Filtragem não local de ruídos em imagens de tomografia computadorizada

Abstract

Computed tomography (CT) is widely used in medical diagnostics due to its ability to generate high-precision three-dimensional images. However, the presence of Poisson and Gaussian noise can significantly degrade the quality of these images, making accurate diagnoses more challenging. This study investigates the performance of non-local filters, particularly Non-Local Means and Block-Matching and 3D Filtering, in reducing noise in CT images. To achieve this, a methodology was implemented that applies these filters to images with varying noise levels, evaluating the results using the PSNR and SSIM metrics. Additionally, the impact of the Anscombe Transform on the linearization of Poisson noise was explored, enhancing the algorithms’ effectiveness. The results demonstrate that BM3D, combined with the Anscombe Transform, delivered the best performance in most scenarios, emerging as a promising tool for medical imaging applications.