Adaptive fusion based deep learning framework for restoring underwater image quality using multi scale attention...

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• Underwater images suffer from degradation like blurring, low contrast, and color deviation.
• Restoring underwater images is crucial for various practical applications.
• Traditional methods struggle with complex underwater image degradations.

Purpose of the Study:

• To develop an effective method for restoring underwater images.
• To improve visibility and overall quality of underwater images.
• To introduce the Efficient Restoration of Underwater Images Using Multi-Scale Attention Features (ERUI-MSAF) model.

Main Methods:

• Adaptive Bilateral Filtering (ABF) for noise reduction and pre-processing.
• ERUI-MSAF model integrating channel and spatial attention features.
• Fusion of Deep Wavenet (DWN) for spatial features and EfficientNet for channel features.

Main Results:

• The ERUI-MSAF model adaptively emphasizes informative features and regions.
• Achieved superior Peak Signal-to-Noise Ratio (PSNR) values of 34.258 and 29.0073 on EUVP and UIEB datasets.
• Demonstrated high performance and computational efficiency compared to existing models.

Conclusions:

• The proposed ERUI-MSAF model is effective for underwater image restoration.
• The integration of multi-scale attention features significantly improves image quality.
• The method offers a promising solution for enhancing underwater imagery.