Low-light image enhancement method based on illumination-reflection decomposition and multi-discriminator

By employing illumination-reflection decomposition and a multi-discriminator approach, the problem of insufficient iterative enhancement capability in low-light image enhancement is solved, achieving coordinated restoration of brightness, color, and detail, thereby improving the realism and stability of the image.

CN122115224APending Publication Date: 2026-05-29TIANJIN UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
TIANJIN UNIV
Filing Date
2026-02-02
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing low-light image enhancement methods suffer from insufficient iterative enhancement capabilities and a lack of effective supervision in terms of brightness restoration and structure preservation, leading to problems such as noise amplification, loss of detail, and artifacts.

Method used

We employ a method based on illumination-reflection decomposition and multiple discriminators, using a temperature-gated cross-attention module, a dual-stream dynamic enhancement module, and a DCT-enhanced hierarchical fusion module for feature decomposition and enhancement. We also introduce a multi-view adversarial discrimination network and a progressive light and color degradation device to construct a multi-dimensional supervision mechanism.

Benefits of technology

It achieves coordinated restoration of image brightness, color, and detail, improves the realism and stability of the enhancement results, effectively suppresses noise amplification and artifacts, and conforms to the laws of human visual perception.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122115224A_ABST
    Figure CN122115224A_ABST
Patent Text Reader

Abstract

The application discloses a low-illumination image enhancement method based on illumination-reflection decomposition and multiple discriminators, and belongs to the technical field of image processing and computer vision; the method comprises the following steps: extracting low-illumination image features, and dividing luminance branches and reflection branches; realizing two-branch feature information interaction through a temperature-gated cross-attention module; introducing a double-flow dynamic enhancement module into the luminance branch to separate illumination degradation and real content information, and introducing a DCT enhanced hierarchical fusion module into the reflection branch to extract multi-scale high-frequency texture features; then, an iterative mechanism is adopted to update the enhancement result in multiple rounds, and the image quality is gradually optimized; finally, a multi-view adversarial discrimination network is constructed, the enhancement image is discriminated from luminance, contrast and the like by relying on a three-attribute discrimination module, and a gradual light color degenerator is combined to implement reverse degradation supervision. The method is cooperatively constrained by forward enhancement and reverse degradation, can effectively inhibit noise amplification and color distortion while improving the luminance of the image, and improves the authenticity and naturalness of the enhancement result.
Need to check novelty before this filing date? Find Prior Art