PCAFA-Net: A Physically Guided Network for Underwater Image Enhancement with Frequency-Spatial Attention.

Area of Science:

• Computer Vision
• Image Processing
• Remote Sensing

Background:

• Underwater images suffer degradation (color shifts, blur, low contrast) due to light scattering and environmental factors.
• Existing physical models and deep learning methods struggle with diverse conditions and data limitations.
• Current approaches often neglect spectral differences and frequency domain information for image enhancement.

Purpose of the Study:

• To develop a novel deep learning network for robust underwater image enhancement.
• To address limitations of conventional methods by incorporating physical guidance and multi-domain information.
• To improve the quality of underwater images for better visual analysis and applications.

Main Methods:

• Introduced PCAFA-Net, a physically guided network for underwater image enhancement.
• Developed three key modules: Adaptive Gradient Simulation Module (AGSM), Adaptive Color Range Adjustment Module (ACRAM), and Frequency-Spatial Strip Attention Module (FSSAM).
• Utilized adaptive adjustments across RGB, Lab, and HIS color spaces and integrated frequency-spatial attention.

Main Results:

• PCAFA-Net demonstrated superior performance in enhancing underwater images compared to existing methods.
• Experimental results on three datasets showed significant improvements in both subjective and objective evaluations.
• The network effectively addressed color shifts, blur, and contrast reduction in degraded images.

Conclusions:

• PCAFA-Net offers an effective solution for underwater image enhancement by leveraging physical guidance and multi-domain feature extraction.
• The proposed method shows promise for applications requiring high-quality underwater imagery.
• Future work could explore further optimization and real-time implementation of the network.