Single image de-raining by multi-scale Fourier Transform network

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Single-image de-raining is challenging due to streak variability.
• Data-driven methods often require large datasets and lack interpretability.

Purpose of the Study:

• To propose a novel, interpretable, and efficient single-image de-raining method.
• To leverage Fourier Transform prior knowledge for improved rain streak removal.

Main Methods:

• Decomposing rainy images into amplitude and phase components using Fourier Transform.
• Applying data-driven algorithms to process amplitude and phase components separately.
• Reconstructing enhanced features and refining with a multi-scale neural network with attention mechanisms.

Main Results:

• The proposed method significantly outperforms state-of-the-art de-raining approaches.
• Achieved superior qualitative and quantitative evaluations in experimental results.
• Demonstrated improved image clarity and robustness in de-raining.

Conclusions:

• The novel approach effectively combines Fourier Transform and data-driven techniques.
• Offers a more interpretable and efficient solution for single-image de-raining.
• Represents a significant advancement in image restoration research.