External Prior Guided Internal Prior Learning for Real-World Noisy Image Denoising

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Existing image denoising methods rely on external data or the noisy image itself for prior learning.
• External priors may lack adaptivity, while internal priors can be compromised by noise interference.
• Real-world noise complexity challenges traditional denoising models based on simple distributions.

Purpose of the Study:

• To develop an adaptive image denoising method for complex real-world noise.
• To propose an external prior guided internal prior learning strategy.
• To enhance the accuracy and robustness of image denoising.

Main Methods:

• Learning external priors from a dataset of clean natural images.
• Learning internal priors from the noisy image, guided by the external priors.
• Formulating priors as orthogonal dictionaries for efficient image reconstruction.

Main Results:

• The proposed method achieved highly competitive denoising performance.
• Outperformed state-of-the-art denoising methods, including those specialized for real-world noise.
• Demonstrated effectiveness on various real-world noisy image datasets.

Conclusions:

• The external prior guided internal prior learning method is effective for real-world noisy image denoising.
• This approach offers a robust solution for complex noise removal.
• The method shows significant improvements over existing techniques.