Progressive Luminance Residual Diffusion with Manifold-Based Representation Refinement for Low-Light Image Enhancement

Low-light image enhancement aims to recover visually natural and detail-preserving images from scenes captured under insufficient illumination. However, existing methods often improve brightness at the cost of noise amplification, structural degradation, and locally inconsistent textures, making it difficult to simultaneously preserve dark-region details, edge fidelity, and color stability. To address these issues, we propose a Luminance Residual Diffusion Network (LumiRD), a unified framework that combines degradation-aware representation refinement with progressive residual restoration. Specifically, an Illumination Manifold Unfolding Module (IMUM) is designed to jointly exploit spatial, gradient, and frequency-domain cues, enabling suppressed responses in low-light features to be adaptively unfolded and enhancing the structural discriminability of dark regions. Building on this, a Residual Diffusion Module (RDM) performs luminance-residual-guided restoration by integrating global exposure correction with local detail compensation, thereby alleviating luminance accumulation, texture drift, and over-enhancement introduced by direct enhancement strategies. Through the collaboration of IMUM and RDM, the proposed model first regularizes low-light feature distributions and then progressively restores unrecovered components, leading to improved structural consistency and robustness under complex illumination variations. Experimental results on multiple public low-light datasets demonstrate that LumiRD achieves superior performance in both quantitative evaluation and visual quality, providing a favorable balance among brightness enhancement, detail recovery, and noise suppression.