Underwater image enhancement using hybrid transformers and evolutionary particle swarm optimization.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Underwater imaging faces challenges like poor visibility, color distortion, and light scattering.
• Existing methods struggle to effectively address these complex underwater image degradation issues.

Purpose of the Study:

• To develop and evaluate a novel framework for enhancing underwater image quality.
• To improve the clarity, color accuracy, and overall visual fidelity of underwater imagery.

Main Methods:

• A Hybrid Transformer Network (HTN) combined with Particle Swarm Optimization (PSO) was developed (HTN-PSO).
• The framework integrates CNNs and transformers for feature extraction and dependency modeling.
• PSO optimizes transformer parameters for maximum image enhancement.

Main Results:

• HTN-PSO demonstrated superior performance on benchmark datasets (RUIE, EUVP, UWGAN, UIEB).
• Achieved a 12% increase in Underwater Image Quality Measure (UIQM) and up to 15% reduction in Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE).
• Outperformed established methods like Uformer and Restormer.

Conclusions:

• The HTN-PSO framework offers a significant advancement in underwater image enhancement.
• It provides a robust solution for improving image quality in exploration, research, and surveillance.
• The proposed method shows superiority over traditional and existing neural network-based approaches.