US 12,189,720 B2
Image analysis device and method, and method for generating image analysis model used for same
Taewan Kim, Seoul (KR); Gunhee Kim, Seoul (KR); and Soochan Lee, Seoul (KR)
Assigned to SK Telecom Co., Ltd., Seoul (KR)
Filed by SK TELECOM CO., LTD., Seoul (KR); and Seoul National University R&DB Foundation, Seoul (KR)
Filed on Sep. 3, 2021, as Appl. No. 17/466,400.
Application 17/466,400 is a continuation of application No. PCT/KR2020/000291, filed on Jan. 7, 2020.
Claims priority of application No. 10-2019-0026852 (KR), filed on Mar. 8, 2019.
Prior Publication US 2021/0397902 A1, Dec. 23, 2021
Int. Cl. G06F 18/21 (2023.01); G06F 18/23211 (2023.01); G06F 18/2431 (2023.01); G06N 5/04 (2023.01)
CPC G06F 18/2185 (2023.01) [G06F 18/23211 (2023.01); G06F 18/2431 (2023.01); G06N 5/04 (2013.01)] 14 Claims
OG exemplary drawing
 
1. An image analysis method, comprising:
receiving a target image for processing by an image analysis model composed of at least one expert generated by learning a training image;
analyzing the target image based on an output value from the at least one expert to generate an analysis result;
determining validity of the analysis result for the target image based on a probability of an input target image belonging to an input category of the at least one expert by at least:
obtaining a likelihood of an input and a class with respect to each of the at least one expert;
obtaining a prior probability of each of the at least one expert; and
obtaining a probability of the input target image belonging to the input category of the at least one expert based on the obtained likelihood and the obtained prior probability; and
responsive to determining that the analysis result for the target image is not valid, classifying the target image into a new category so that a new expert for the new category is trained using the target image.