On the Accuracy of Denoising Algorithms in Medical Imaging: A Case Study

Abstract

Anisotropic diffusion-based filters are a widespread used resource for medical image denoising because they are designed to preserve the image details during noise removal. This paper aims at providing a quantitative evaluation of this important feature without the inaccuracies of the commonly adopted full-reference metrics. For the first time, the true value of detail preservation yielded by an anisotropic diffusion filter is formally derived from the filter theory. Many computer simulations are reported in the paper in order to study how values and locations of errors representing filtering distortion depend upon the parameter settings.