Dual-path joint correction network for underwater image enhancement.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Underwater images suffer from significant quality degradation, including color distortion and loss of detail.
• These issues are primarily caused by light absorption and scattering from suspended particles in water.
• Existing methods struggle to effectively address both color shift and blur simultaneously.

Purpose of the Study:

• To propose a novel network, the Dual-path Joint Correction Network (DJC-NET), for enhancing degraded underwater images.
• To simultaneously address color distortion and detail loss in underwater imagery.
• To preserve unique properties of underwater images through a dual-branch approach.

Main Methods:

• Developed a Dual-path Joint Correction Network (DJC-NET) with two distinct branches: one for light absorption correction and another for light scattering correction.
• The light absorption branch employs a triplet color feature extraction module to balance R, G, and B channel distributions.
• The light scattering branch utilizes a dual-dimensional attention mechanism for enhanced texture and detail recovery, with features fused using a multi-scale U-net.

Main Results:

• DJC-NET effectively corrects color shifts and recovers lost details in underwater images.
• The dual-branch architecture successfully addresses both light absorption and scattering issues.
• Experimental results show superior performance compared to state-of-the-art methods across diverse underwater scenes.

Conclusions:

• The proposed DJC-NET offers a robust solution for underwater image enhancement.
• The network's dual-path design effectively tackles the complex degradation factors in underwater environments.
• DJC-NET demonstrates significant improvements in visual quality and objective metrics for underwater imagery.