Underwater Image Enhancement Framework with Deep Background Light Estimation Module

Underwater image enhancement is a challenging task due to the absorption and scattering of light, resulting in color distortion and reduced contrast. In this study, we develop a novel variant of the recently proposed HUWIE-Net (Hybrid Underwater Image Enhancement Network), a hybrid deep learning-based framework which performed both pixel-level color correction and physics-informed dehazing with outstanding results for underwater image enhancement. The novel variant proposed here, named HUWIE-BL-Net, incorporates a new deep learning-based background light estimation module that adaptively models spatial variations in the illumination. The inclusion of this new module results in improved enhancement performance. Experimental results on real-world underwater images demonstrate considerably better results for the proposed HUWIE-BL-Net in terms of both quantitative metrics and perceptual consistency across different underwater scenarios.