SGUIE-Net

SGUIE-Net enhances underwater images by using semantic attention and multi-scale perception to correct color distortion and restore blurred details caused by wavelength-dependent light attenuation, refraction, and scattering for improved analysis of underwater imagery.

Key Features:

• Semantic Information Integration: Leverages semantic information as high-level guidance for region-wise enhancement feature learning to target specific areas within an image.
• Semantic Region-Wise Enhancement Module: Implements a module that learns local enhancement features for distinct semantic regions using multi-scale perception.
• Multi-Scale Perception: Processes images at multiple scales to capture both global and local features for comprehensive enhancement.
• Feature Fusion: Integrates complementary local features with global enhancement features extracted at the original image scale to produce semantically consistent outputs.

Scientific Applications:

• Marine Biology: Enhances image clarity and color accuracy to support analysis of aquatic ecosystems and biodiversity.
• Environmental Monitoring: Provides improved underwater imagery to support monitoring and analysis of environmental conditions.
• Oceanographic Research: Improves visual data quality for oceanographic studies that rely on underwater imaging.

Methodology:

Employs a structured learning approach to learn from limited paired training samples by establishing effective learning mechanisms that extract richer supervised information; the network architecture balances local and global feature extraction, and performance was evaluated via experiments on publicly available datasets and a newly proposed dataset.