An image super-resolution system based on reference images

CN114359044BActive Publication Date: 2024-12-13SOUTH CHINA UNIV OF TECH
2 Cites 0 Cited by

Patent Information

Application Number
CN202111488253.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-12-07
Publication Date
2024-12-13
Estimated Expiration
2041-12-07

AI Technical Summary

Technical Problem

The existing image super-resolution method based on reference images is not highly correlated with the input image, and the model performance is degraded, ignoring the high-frequency information in the low-resolution image space, resulting in poor texture detail restoration effect. .

Method used

A system including feature extraction module, texture migration module and image reconstruction module was designed. Through shallow and deep feature extraction, texture migration and reconstruction techniques, combined with attention mechanism and residual connection, it effectively utilizes the high frequency in reference images. Information to make up for the missing texture details in low-resolution images.

Benefits of technology

Given a reference image with correlation, the system can better restore the texture details of low-resolution images and improve the super-resolution reconstruction effect. It is suitable for real-world dual-camera datasets and video super-resolution tasks, with better market and application value.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114359044B_ABST
    Figure CN114359044B_ABST
Patent Text Reader

Abstract

The present invention discloses an image super-resolution system based on a reference image, including a feature extraction module, a texture migration module and an image reconstruction module; the feature extraction module is used to extract features from an input low-resolution image to generate a feature map corresponding to the low-resolution image, the texture migration module migrates high-definition texture details from a reference high-resolution image and fuses them into the feature map extracted from the low-resolution image, and the image reconstruction module reconstructs a high-resolution image of the size of a target image from the fused feature map. The present invention can achieve performance superior to that of a single image super-resolution method, and can better restore the lost texture details of a low-resolution image when a reference image with a certain correlation is given. The present invention can be extended to real-world dual-camera datasets and other tasks, such as video super-resolution and video denoising, and has potential market and application value.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Image super-resolution reconstruction method based on fused attention mechanism residual network

    CN111192200A

  • Image super-resolution method based on cross-scale texture migration

    CN113112401A