Image Denoising Using a Compressive Sensing Approach Based on Regularization Constraints.

Area of Science:

• Image processing
• Signal processing
• Remote sensing
• Medical imaging

Background:

• Large data volumes in remote sensing and medical imaging necessitate efficient compression.
• Compressed sensing (CS) enables signal acquisition with fewer measurements but often struggles with reconstruction detail.
• Existing CS reconstruction methods can smooth image textures and details.

Purpose of the Study:

• To develop an improved compressed sensing reconstruction method.
• To enhance image denoising and preserve fine details, textures, and edges.
• To address limitations of current reconstruction techniques in maintaining image fidelity.

Main Methods:

• Proposed a novel CS reconstruction algorithm combining total variation regularization and non-local self-similarity constraints.
• Utilized an augmented Lagrangian for optimization, avoiding non-linearity and non-differentiability issues.
• The algorithm, termed denoising-compressed sensing by regularization (DCSR), performs both reconstruction and denoising.

Main Results:

• The DCSR algorithm demonstrated superior performance in denoising and preserving image details compared to state-of-the-art methods.
• Achieved up to a 25% gain in denoising efficiency and visual quality, measured by PSNR and SSIM.
• Evaluated against Nesterov's algorithm, group-based sparse representation, and wavelet-based methods.

Conclusions:

• The proposed DCSR method effectively reconstructs compressed images while simultaneously denoising them.
• The combination of total variation and non-local self-similarity significantly improves texture and edge preservation.
• DCSR offers a robust solution for high-fidelity image reconstruction in data-intensive applications.