WEDM: Wavelet-Enhanced Diffusion with Multi-Stage Frequency Learning for Underwater Image Enhancement.

Area of Science:

• Computer Vision
• Image Processing
• Marine Science

Background:

• Underwater images suffer degradation from light absorption and scattering, causing color distortion and loss of detail.
• Existing spatial-domain methods often overlook crucial frequency-domain information for texture and edge restoration.

Purpose of the Study:

• To propose a novel underwater image enhancement (UIE) framework, the Wavelet-based Enhancement Diffusion Model (WEDM).
• To leverage frequency-domain analysis and diffusion models for improved UIE.
• To enhance visual quality for marine exploration and ecological monitoring applications.

Main Methods:

• The proposed WEDM framework integrates frequency-domain decomposition with diffusion models.
• It features a Wavelet Color Compensation Module (WCCM) for LAB color correction using discrete wavelet transform.
• A Wavelet Diffusion Module (WDM) employs wavelet-based operations, replacing convolutions to preserve multi-scale frequency features.

Main Results:

• The WEDM framework effectively reduces noise amplification and high-frequency blurring by combining residual denoising diffusion with frequency-specific processing.
• Ablation studies confirmed the critical contributions of both WCCM and WDM to enhanced color fidelity and texture details.
• The framework provides a robust solution for underwater visual tasks.

Conclusions:

• The Wavelet-based Enhancement Diffusion Model (WEDM) offers a significant advancement in underwater image enhancement.
• Its integration of wavelet transforms and diffusion models effectively addresses degradation issues.
• WEDM shows strong potential for applications in marine exploration and ecological monitoring.