Salt-and-Pepper Noise Removal and Detail Preservation Using Convolution Kernels and Pixel Neighborhood
Zayed M. Ramadan · American Journal of Signal Processing · 2014
This paper introduces a method for removal of salt-and-pepper impulsive noise from images while preserving edges and fine details. The method consists of two stages: detection and filtering. In the detection stage, two conditions must be satisfied for a pixel to be considered noisy. The first condition is based on convolution of the corrupted image with four convolution kernels and the second depends on the pixel under consideration in the sliding window and its neighborhood. In the filtering stage, the conventional median filtering is used except that only pixels that are considered noise-free in the sliding window of the detection stage are included in the calculations of the median value that replaces the corrupted pixel value. Small size of sliding windows and wide range of noise densities are used in this paper. Simulation results using many images of different features show superior results of the proposed method over other well-known methods in the literature of image restoration.