A Hybrid Image Denoising Framework using Adaptive Clustering and Residual Analysis

Image denoising is an important preprocessing step in medical imaging, reducing noise while preserving essential structural information. But existing denoising methods exhibit inconsistent performance across different imaging modalities and noise levels.

Experiments are performed on HRCT and MRI data corrupted with Gaussian noise at various noise levels (low to high). A hybrid approach has been proposed by using the method of noise concept. This study also presents a comparative evaluation of various image denoising techniques using both reference and non-reference metrics.

Experimental results demonstrate that the proposed method exhibits consistent, symmetric degradation in PSNR and SSIM as the noise level increases. For dataset1, the proposed method achieves a PSNR of 31.62 and 24.98 at noise levels of 10 and 40, respectively. Similarly, for dataset2, the PSNR is 36.54 at noise level 10 and 28.68 at 40.

Although several denoising methods have been proposed, most perform well only in low-noise conditions, with performance improving as the noise level decreases. Based on both reference-based (PSNR, SSIM, and entropy) and no-reference perceptual quality features (NIQE, BRISQUE, and PIQE), the research provides an in-depth comparative study of various filters, bilateral filter variants, and advanced hybrid denoising methods. One of the primary strengths of the proposed method is its superior performance at high noise levels.

These results indicate the effectiveness and reliability of the proposed method for denoising high-quality images, particularly in challenging high-noise scenarios. The quantitative and perceptual quality results show that it is an effective preprocessing tool for medical imaging.