Nonlocal means image denoising using orthogonal moments.

Area of Science:

• Digital Image Processing
• Computational Imaging
• Medical Image Analysis

Background:

• Image noise significantly degrades visual quality and diagnostic accuracy in medical imaging.
• Traditional denoising methods often struggle to preserve image details while effectively removing various noise types.
• The nonlocal means (NLM) algorithm offers a robust framework for image denoising by leveraging patch similarity.

Purpose of the Study:

• To propose and evaluate a novel image denoising method in the moment domain.
• To enhance the performance of the modified nonlocal means (NLM) algorithm using Krawtchouk moments for noise reduction.
• To assess the effectiveness of the proposed method on synthetic and real clinical images corrupted by different noise models.

Main Methods:

• Development of a modified nonlocal means (NLM) algorithm incorporating Krawtchouk moments for neighborhood similarity evaluation.
• Quantitative validation of denoising performance using Peak Signal-to-Noise Ratio (PSNR), Structural Similarity (SSIM) index, and Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE).
• Experimental evaluation on synthetic and real clinical images contaminated with Gaussian, Poisson, and Rician noise.

Main Results:

• The proposed Krawtchouk moment-based NLM algorithm demonstrated superior performance compared to Zernike-based denoising.
• Significant improvements were observed in PSNR (3.1 dB), SSIM (0.1285), and BRISQUE (4.23) scores.
• The method showed competitive results across varying noise levels and types, outperforming existing techniques.

Conclusions:

• The proposed moment domain image denoising method effectively reduces noise while preserving image quality.
• Krawtchouk moments provide an efficient way to evaluate neighborhood similarity in the NLM framework for enhanced denoising.
• The method shows promise for clinical applications requiring high-fidelity image reconstruction.