An image restoration method based on dual cue guided transformer

By employing a dual-cue guided Transformer-based image restoration method, which utilizes an enhanced dynamic decoupler and high- and low-frequency cue modulators to process image features, the shortcomings of existing methods in restoring high-frequency and low-frequency features are addressed, achieving higher-precision image restoration results.

CN118396858BActive Publication Date: 2026-07-03NANKAI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
NANKAI UNIV
Filing Date
2024-03-14
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing Transformer-based image restoration methods are ineffective at uncovering high-frequency local details and low-frequency nonlocal structures, making it difficult to simultaneously simulate both.

Method used

An image restoration method based on dual-cue guided Transformer is adopted. The image features are decoupled into low-frequency and high-frequency features through an enhanced dynamic decoupler module, and these features are processed by high-frequency and low-frequency cue modulators respectively. The image is reconstructed by combining a multi-head self-attention mechanism and an encoder-decoder network.

Benefits of technology

It improves the accuracy of image restoration, enabling better recovery of high-frequency local details and low-frequency non-local structures, resulting in clearer images.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118396858B_ABST
    Figure CN118396858B_ABST
Patent Text Reader

Abstract

The present application relates to the technical field of image processing, and more particularly to an image restoration method based on a dual prompt guided Transformer, which extracts first and second shallow features of an image through a convolution layer; obtains low-frequency features and high-frequency features by decoupling the first shallow features through an enhanced dynamic decoupler module; obtains high-frequency prompt features by processing the high-frequency features through a high-frequency prompt modulator; obtains low-frequency prompt features by processing the low-frequency features through a low-frequency prompt modulator; improves the second shallow features into deep features through a decoder and an encoder network, modifies the low-frequency prompt features and the high-frequency prompt features according to the deep features to obtain low-frequency output features and high-frequency output features; and inputs the low-frequency output features and the high-frequency output features into an image reconstruction module to generate a reconstructed image. The features in different frequency ranges under multi-scale resolution are used to prompt the model to restore clearer images, thereby improving the accuracy of image restoration and achieving better restoration effect.
Need to check novelty before this filing date? Find Prior Art