Deep Underwater Image Quality Assessment With Explicit Degradation Awareness Embedding.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Underwater Image Quality Assessment (UIQA) is a critical research area.
• Current deep learning models for UIQA face challenges due to limited supervision from single quality scores.
• Diverse image content and distortions can lead to similar quality scores, hindering model training.

Purpose of the Study:

• To develop a novel deep UIQA model with Explicit Degradation Awareness embedding (EDANet).
• To enhance UIQA model learning by incorporating detailed degradation-aware information.
• To improve the accuracy and robustness of underwater image quality evaluation.

Main Methods:

• A two-stage training strategy is employed for EDANet.
• A Degradation Information Discovery subnetwork (DIDNet) is pre-trained to generate a residual map characterizing local image degradation.
• Intermediate features from DIDNet are embedded into a Degradation-guided Quality Evaluation subnetwork (DQENet) for quality prediction.

Main Results:

• EDANet demonstrates superior performance compared to 18 state-of-the-art UIQA methods.
• Extensive comparisons on two benchmark datasets validate the effectiveness of EDANet.
• The model successfully leverages degradation-aware information for enhanced quality prediction.

Conclusions:

• EDANet offers a significant advancement in underwater image quality assessment.
• Explicit degradation awareness effectively guides deep learning models for UIQA.
• The proposed method provides a more robust and accurate approach to evaluating underwater image quality.