A Low-Light Image Enhancement Method Based on Frequency Domain Transformer

By using the frequency domain Transformer method, image features are mapped to the frequency domain for self-attention operations. Combined with a frequency domain guided feedforward mechanism, the challenges of brightness enhancement and detail restoration in low-light image enhancement are solved, achieving efficient and high-fidelity image enhancement results.

CN122694728BActive Publication Date: 2026-11-10NAVAL AVIATION UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202511657559.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-11-13
Publication Date
2026-11-10
Estimated Expiration
2045-11-13

AI Technical Summary

Technical Problem

Existing low-light image enhancement methods face significant challenges in terms of brightness enhancement, structure preservation, and computational efficiency. Traditional methods lack adaptive modeling capabilities, CNN structures are limited by local receptive fields, and while Transformers have global modeling capabilities, they ignore frequency domain information, resulting in insufficient enhancement results in high-frequency detail and texture restoration.

Method used

We employ a frequency-domain Transformer-based approach, which maps features to the frequency domain for self-attention operations and combines this with a frequency-domain guided feedforward mechanism to achieve efficient modeling of the global context and frequency domain information of the image. We utilize frequency-domain attention to improve image brightness and detail recovery while reducing computational complexity.

Benefits of technology

It significantly improves the brightness and contrast of low-light images while restoring image details and texture features, avoiding the problems of inconsistent enhancement results and high computational cost, and maintaining high fidelity and computational efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122694728B_ABST
    Figure CN122694728B_ABST
Patent Text Reader

Abstract

This application relates to the field of computer vision and image processing technology, specifically providing a low-light image enhancement method based on frequency domain Transformer. The method includes: firstly, shallow feature extraction is performed on the original low-light image, followed by input into an encoder-decoder network. This network employs a frequency domain Transformer module, which uses Fast Fourier Transform to completely map the query, key, and value to the frequency domain. A self-attention mechanism is then executed in the frequency domain to calculate attention weights and perform weighted modulation on the value vector. Finally, an inverse transform maps the image back to the spatial domain. This invention efficiently captures frequency domain spatial information and models global contextual information to effectively enhance image features, improve high-quality image restoration, and achieve high-performance low-light image enhancement.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Moving image deblurring model based on Fourier transform

    CN117115040A

  • Remote sensing image super-resolution reconstruction method based on multi-stage space-frequency combination

    CN118052712A