Underwater Image Restoration Based on Light Attenuation Prior and Scene Depth Fusion Model

Xu Zhu,Yujie Li,Huimin Lu

DOI: https://doi.org/10.1007/978-3-031-47634-1_4

2023-01-01

Abstract:Underwater images often suffer from blurry details, color distortion, and low contrast due to light absorption and scattering in water. Existing restoration technologies use a fixed attenuation coefficient value, which fails to account for the uncertainty of the water body and leads to suboptimal restoration results. To address these issues, we propose a scene depth fusion model that considers underwater light attenuation to obtain a more accurate attenuation coefficient for image restoration. Our method employs the quadtree decom-position method and a depth map to estimate the background light. We then fuse and refine the depth map, compute the attenuation coefficient of the water medium for a more precise transmission map, and apply a reversed underwater imaging model to restore the image. Experiments demonstrate that our method effectively enhances the details and colors of underwater images while improving the contrast. Moreover, our method outperforms several state-of-the-art methods in terms of both accuracy and quality, showing its superior performance.