能见度
水下
计算机视觉
人工智能
计算机科学
图像复原
遥感
光学
地质学
图像处理
物理
图像(数学)
海洋学
作者
Frans Shafuda,Hayato Kondo
标识
DOI:10.1109/joe.2025.3562630
摘要
In a turbid water environment, captured images suffer from degradation due to poor visibility and low contrast caused by the scattering and absorption of light. This can be further exacerbated by foreground occlusion due to suspended macro particles. Artificial light sources are often used to improve visibility, but using single point light sources, such as spotlights results in captured images with regions of strong backscatter and regions of low signal, negatively affecting conventional underwater image restoration methods. To address these challenges, we introduce a novel scene illumination and scan approach that uses two parallel light emitting diode light strips and a single camera at the center to ensure uniform scene illumination with a defined distribution pattern. We scan and capture scene images in a synthetic aperture imaging fashion, obtaining multiple images with overlapping views and high spatial coherence. Based on the introduced scene illumination and scan approach, we propose a turbidity removal method that uses multiview image information and a defined illumination pattern to estimate backscatter, as well as an occlusion compensation method that takes advantage of spatial coherence among images to remove foreground occlusion. Experimental results demonstrate the effectiveness of our approach in restoring visibility of turbid and occluded scenes.
科研通智能强力驱动
Strongly Powered by AbleSci AI