Underwater Image Enhancement using Convolutional Block Attention Module

Abstract

Underwater images include poor contrast, fuzzy features, and colour distortion due to light scattering, refraction and absorption by unwanted dust particles in water. This research demonstrates that assigning the appropriate receptive field size context depending on the traversal scope of the color channel can result in a significant performance boost for the objective of underwater image enhancement. It's critical to reduce non-uniform multiple contextual elements, and also boost the model's representational potential. So to dynamically modify the learnt multi-contextual characteristics, we included an attentive skip method. The suggested framework is improved via pixel wise and feature based cost functions. Experiments and comparisons with existing deep learning models and conventional approaches validate the framework for underwater image enhancement. The proposed framework is superior according to comparison results.