Learning mapping by curve iteration estimation For real-time underwater image enhancement

Area of Science:

• Computer Vision
• Image Processing

Background:

• Underwater image quality is degraded by light attenuation and scattering.
• Existing enhancement algorithms often lack real-time performance and are computationally intensive.

Purpose of the Study:

• To develop a lightweight and efficient framework for underwater image enhancement.
• To overcome the limitations of current algorithms in practical underwater vision tasks.

Main Methods:

• A novel iterative curve estimation approach is proposed to learn image mappings.
• A parameter estimation network, CieNet, is utilized to learn curve parameters.
• A set of loss functions guides the parameter learning process.

Main Results:

• The proposed method achieves superior performance over existing algorithms in quantitative metrics and visual quality.
• The lightweight network design significantly reduces computational resource requirements.
• The method demonstrates extremely short running times, facilitating real-time applications.

Conclusions:

• The developed framework offers an effective and efficient solution for underwater image enhancement.
• Its lightweight nature and real-time capabilities make it highly applicable for integration into various devices.
• This research advances the field of underwater computer vision by providing a practical and performant enhancement technique.