US 11,989,264 B2
Automatic generation system of training image and method thereof
Tien-He Chen, Taoyuan (TW); Che-Min Chen, Taoyuan (TW); and Jia-Wei Yan, Taoyuan (TW)
Assigned to DELTA ELECTRONICS, INC., Taoyuan (TW)
Filed by DELTA ELECTRONICS, INC., Taoyuan (TW)
Filed on Jun. 2, 2023, as Appl. No. 18/205,490.
Application 18/205,490 is a continuation of application No. 17/472,951, filed on Sep. 13, 2021, granted, now 11,709,913.
Claims priority of provisional application 63/091,857, filed on Oct. 14, 2020.
Claims priority of application No. 202011389588.6 (CN), filed on Dec. 1, 2020.
Prior Publication US 2023/0306080 A1, Sep. 28, 2023
This patent is subject to a terminal disclaimer.
Int. Cl. G06F 18/214 (2023.01); G06F 18/21 (2023.01); G06F 18/2415 (2023.01); G06V 10/56 (2022.01)
CPC G06F 18/2148 (2023.01) [G06F 18/2193 (2023.01); G06F 18/2415 (2023.01); G06V 10/56 (2022.01)] 14 Claims
OG exemplary drawing
 
1. An automatic generation method of a training image, comprising:
a) acquiring a plurality of container images and selecting one of the container images to execute a target-adding process to generate a training image,
wherein the target-adding process includes:
a1) acquiring a target image;
a2) adding the target image to the selected container image as a candidate image;
a3) computing a reliability of the candidate image, wherein the reliability corresponds to a recognition difficulty of the target image in the candidate image, and the recognition difficulty positively correlates to a similarity between the target image and a surrounding image of target image in the candidate image; and
a4) repeatedly performing the step a1) to the step a3) until the reliability of the candidate image meets a threshold condition for generating the training image; and
b) recording the training image and a target data, wherein the target data includes at least one of a target category and a target position of the target image;
wherein computing the reliability of the candidate image includes executing a second object-recognizing process, and the second object-recognizing process includes:
a311) obtaining a first probability of each target category corresponding to an image block of the target image of the candidate image, determining the target category with a highest first probability, and computing a category recognition score of the target image in the candidate image based on the highest first probability; and
a312) obtaining a second probability of each range fully covering the target image, determining the target position of the target image based on the range with a highest second probability, and computing a completeness recognition score based on the highest second probability; and
a313) computing the reliability of the candidate image based on the category recognition score and the completeness recognition score of the target image.