Underwater Image Restoration Based on Marine Snow Removal Techniques

Image capture and analysis in underwater environments have applications in many fields, such as underwater archaeological exploration, deep-sea biological research, oil and gas resource exploration, and path navigation of intelligent underwater robots. Existing underwater image restoration methods usually face problems such as high model complexity and the inability to be deployed on actual underwater operating platforms. This paper proposes an efficient and fast underwater image restoration method called UIRF-Net to address these issues. This method can restore underwater images and remove marine snow phenomena. UIRF-Net makes the deep learning network more lightweight by using modules such as channel extraction prior information and RGB three-channel color enhancement, improving the image color while ensuring the image quality and processing speed. Through experiments on the UIEB-Withsnow dataset, it is verified that UIRF-Net can produce clear underwater images from photos containing complex marine snow noise. This study provides a comprehensive enhancement solution for the field of underwater image restoration, which significantly improves the clarity of underwater photos.

I. Introduction

Marine Snow Noise refers to the white bright spots caused by the scattering and reflection of light by organic and inorganic particles (such as plankton, debris, mineral particles, etc.) suspended in the water in underwater images or videos. These particles are suspended in the ocean, forming a visual effect like snowflakes falling in the air, so it is called "marine snow". Marine snow is mainly composed of biological debris^[1], excrement, microbial communities, etc., and includes mud and mineral particles carried by underwater robots and underwater vehicles during operation. These particles are injected into the ocean through rivers or enter the ocean through seabed activities. Bubbles generated during underwater activities or natural processes will also increase the amount of marine snow. Marine snow noise significantly impacts the quality of underwater images by causing blurred vision and reduced contrast, which in turn hampers the effectiveness of underwater vision systems. To address this problem, researchers have proposed many methods in history, including traditional algorithms, deep learning techniques, and other image enhancement technologies. These studies have made notable progress in eliminating snow noise, significantly improving the quality and processing of underwater images.