Single image super-resolution method and system based on deep coupled enhancement network

By using a deep coupling enhancement network approach, combined with Haar wavelet decomposition and global feature extraction from pre-trained EDSR, adaptive fusion of high-frequency shallow features and global features is achieved. This solves the problem of poor image detail reconstruction in complex structures in existing technologies, and improves image reconstruction accuracy and cross-scene adaptability.

CN122415338APending Publication Date: 2026-07-17QINGHAI UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
QINGHAI UNIVERSITY
Filing Date
2026-05-15
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing single-image super-resolution techniques are poor at reconstructing details in images with complex structures, failing to balance computational efficiency with detail reconstruction accuracy.

Method used

A method based on deep coupling enhancement network is adopted, which extracts high-frequency shallow feature maps through Haar wavelet decomposition and deep convolution, and fuses them with the global feature map of pre-trained EDSR. The residual deep coupling module is used to realize the adaptive fusion of high-frequency shallow features and global features. Finally, the HR image is reconstructed through subpixel convolution upsampling.

Benefits of technology

It significantly improves image reconstruction accuracy and resource utilization, adapts to the semantics and high-frequency features of different images, and enhances cross-scene adaptability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122415338A_ABST
    Figure CN122415338A_ABST
Patent Text Reader

Abstract

The application provides a single image super-resolution method and system based on a deep coupling enhanced network, and relates to the technical field of image processing. An auxiliary global feature extraction module is introduced, which can capture the global context information of the image and avoid the loss of texture and edge details. Through a deep coupling module, adaptive fusion of high-frequency shallow features and global features is realized, avoiding the problems of limited receptive field and high computational complexity. The system performs two fusions on high-frequency shallow feature maps and global feature maps, i.e., first generating semantic feature maps and then generating high-resolution image feature maps, ensuring the full use and synergistic effect of local details and global semantic information. This deep coupling feature fusion mechanism enables the system to better adapt to the semantics and high-frequency features of different images, thereby showing significant progress in reconstruction accuracy, resource utilization and cross-scene adaptability.
Need to check novelty before this filing date? Find Prior Art