CSUID - Comprehensive synthetic underwater image dataset.

Area of Science:

• Computer Vision
• Image Processing
• Oceanography

Background:

• Underwater environments present significant challenges for image acquisition due to light scattering, absorption, and distortion.
• Common underwater image degradations include color cast, reduced contrast, blurring, and light attenuation.
• Developing robust algorithms for underwater image enhancement requires realistic and diverse training data.

Purpose of the Study:

• To introduce a comprehensive dataset of synthetic underwater images for research.
• To simulate various underwater optical conditions and their combined effects.
• To provide a benchmark for evaluating underwater image restoration algorithms.

Main Methods:

• Generation of 150,000 synthetic underwater images from 100 ground-truth images.
• Application of diverse combinations of degradation effects (color cast, blurring, low-light, contrast reduction) at varying severity levels.
• Calculation of 21 distinct focus metrics for each image to quantify image quality.

Main Results:

• A large-scale dataset approximating real-world underwater image conditions was successfully created.
• The dataset captures a wide range of underwater visual challenges, offering significant variability.
• Quantitative analysis using 21 focus metrics provides valuable data for algorithm development.

Conclusions:

• The proposed synthetic dataset effectively models underwater image degradation.
• This resource is expected to advance the development of image restoration techniques for underwater applications.
• The comprehensive focus metrics enable rigorous evaluation of image enhancement algorithms.