An Image Denoising Algorithm Based on Modified Nonlinear Filtering

Abstract

Nonlinear filters have been widely used in noise reduction and smoothing of grayscale images. Nevertheless, these kinds of filters have limitation in image denoising because, while smoothing, in addition to the noise, they remove important features from the image. In this research, a new image denoising algorithm is proposed to retain important features and details while eliminating noise from the image. This algorithm is based on the use of nonlinear filters for smoothing grayscale images corrupted by random noise. The basic idea behind the proposed algorithm is to split noisy image into features and noise. The proposed method differs from ordinary nonlinear filters in the use of the idea of image noise reduction through iterative application of smoothing with successively different pixel neighborhood block sizes. Then, it calculates residuals between the original image and the smoothed versions with different pixel neighborhood sizes. The algorithm discards regions in the residuals that it ‘sees’ to contain noise by thresholding and some other cleaning operations. Then, a denoised image will be created by recombining the processed residual images with a selected smoothed version of the original image. Therefore, detect non-noise edges and features in the image, adapt and modify their behavior near the edges in order to preserve them. The proposed method outperformed the ordinary nonlinear filters by improving the subjective appearance of grayscale images corrupted by noises like Gaussian and in producing higher PSNR and lower MSE

Copyright © 2013 Praise Worthy Prize - All rights reserved.