Swin-Transform image denoising method and system based on channel attention

An attention and image technology, applied in the field of image processing, can solve the problems of easy loss of input noise image details, high computational memory and time consumption, and achieve the effect of overcoming high computational memory and time consumption

Active Publication Date: 2022-03-04
SUZHOU UNIV
View PDF5 Cites 15 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] For this reason, the technical problem to be solved by the present invention is to overcome the problem that the image denoising method based on de

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Swin-Transform image denoising method and system based on channel attention
  • Swin-Transform image denoising method and system based on channel attention
  • Swin-Transform image denoising method and system based on channel attention

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0041] The present embodiment provides a Swin-Transformer image denoising method based on channel attention, comprising the following steps:

[0042] Step S1: Obtain the original high-definition image data set, preprocess the original high-definition image data set, and obtain multiple pairs of training data sets of noisy images and high-definition images for the denoising network model.

[0043] Specifically, the present invention is trained on 800 training images of the image denoising training data set DIV2K, and tested on the public benchmark data sets BSD68 and Set12 of image denoising. DIV2K is a high-quality (2K resolution) image dataset consisting of 800 training images, 100 validation images and 100 test images. Set12 has 12 different scene noise images, and BSD68 has 68 different natural scene noise images.

[0044] Add Gaussian noise to 800 high-definition images in DIV2K, and generate 800 pairs of noise / clear images as the original training set D. Convert all ima...

Embodiment 2

[0071] Based on the same inventive concept, this embodiment provides a Swin-Transformer image denoising system based on channel attention, the problem-solving principle of which is similar to the xx method described above, and repeated descriptions will not be repeated.

[0072] This embodiment provides a Swin-Transformer image denoising system based on channel attention, including:

[0073] The data preparation module is used to obtain the original high-definition image data set, preprocess the original high-definition image data set, and obtain a training data set of noisy images and high-definition images for denoising network model training;

[0074] The shallow feature extraction module is used to input the noisy image to the shallow feature extraction network in the denoising network model to extract feature information to obtain a shallow feature map;

[0075] The deep feature extraction module is used to use the shallow feature map as the input of the deep feature extr...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention relates to a Swindow-Transform image denoising method and a Swindow-Transform image denoising system based on channel attention. According to the method, a noise image is input into a trained and optimized denoising network model, and a shallow feature extraction network in the denoising network model firstly extracts shallow feature information such as noise and channels of the noise image; inputting the extracted shallow feature information into a deep feature extraction network in a denoising network model to obtain deep feature information, and inputting the shallow feature information and the deep feature information into a reconstruction network of the denoising network model for feature fusion to obtain a pure image. The problems that in the prior art, an image denoising method based on a deep convolutional neural network is prone to losing input noise image details, and high computation memory and time consumption are caused are solved.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a Swin-Transformer image denoising method and system based on channel attention. Background technique [0002] Image denoising is an important low-level computer vision task that has great promise in photography, medical imaging, biology, and many other fields. The purpose of image denoising is to restore a noisy image to a clean, noise-free image. In recent years, due to the great success of deep learning in computer vision, convolutional neural network (CNN) has been applied to image denoising tasks and achieved impressive performance. Currently, most state-of-the-art image denoising methods are based on CNN and achieve satisfactory results. For example, Residual non-local attention networks (RIDNet) were proposed to solve the denoising problem of real images. RIDNet is a single-stage denoising network with feature attention. However, RIDNet lacks adaptabil...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
IPC IPC(8): G06T5/00G06T3/40G06V10/40G06V10/774G06V10/80G06V10/82G06K9/62G06N3/04G06N3/08
CPCG06T5/002G06T3/4007G06N3/08G06T2207/20081G06T2207/20084G06N3/045G06F18/214G06F18/253
Inventor 张莉代强赵雷王邦军
Owner SUZHOU UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap