IMAGE SUPER-RESOLUTION RECONSTRUCTION METHOD EMPLOYING ADAPTIVE ADJUSTMENT

WIPO Patent Application WO/2021/185225

A1

An image super-resolution reconstruction method employing adaptive adjustment, pertaining to the technical field of image processing. A basic framework comprises an adversarial training model involving a generative adversarial network. The training model consists of a generative model and a discriminant model competing with each other. The generative model is responsible for generating a high-resolution image. The discriminant model is used to determine whether an input image is generated or is a sample acquired from a high-resolution database. With the recognition ability gradually improved, the discriminant model transmits information to the generative model, and by optimizing a loss function, the high-resolution image generated by the generative model is closer to a real sample. As the quality of the generated image improves, the loss of the discriminant model increases. In addition, since the recognition ability of the discriminant model is continuously improved, when the discriminant model cannot distinguish between the generated image and the real sample, the generative model completes a super-resolution task. The invention improves the ability to express a model feature, and achieves a good super-resolution reconstruction effect.

JIANG DAIHONG (CN)
ZHANG SANYOU (CN)
DAI LEI (CN)

PCT/CN2021/080920

September 23, 2021

March 16, 2021

Click for automatic bibliography generation  

UNIV XUZHOU TECHNOLOGY (CN)

G06N3/04; G06T3/40

CN111507898A	2020-08-07
CN110136063A	2019-08-16
CN109978762A	2019-07-05
CN107154023A	2017-09-12
CN109615582A	2019-04-12
CN109993698A	2019-07-09
US20180075581A1	2018-03-15

XUZHOUSHI SANLIAN PATENT AGENCY (CN)

View/Download PDF      PDF Help

Previous Patent: SPEAKER AND ELECTRONIC DEVICE

Next Patent: ROCKING CHAIR FOR CHILDREN