PRIDNet based Image Denoising for Underwater Images

Abstract

Underwater image enhancement has become a popular research topic due to its importance in aquatic robotics and marine engineering. However, the underwater images frequently experience signal-dependent speckle noise when transmitting and acquiring data, which can limit certain applications such as detection, object tracking. In the recent years, the existing underwater image enhancement algorithms efficiency has been analysed and evaluated on a small number of carefully chosen real-world images or synthetic datasets. As such, it is challenging to predict how these algorithms might function with images acquired in the wild under various circumstances. This paper introduces a new solution for noise removal from underwater images called Pyramid Real Image Noise Removal Network (PRIDNet) with patches.PRIDNet is a three-level network design using image patches. The tests were carried out on a dataset of actual noisy images demonstrate that, in terms of quantitative metrics, our proposed denoising model reduction performs better with the exixting denoisers. We determine the effectiveness and constraints of existing algorithms using benchmark assessments and the suggested model, offering valuable information for further studies on underwater image enhancement.