Ldf-Net: Dual-Fusion Network Underwater Image Enhancement Based on Lab Color Space

Underwater images often suffer from color distortion, low contrast, and blur due to wavelength-dependent light absorption and scattering in water. To address these challenges, we propose LDF-Net (Lab Dual-Fusion Network), an enhancement model in the CIELAB color space that explicitly separates luminance (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{L}$</tex>) and chrominance (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{a b}$</tex>) information into two parallel branches. LDF-Net combines multi-scale shallow feature extraction, DBlock+CBAM-based mid-level enhancement, and MFM-based cross-level fusion to jointly optimize contrast restoration and color correction. LDF-Net allows the luminance branch to focus on dehazing and detail recovery while the chrominance branch corrects color casts and boosts saturation; adaptive weighting in the MFM modules ensures effective collaboration between these branches. Extensive experiments on public underwater datasets demonstrate that LDF-Net outperforms mainstream methods in PSNR, SSIM, UIQM, and UCIQE, while exhibiting strong generalization under varying water conditions. The code is available at https://github.com/losper5/LDF-Net.