研究目的
To present a new, large-scale database for evaluating underwater image processing methods, specifically for enhancement and restoration, using real turbidity water to simulate realistic underwater environments.
研究成果
The NWPU underwater image database provides a large-scale, realistic dataset for evaluating underwater image processing methods. It includes ground-truth images and supports quantitative assessment using multiple quality metrics. The use of real turbidity water enhances authenticity, and the database can facilitate future research in underwater image enhancement and restoration.
研究不足
The use of a turbidimeter may introduce measurement errors in turbidity levels. The database is limited to a controlled tank environment and may not fully represent all real-world underwater scenarios. Only one enhancement algorithm (MSRCR) is evaluated in the baseline.
1：Experimental Design and Method Selection:
The experiment involves designing an underwater image capturing system to create a database with controlled turbidity, lighting, and distance conditions. The MSRCR algorithm is used for image enhancement, and four image quality assessment methods (PSNR, SSIM, NIQE, UIQM) are employed for evaluation.
2：Sample Selection and Data Sources:
40 objects are used, each captured under 6 turbidity levels (0,5,10,15,20,25 NTU), 4 lighting conditions, and 6 distances (30cm to 55cm in 5cm intervals). Ground-truth images are captured in clear water and air.
3：List of Experimental Equipment and Materials:
A custom water tank (150cm x 50cm x 50cm), a camera (RS-A2300-GM50/GC50 by Microview), LED lights, a turbidimeter for measuring water turbidity, and a computer for data transfer and image capture.
4：Experimental Procedures and Operational Workflow:
The camera and objects are fixed in position. High turbidity water is prepared using real lake water mixed with clear water. Images are captured sequentially by diluting turbidity, changing lighting, and adjusting distance. Physical alignment ensures pixel-level consistency.
5：Data Analysis Methods:
The MSRCR algorithm is applied to enhance images. Quality is assessed using PSNR, SSIM (full-reference), NIQE, and UIQM (no-reference) metrics to evaluate performance.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
