Structure Adaptive Filtering for Edge-Preserving Image Smoothing

Abstract

In this paper, we propose a new edge-preserving image smoothing technique. A simple and effective scheme that classifies a pixel as situating on a corner, an edge or a plane has been developed. For the central pixel to be processed, nine adjacent support regions are constructed, leading to nine dimensional variation. Then the selected support region is adaptively determined by the coefficient of variation and variance, and finally the center pixel is updated iteratively according to the selected support region. More specifically, we show that a pixel at a location with very small variation is very likely situating on a plane (a smooth region). Otherwise, When the coefficient of variation is larger than the mean, then it is likely an edge pixel, otherwise it is a corner pixel. We adaptively select the appropriate filtering windows based on the local image structures to achieve excellent edge-preserving image smoothing. We present experimental results to show the effectiveness of our new technique.