Textured Image Denoising Using Dominant Neighborhood Structure

Abstract

This paper presents a new technique to texture image denoising using windowed nonlocal means with dominant neighborhood structure. The proposed method mainly addresses the incurred high blurring when the windowed nonlocal means is applied to texture images corrupted by high noise levels. The dominant neighborhood structure has been recently developed and successfully applied to texture classification (Khellah in IEEE Trans Image Process 20(11), 2011). Dominant neighborhood structure is an estimated global map representing the measured intensity similarity between any given image pixel and its surrounding neighbors within a certain search window. The map is used in this work to exclude all insignificant patches within the search window when computing the nonlocal means weights. The method is found to be quantitatively and qualitatively effective in denoising texture images when corrupted by high noise levels. In addition, the method highly reduces the computational requirements of the windowed nonlocal means filter.