Underwater Image Quality Assessment: Subjective and Objective Methods

Pengfei Guo,Lang He,Shuangyin Liu,Delu Zeng,Hantao Liu

DOI: https://doi.org/10.1109/tmm.2021.3074825

IF: 7.3

2022-01-01

IEEE Transactions on Multimedia

Abstract:Underwater image enhancement plays a critical role in marine industry. Various algorithms are applied to enhance underwater images, but their performance in terms of perceptual quality has been little studied. In this paper, we investigate five popular enhancement algorithms and their output image quality. To this end, we have created a benchmark, including images enhanced by different algorithms and ground truth image quality obtained by human perception experiments. We statistically analyse the impact of various enhancement algorithms on the perceived quality of underwater images. Also, the visual quality provided by these algorithms is evaluated objectively, aiming to inform the development of objective metrics for automatic assessment of the quality for underwater image enhancement. The image quality benchmark and its objective metric are made publicly available.

computer science, information systems,telecommunications, software engineering

What problem does this paper attempt to address?

The main objective of this paper is to evaluate and compare the performance of current popular underwater image enhancement algorithms and to propose a new objective quality measurement method. Specifically:

1. **Algorithm Comparative Study**: The paper selects five advanced underwater image enhancement algorithms (including both physics-based and non-physics-based methods), describes, compares, and analyzes these algorithms, and explores their effectiveness in processing underwater images.

2. **Establishing a Benchmark Dataset**: To evaluate the quality of output images from different algorithms, the researchers constructed a benchmark database containing 40 original images and their enhanced versions by different algorithms. These images cover complex underwater environment characteristics, such as color shift, sharpness degradation, and contrast reduction.

3. **Subjective and Objective Quality Evaluation**: Well-controlled perceptual experiments were conducted to obtain human subjective ratings of the enhanced image quality. Additionally, a novel objective quality metric was proposed to automatically assess the quality of underwater image enhancement results. This metric relies solely on the enhanced images themselves, without the need for reference to "perfect" or "highest" quality standard images.

In summary, this paper aims to advance the related field by systematically comparing various underwater image enhancement techniques and combining subjective and objective evaluation methods, particularly providing new insights and technical means for automatic image quality assessment.