Image Denoising via Bandwise Adaptive Modeling and Regularization Exploiting Nonlocal Similarity

Area of Science:

• Computer Vision
• Signal Processing
• Image Restoration

Background:

• Traditional image denoising methods often struggle with non-stationary image signals.
• Global modeling approaches fail to capture the diverse characteristics across different image regions and transform bands.
• Existing algorithms may not effectively leverage local image correlations for accurate parameter estimation.

Purpose of the Study:

• To develop a novel image denoising algorithm that improves image quality through adaptive signal modeling and regularization.
• To address the limitations of global models by employing content-dependent adaptive models for image patches.
• To enhance denoising performance by accurately estimating distribution parameters for each band within individual patches.

Main Methods:

• Proposes an adaptive signal modeling and regularization technique for image denoising.
• Utilizes bandwise distribution modeling in the transform domain for image patch regularization.
• Employs non-local correlation to gather similar patches for adaptively estimating distribution parameters (expectation and variance) for each band.
• Restores the image using bandwise adaptive soft-thresholding based on a Laplacian approximation.

Main Results:

• The proposed algorithm demonstrates superior performance compared to several state-of-the-art denoising methods.
• Objective quality metrics show significant improvements in image restoration.
• Perceptual quality assessments confirm the effectiveness of the denoising scheme.

Conclusions:

• Adaptive signal modeling and bandwise regularization offer a robust approach to image denoising.
• The content-dependent adaptive models significantly enhance denoising accuracy over global methods.
• The algorithm effectively handles image non-stationarity and transform band diversity, leading to high-quality image restoration.