Medical image denoising from similar patches derived by Rough Set

Abstract

Current state-of-the-art research on denoising involves patch similarity. The similar patches are obtained either from image itself or from dictionary of patches. This paper proposes a new way to find similar patches from a given image using Rough Set Theory (RST). Search for similar patches is usually restricted locally. However, a global search could fetch patches which are more similar. The current RST based approach is enabling such search global and hence satisfying the Non-local principal which is the basis for patch based denoising. Like a few other denoising techniques, the framework of nonlocal means and principal component analysis both are then utilized to denoise medical images. The main essence of the current work reflects true sense of non-locality of similar patches. Exhaustive experiments clearly indicate comparability of the current proposal to the state-of-the-art methods in the light of several evaluation measures.