Twice Mixing: A rank learning based quality assessment approach for underwater image enhancement
作者：
Highlights：
• We present a rank learning framework for UIE-IQA based on an elaborately designed self-supervision mechanism. It is also the first time that using deep learning approaches to address the UIE-IQA problem.
• We construct a dataset with over 2200 raw underwater images and their high-quality and low-quality enhanced versions.
• Extensive experiments on both synthetic and real-world UIE-IQA databases demonstrate that our method outperforms other methods significantly, and is more suitable for real applications.
摘要
•We present a rank learning framework for UIE-IQA based on an elaborately designed self-supervision mechanism. It is also the first time that using deep learning approaches to address the UIE-IQA problem.•We construct a dataset with over 2200 raw underwater images and their high-quality and low-quality enhanced versions.•Extensive experiments on both synthetic and real-world UIE-IQA databases demonstrate that our method outperforms other methods significantly, and is more suitable for real applications.
论文关键词：Underwater image,Quality assessment,Mixing,Rank learning,Siamese Network
论文评审过程：Received 13 March 2021, Revised 19 November 2021, Accepted 21 December 2021, Available online 28 December 2021, Version of Record 12 January 2022.
论文官网地址：https://doi.org/10.1016/j.image.2021.116622
