A Detailed Preserving Medical Image Denoising Using Cluster-Wise PCA Thresholding and Iterative Mean Filtering

Abstract

Image denoising is a vital process in medical imaging that involves removing noise or distortions introduced during image acquisition. Random noise can degrade image quality and reduce contrast, making accurate analysis more challenging. This paper presents a detail-preserving medical image denoising framework based on a hybrid methodology that combines cluster-wise Principal Component Analysis (PCA) thresholding with Iterative Mean Filtering (IMF). The process begins with noise level estimation followed by the extraction of overlapping image patches. These patches are then grouped using K-means++ clustering to form structurally homogeneous clusters, enabling localized processing. Within each cluster, PCA is applied to transform the data into a low-dimensional subspace, and an adaptive threshold is computed using the Marchenko Pastur (MP) law to separate noise from significant image features. The denoised patches are then reconstructed by applying inverse PCA. To further enhance structural continuity and reduce residual noise, Iterative Mean Filtering is employed both before and after PCA denoising. The final denoised image is reconstructed through weighted averaging of the processed patches. The modular structure of the algorithm allows dynamic thresholding, cluster-specific denoising, and iterative refinement, making the approach adaptive to varying noise levels and anatomical structures. This methodology is evaluated against state-of-the-art denoising methods using both full-reference metrics (PSNR, SSIM, FSIM, Entropy) and no-reference quality measures (BRISQUE, NIQE, PIQE). Results demonstrate that the proposed method achieves superior noise suppression while preserving fine details, offering an efficient and interpretable solution for clinical image Enhancement.