Deep Learning-Driven Parameter Adaptation for Underwater Image Restoration

Abstract

In this paper we propose a learning-based approach to enhance underwater image quality by optimizing parameters and applying intensity transformations. Our methodology involves training a CNN Regression model on diverse underwater images to learn enhancing parameters, followed by applying intensity transformation techniques. In order to evaluate our approach, we conducted experiments using well-known underwater image datasets found in the literature, comprising real-world subaquatic images and we propose a novel underwater image dataset, composed by 276 images from Amazon turbid water rivers. The results demonstrate that our approach achieves an impressive accuracy rate in three different underwater image datasets. This high level of accuracy showcases the robustness and efficiency of our proposed method in restoring underwater images.