Abstract
Low-light image enhancement remains a challenging task, particularly inpreserving object edge continuity and fine structural details under extremeillumination degradation. In this paper, we propose a novel model, DRWKV(Detailed Receptance Weighted Key Value), which integrates our proposed GlobalEdge Retinex (GER) theory, enabling effective decoupling of illumination andedge structures for enhanced edge fidelity. Secondly, we introduce Evolving WKVAttention, a spiral-scanning mechanism that captures spatial edge continuityand models irregular structures more effectively. Thirdly, we design theBilateral Spectrum Aligner (Bi-SAB) and a tailored MS2-Loss to jointly alignluminance and chrominance features, improving visual naturalness and mitigatingartifacts. Extensive experiments on five LLIE benchmarks demonstrate that DRWKVachieves leading performance in PSNR, SSIM, and NIQE while maintaining lowcomputational complexity. Furthermore, DRWKV enhances downstream performance inlow-light multi-object tracking tasks, validating its generalizationcapabilities.