Image denoising by exploring external and internal correlations.

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Single image denoising is challenging due to limited data within a single noisy image.
• Existing methods often struggle with accurate patch selection and noise reduction.

Purpose of the Study:

• To propose a novel image denoising scheme utilizing both internal and external image correlations.
• To enhance denoising performance by effectively utilizing web images.

Main Methods:

• Constructing internal and external data cubes from similar patches in noisy and web images.
• Employing a two-stage denoising strategy with graph-based optimization for external denoising and frequency truncation for internal denoising.
• Utilizing transform domain filtering and fusing results from internal and external denoising stages.

Main Results:

• The proposed method achieves superior performance compared to state-of-the-art denoising schemes.
• Experimental results demonstrate significant improvements in both subjective and objective quality measurements.
• The method achieves over 2 dB gain compared to BM3D across various noise levels.

Conclusions:

• The novel image denoising scheme effectively leverages internal and external correlations for superior noise reduction.
• The two-stage filtering approach, enhanced by preliminary denoising results, improves patch matching and parameter estimation.
• The method offers a robust and high-performing solution for single image denoising problems.