Patent · US Active

Channel attention-based swin-transformer image denoising method and system

US12367543B2 · kind B2 · utility

0Cited by
0References
9Claims
0Family size

Assignee

Inventors

Key dates

Filing dateDec 30, 2021
Grant dateJul 22, 2025
Priority date
Expiry dateDec 30, 2041

Classification

  • Technology area (CPC G)Physics
  • CPC primaryG06T2207/20224
  • WIPO fieldComputer technology
  • WIPO sectorElectrical engineering

Abstract

The invention provides a channel attention (CA)-based Swin-Transformer image denoising method and system. A noisy image is inputted into a trained and optimized denoising network model. A shallow layer feature extraction network in the denoising network model first extracts shallow layer feature information such as noise and channels in the noisy image. The extracted shallow layer feature information is then inputted into a deep layer feature extraction network in the denoising network model to acquire deep layer feature information. Subsequently, the shallow layer feature information and the deep layer feature information are inputted into a reconstruction network of the denoising network model to perform feature fusion, so that a clear image can be obtained, thereby overcoming problems that an image denoising method based on a deep convolutional neural network is prone to a loss of details in an inputted noisy image and high computing memory and time consumption.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.