Title:
WEIGHT-SHARING DOUBLE-REGION GENERATIVE ADVERSARIAL NETWORK AND IMAGE GENERATION METHOD THEREFOR
Document Type and Number:
WIPO Patent Application WO/2022/094911
Kind Code:
A1
Abstract:
Disclosed in the present invention are a weight-sharing double-region generative adversarial network and an image generation method therefor. The generative adversarial network comprises a non-artifact region processing module and an artifact region processing module, wherein the non-artifact region processing module comprises a first feature generator, a first feature parser, and a first discriminator, and the artifact region processing module comprises a second feature generator, an artifact generator, a reconstruction parser, a second feature parser, and a second discriminator. In the present invention, the same generator and parser are repeatedly used for training for multiple times, such that the sharing performance of image features of the generator and the parser can be generated; the two regions are used for training a model, such that both supervised training and unsupervised training can be performed. In the present invention, motion artifacts of a medical image can be eliminated, and the peak signal-to-noise ratio and the structural similarity of the image are improved while the image features are generated, thereby obtaining a clearer medical image meeting diagnosis requirements.
Inventors:
HU ZHANLI (CN)
ZHENG HAIRONG (CN)
LIANG DONG (CN)
LIU XIN (CN)
DENG FUQUAN (CN)
ZHENG HAIRONG (CN)
LIANG DONG (CN)
LIU XIN (CN)
DENG FUQUAN (CN)
Application Number:
PCT/CN2020/127030
Publication Date:
May 12, 2022
Filing Date:
November 06, 2020
Export Citation:
Assignee:
SHENZHEN INST ADV TECH (CN)
International Classes:
G06T5/00; G06N3/04; G06N3/08; G06T11/00
Foreign References:
| CN110675461A | 2020-01-10 | |||
| CN110570492A | 2019-12-13 | |||
| US20190128989A1 | 2019-05-02 | |||
| US20190377047A1 | 2019-12-12 | |||
| US20190197358A1 | 2019-06-27 |
Attorney, Agent or Firm:
BEIJING CHENGHUI LAW FIRM (CN)
Download PDF: