DMFourLLIE: Dual-Stage and Multi-Branch Fourier Network for Low-Light Image Enhancement

TL;DR

DMFourLLIE tackles color distortion and noise in low-light image enhancement by rethinking Fourier-domain processing: instead of indiscriminately amplifying the amplitude, it preserves structure through phase refinement guided by infrared priors and luminance-aware amplitude modulation. The method splits into a first Fourier reconstruction stage that fuses cross-modal priors and a second spatial-texture reconstruction stage that combines multi-scale spatial convolutions with Fourier convolutions. A four-term loss balances fidelity, perception, and luminance-guided guidance to optimize both stages jointly. Across LOL-v2-Real, LOL-v2-Syn, and LSRW-Huawei, DMFourLLIE achieves state-of-the-art results with robust color fidelity, detail preservation, and improved downstream tasks such as object detection in low-light scenes. The work also demonstrates the value of infrared information and luminance priors in guiding Fourier-domain enhancement, paving the way for more integrated cross-modal approaches in LLIE.

Abstract

In the Fourier frequency domain, luminance information is primarily encoded in the amplitude component, while spatial structure information is significantly contained within the phase component. Existing low-light image enhancement techniques using Fourier transform have mainly focused on amplifying the amplitude component and simply replicating the phase component, an approach that often leads to color distortions and noise issues. In this paper, we propose a Dual-Stage Multi-Branch Fourier Low-Light Image Enhancement (DMFourLLIE) framework to address these limitations by emphasizing the phase component's role in preserving image structure and detail. The first stage integrates structural information from infrared images to enhance the phase component and employs a luminance-attention mechanism in the luminance-chrominance color space to precisely control amplitude enhancement. The second stage combines multi-scale and Fourier convolutional branches for robust image reconstruction, effectively recovering spatial structures and textures. This dual-branch joint optimization process ensures that complex image information is retained, overcoming the limitations of previous methods that neglected the interplay between amplitude and phase. Extensive experiments across multiple datasets demonstrate that DMFourLLIE outperforms current state-of-the-art methods in low-light image enhancement. Our code is available at https://github.com/bywlzts/DMFourLLIE.

Figures (10)

• Figure 1: Comparison with State-of-the-Art Two-Stage Fourier-Based Methods. (a) displays initial results from FECNet four2 and FourLLIE four1, which exhibit color distortion and noise. Our method, in contrast, enhances brightness and preserves color fidelity, eliminating noise. (b) shows the superior noise suppression and color accuracy of our approach, particularly in highlighted areas.
• Figure 2: Our Motivations. (a): Illustrates issues from replacing the amplitude component of a low-light image with that of a normally illuminated image, resulting in $I_{a}$. (b): Highlights the importance of the phase component, as shown by the distorted image $I_{b}$ from replacing the phase component.
• Figure 3: Overall framework of DMFourLLIE.
• Figure 4: Structure of the Fourier Reconstruction Stage.
• Figure 5: Structure of the Spatial and Texture Reconstruction Stage.