DCD-UIE: Decoupled Chromatic Diffusion Model for Underwater Image Enhancement.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Underwater images suffer from color distortion and structural degradation, hindering visual perception and analysis.
• Traditional underwater image enhancement (UIE) methods struggle to simultaneously address both color and structure issues due to their coupling in RGB space.

Purpose of the Study:

• To propose a novel guided diffusion model for UIE that effectively decouples and enhances both color and structure.
• To develop a framework that collaboratively optimizes color and structure under various underwater degradation scenarios.

Main Methods:

• A guided diffusion model utilizing the HSV color space to naturally separate color (H, S) and structure (V).
• An adaptive perceptual guidance module (APGM) generating orthogonal color and structure guides for the diffusion model.
• A decoupled loss optimization module with independent loss functions for supervising color and structure restoration.
• A closed-loop optimization framework combining forward decoupled guidance and backward decoupled supervision.

Main Results:

• The proposed method significantly outperforms existing state-of-the-art UIE approaches across diverse underwater scenes.
• Demonstrated superior performance in enhancing images affected by color casts and haze.
• Achieved excellent results on no-reference image quality assessment metrics, indicating high perceptual quality.

Conclusions:

• The decoupled guided diffusion model effectively addresses the coupled challenges of color distortion and structural degradation in UIE.
• The HSV color space and decoupled optimization framework provide a robust solution for restoring high-quality underwater images.
• The method shows promise for various applications requiring clear and visually accurate underwater imagery.