F 18/211 (2023.01); G06F 18/2411 (2023.01); G06F 18/2135 (2023.01); G06V 10/771 (2022.01); G06V 10/77 (2022.01); G06V 10/82 (2022.01); G06T 11/00 (2006.01); G06N 20/00 (2019.01); G06N 3/02 (2006.01)
CPC G06N 3/084 (2013.01) [G06F 18/211 (2023.01); G06F 18/21355 (2023.01); G06F 18/2411 (2023.01); G06F 40/40 (2020.01); G06T 11/003 (2013.01); G06V 10/771 (2022.01); G06V 10/7715 (2022.01); G06V 10/82 (2022.01); G06V 40/20 (2022.01); G06F 2218/10 (2023.01); G06F 2218/14 (2023.01); G06N 3/02 (2013.01); G06N 20/00 (2019.01)] 7 Claims
OG exemplary drawing
 
1. A method of translating a source image of a first domain into a virtual image of a second domain which is different from the first domain, the method comprising:
receiving a source image of a first domain;
obtaining a destination image of a second domain;
identifying a plurality of features in the source image of the first domain and features of the destination image;
translating the source image of the first domain into a first virtual image of the second domain using a pre-trained neural network, including:
classifying the plurality of features in the source image of the first domain into effective features of the source image and erroneous features of the source image based on whether each feature of the plurality of features corresponds to the features of the destination image of the second domain, wherein the effective features correspond to the features of the destination image and the erroneous features do not correspond to the features of the destination image;
generating the first virtual image of the second domain by synthesizing the effective features of the source image with the features of the destination image and removing the erroneous features of the source image; and
identifying a quality of the first virtual image by calculating a similarity between the first virtual image and the destination image and a similarity between the source image and a first reconstruction image obtained by inverting the first virtual image into the first domain, wherein the quality of the first virtual image is associated with a level at which the erroneous features are removed in the first virtual image.