P3C-DNet: Pseudo-Groundtruth Contrastive Learning With Color Calibration Dehazing Network.

Area of Science:

• Computer Vision
• Image Processing

Background:

• Supervised dehazing methods struggle with real-world images due to reliance on synthetic data.
• Existing methods face challenges in generalization, color distortion, and handling complex haze scenarios.

Purpose of the Study:

• To develop a novel unsupervised framework for accurate and robust real-world image dehazing.
• To overcome limitations of existing methods in generalization and color fidelity.

Main Methods:

• Proposed P3C-DNet (Pseudo-groundtruth Contrastive learning with Color Calibration Dehazing Network) framework.
• Introduced Pseudo-groundtruth Contrastive Supervision (PCS) for generating training data.
• Integrated a codebook-based matching mechanism and a Dynamic Color Restoration Block (DCRB).

Main Results:

• P3C-DNet achieved superior performance in haze removal compared to existing methods.
• Demonstrated significant improvements in color fidelity and detail preservation.
• Set a new benchmark for real-world image dehazing tasks.

Conclusions:

• The proposed unsupervised framework effectively addresses limitations of supervised methods for real-world dehazing.
• P3C-DNet offers a robust solution for complex haze scenarios with improved visual quality.
• The method shows strong potential for practical applications in image restoration.