An adaptive image super-resolution restoration method fusing multi-level complementary features

By integrating multi-level complementary features into an adaptive image super-resolution restoration method, and utilizing a dynamic parameter generation network to adaptively adjust the image super-resolution mapping model, the problem of insufficient single feature description in existing technologies is solved, achieving a more accurate and vivid image super-resolution restoration effect.

CN115293983BActive Publication Date: 2026-05-29TAIYUAN UNIVERSITY OF TECHNOLOGY
View PDF 1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
TAIYUAN UNIVERSITY OF TECHNOLOGY
Filing Date
2022-08-05
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing deep learning-based image super-resolution restoration methods rely on only a single feature of the image when adaptively adjusting the image super-resolution mapping model. This results in an inability to effectively reconstruct high-frequency information of the image and makes it difficult to accurately describe the differences in mapping relationships between different types of images.

Method used

An adaptive image super-resolution restoration method that integrates multi-level complementary features is adopted. The method extracts and integrates multi-level complementary features of the image through a dynamic parameter generation network to generate dynamic parameters, adaptively adjusts the image super-resolution mapping model, and optimizes the network using a loss function until convergence.

Benefits of technology

It achieves more accurate and vivid image super-resolution restoration results, improves image super-resolution restoration performance, and can more effectively distinguish the super-resolution mapping relationship of different types of images.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115293983B_ABST
    Figure CN115293983B_ABST
Patent Text Reader

Abstract

The application relates to the field of image super-resolution restoration, and discloses an adaptive image super-resolution restoration method fusing multi-level complementary features, which comprises the following steps: modeling an image super-resolution mapping model by using a dynamic parameter neural network; acquiring image color, gradient, texture and semantic multi-level features by using a feature extraction network; proposing a complementary feature selection and fusion method by using a sorting distance and uncertainty measurement; adaptively generating image super-resolution network dynamic parameters by fusing the multi-level complementary features; and adaptively adjusting the image super-resolution mapping model in a feature mapping mode by the dynamic parameters. By introducing the multi-level complementary features in the image super-resolution restoration, the image super-resolution mapping model can be adaptively adjusted, and the image texture information can be more accurately and vividly restored.
Need to check novelty before this filing date? Find Prior Art