Enhancing underwater image clarity: a multi-branch network with commonality-specificity fusion

Abstract

Underwater images often suffer from degradation due to light attenuation and scattering, posing challenges for vision-based tasks. This paper introduces a novel Edge-Guided Multi-Branch Commonality-Specificity Network (EGMCS-Net) to enhance underwater image quality. The network integrates an RGB branch for initial enhancement and an edge branch for refined edge features, followed by a Commonality-Specificity Fusion Module (CSFM) to achieve complementary feature interaction. To the best of our knowledge, this is the first application of commonality-specificity modeling to multi-branch underwater image enhancement. Extensive experiments on UIEB, LSUI, and C60 datasets demonstrate superior performance, with the best SSIM and LPIPS on UIEB, best PSNR and SSIM on LSUI, and strong generalization on C60.