A Deep Feature Fusion Underwater Image Enhancement Model Based on Perceptual Vision Swin Transformer

Area of Science:

• Computer Vision
• Image Processing
• Marine Technology

Background:

• Underwater optical images are vital for marine exploration and monitoring.
• Image degradation from scattering and absorption reduces contrast and detail.
• Existing methods struggle with severe underwater image distortions.

Purpose of the Study:

• To develop an advanced underwater image enhancement framework.
• To address challenges of reduced contrast, chromatic distortions, and loss of details.
• To improve the quality of underwater optical images for various applications.

Main Methods:

• Proposed a U-shaped framework integrating Swin-Transformer blocks with attention and residual modules.
• Introduced Dual-Window Multi-Head Self-Attention (DWMSA) for context and structure.
• Utilized Global-Aware Attention Map (GAMP) and Feature-Augmentation Residual Network (FARN) for adaptive enhancement.
• Trained the model using Charbonnier, perceptual, and edge losses.

Main Results:

• Achieved state-of-the-art performance on UFO-120 and EUVP datasets.
• Reported high average metrics: PSNR 29.5 dB, SSIM 0.94, LPIPS 0.17, UIQM 3.62, UCIQE 0.59.
• Demonstrated significant improvements in contrast, color restoration, and detail sharpness.

Conclusions:

• The proposed framework effectively enhances degraded underwater images.
• The method outperforms existing techniques in quantitative and qualitative evaluations.
• The results support the framework's utility in marine resource exploration and monitoring.