RHVI-FDD: A Hierarchical Decoupling Framework for Low-Light Image Enhancement

2026-04-07

AI Summary

Low-light images often suffer from severe noise, detail loss, and color distortion, further complicated by the coupling between luminance and chrominance, as well as the entanglement of noise and fine details within chrominance components. To address these challenges, this work proposes RHVI-FDD, a hierarchical decoupling framework. At the macro level, it employs the RHVI transform to achieve robust luminance–chrominance decoupling. At the micro level, a frequency-domain decoupling module leverages the discrete cosine transform to decompose chrominance features into low-, mid-, and high-frequency bands—corresponding respectively to global color tone, local details, and noise—which are then processed by dedicated expert networks and adaptively fused via a gating mechanism. This approach pioneers hierarchical decoupling in low-light enhancement and achieves state-of-the-art performance across multiple datasets, delivering superior quantitative metrics and visual quality.

Abstract

Low-light images often suffer from severe noise, detail loss, and color distortion, which hinder downstream multimedia analysis and retrieval tasks. The degradation in low-light images is complex: luminance and chrominance are coupled, while within the chrominance, noise and details are deeply entangled, preventing existing methods from simultaneously correcting color distortion, suppressing noise, and preserving fine details. To tackle the above challenges, we propose a novel hierarchical decoupling framework (RHVI-FDD). At the macro level, we introduce the RHVI transform, which mitigates the estimation bias caused by input noise and enables robust luminance-chrominance decoupling. At the micro level, we design a Frequency-Domain Decoupling (FDD) module with three branches for further feature separation. Using the Discrete Cosine Transform, we decompose chrominance features into low, mid, and high-frequency bands that predominantly represent global tone, local details, and noise components, which are then processed by tailored expert networks in a divide-and-conquer manner and fused via an adaptive gating module for content-aware fusion. Extensive experiments on multiple low-light datasets demonstrate that our method consistently outperforms existing state-of-the-art approaches in both objective metrics and subjective visual quality.

Problem

• low-light image enhancement
• color distortion
• noise suppression
• detail preservation
• luminance-chrominance coupling

Innovation

• Hierarchical Decoupling
• RHVI Transform
• Frequency-Domain Decoupling
• Low-Light Image Enhancement
• Discrete Cosine Transform