LCH‐Net: A Lightweight OKLCH‐Space Decoupling Network for Archival Image Enhancement

ABSTRACT Dimness and blurriness are common issues in archival images of libraries and archives. Traditional enhancement methods in the sRGB colour space also tend to cause colour distortion and noise amplification. For these reasons, this study puts forward the LCH‐Net framework. This method processes lightness, chroma and hue separately in the OKLCH colour space: the lightness adjustment sub‐network adaptively regulates illumination via a learnable lightness curve; the chroma adjustment sub‐network combines frequency‐domain and spatial‐domain denoising to suppress noise while preserving details; and the hue adjustment sub‐network employs circular convolution to maintain colour continuity. Experiments show that the model requires only 1.89 M parameters and 5.72G FLOPs, yet achieves outstanding performance on datasets such as LOL‑v2‑real (PSNR 23.96, SSIM 0.89, LPIPS 0.143) and LOL‑v2‑synthetic (PSNR 27.23, SSIM 0.96, LPIPS 0.047). It thus provides an efficient and practical solution for enhancing historical images in resource‐constrained scenarios.