水下
计算机科学
人工智能
计算机视觉
能见度
衰减
图像复原
亮度
失真(音乐)
水准点(测量)
频道(广播)
颜色校正
干扰(通信)
图像(数学)
图像处理
光学
物理
地质学
电信
海洋学
放大器
带宽(计算)
大地测量学
作者
Jingchun Zhou,Y. Wang,Chongyi Li,Weishi Zhang
出处
期刊:IEEE Journal of Oceanic Engineering
[Institute of Electrical and Electronics Engineers]
日期:2023-10-01
卷期号:48 (4): 1322-1337
被引量:12
标识
DOI:10.1109/joe.2023.3275615
摘要
The unique physical properties of underwater environments often lead to color distortion, reduced visibility, and loss of detail in underwater images. To address these challenges, we present a novel underwater image restoration technique, known as the multicolor components and light attenuation (MCLA) method. Initially, we introduce an adaptive background light estimation for underwater image restoration (BLEU) approach, utilizing multicolor model conversion to effectively eliminate color casts while minimizing interference from white objects and suspended particles. Following this, we propose a cutting-edge depth-map estimation model grounded in feature priors, which serves to enhance details and restore textures. We then adjust the red channel with a shorter wavelength to correct the depth map and compute the transmission map based on the Lambert–Beer law. Ultimately, by leveraging the obtained background light and transmission map, we can generate clear underwater images using the inverted underwater optical imaging model. The MCLA method not only improves brightness and visibility, but also removes color casts, boosts contrast, enriches details, and reduces artifacts. Experimental results on the widely used UIEB benchmark demonstrate the superiority of MCLA over state-of-the-art techniques in both subjective and objective evaluations. Moreover, our approach exhibits enhanced performance in color correction accuracy and detail and texture restoration.
科研通智能强力驱动
Strongly Powered by AbleSci AI