ReverSea: An Enhancement Technique for Low-Lit Underwater Images

Enhancing low-lit underwater images is a challenging task due to color distortion, low visibility and poor contrast. These issues arise due to the way light behaves in water, leading to a loss of detail and making image enhancement more complex. In this paper, ReverSea, a lightweight Convolutional Neural Network (CNN) is proposed to enhance underwater images by incorporating depth-aware correction using transmission maps. Unlike traditional methods that rely on preprocessing for low-lit images, ReverSea processes them directly, eliminating the need for preprocessing and ensuring adaptability to low-light conditions. The model restores natural colors, improves visibility and reduces distortions through a loss function that combines color constancy, histogram equalization and luminous loss. ReverSea performance is evaluated on the Sea-Thru dataset, where it outperforms state-of-the-art methods in both visual quality and Underwater Image Quality Metric (UIQM). Further, ReverSea effectively handles non-uniform lighting conditions, particularly in deep-sea environments, as demonstrated by its performance on the OceanDark dataset.