Reweighted Low-Rank Matrix Analysis With Structural Smoothness for Image Denoising

Area of Science:

• Computer Vision
• Image Processing
• Matrix Analysis

Background:

• Image denoising is crucial for visual data quality.
• Existing low-rank methods struggle with significant sparse noise.
• Incorporating structural properties can improve denoising.

Purpose of the Study:

• Develop a robust low-rank matrix recovery algorithm for image denoising.
• Enhance structural smoothness and pixel range constraints in image recovery.
• Improve performance against sparse noise and in hyper-spectral imaging.

Main Methods:

• Proposed a mathematical formulation combining nuclear norm, total variation (TV) norm, and norm.
• Utilized iterative alternating direction and fast gradient projection methods.
• Developed an algorithm to solve the non-convex optimization problem.

Main Results:

• The proposed method significantly improves image quality, especially with large random noise.
• Achieved up to 4.21 dB improvement in single-image denoising with 30% sparse noise density.
• Outperformed state-of-the-art low-rank matrix recovery techniques in evaluations.

Conclusions:

• The novel algorithm effectively exploits image low-rank properties and structural smoothness.
• The method demonstrates superior performance in single-image denoising, hyper-spectral denoising, and video background modeling.
• This approach offers a significant advancement in handling noisy image data.