Robustness of Median Filter For Suppression of Salt and Pepper Noise (SPN) and Random Valued Impulse Noise (RVIN)
A. Zubair, and H. Busari. International Journal of Image Processing (IJIP), 12 (1):
12-27(April 2018)
Abstract
Noises in images are caused by many sources. Image de-noising has remained an active area of research. Results of numerical experiments on the robustness of median filter for suppression of Salt and Pepper Noise (SPN) and Random Valued Impulse Noise (RVIN) of varying noise densities are presented and discussed. Varying densities of SPN and RVIN were simulated and used to corrupt five selected test images which have different image frequencies. The corrupted images were filtered with Median Filters which has 3 by 3 kernel size. The effects of larger kernels were also examined. The performance metrics are the Peak Signal to Noise Ratio (PSNR) and Gain. SPN is found to have more adverse effects on images than RVIN. However, the Median filter is found to achieve a higher degree of noise suppression with SPN than RVIN. Effects of SPN and RVIN increase with an increase in noise density. Median filtering of SPN and RVIN corrupted images are found to be satisfactory with 3 by 3 kernel for noise densities up to the maximum of 60% and 40% noise densities respectively. Median filter Gain is found to increase with noise density up 40% and then reduce with further increase in noise density. To some extent, there is some correlation between Median filter gain and test image frequency. Using 5 by 5 kernel may improve noise suppression but the resulting filter image is blurred. 3 by 3 is the optimum kernel size.
%0 Journal Article
%1 zubair2018robustness
%A Zubair, Abdul Rasak
%A Busari, Hammed Oyebamiji
%D 2018
%J International Journal of Image Processing (IJIP)
%K Density, Filter, Frequency, Image Median Noise Noise, Peak Ratio Signal To
%N 1
%P 12-27
%T Robustness of Median Filter For Suppression of Salt and Pepper Noise (SPN) and Random Valued Impulse Noise (RVIN)
%U http://www.cscjournals.org/library/manuscriptinfo.php?mc=IJIP-1143
%V 12
%X Noises in images are caused by many sources. Image de-noising has remained an active area of research. Results of numerical experiments on the robustness of median filter for suppression of Salt and Pepper Noise (SPN) and Random Valued Impulse Noise (RVIN) of varying noise densities are presented and discussed. Varying densities of SPN and RVIN were simulated and used to corrupt five selected test images which have different image frequencies. The corrupted images were filtered with Median Filters which has 3 by 3 kernel size. The effects of larger kernels were also examined. The performance metrics are the Peak Signal to Noise Ratio (PSNR) and Gain. SPN is found to have more adverse effects on images than RVIN. However, the Median filter is found to achieve a higher degree of noise suppression with SPN than RVIN. Effects of SPN and RVIN increase with an increase in noise density. Median filtering of SPN and RVIN corrupted images are found to be satisfactory with 3 by 3 kernel for noise densities up to the maximum of 60% and 40% noise densities respectively. Median filter Gain is found to increase with noise density up 40% and then reduce with further increase in noise density. To some extent, there is some correlation between Median filter gain and test image frequency. Using 5 by 5 kernel may improve noise suppression but the resulting filter image is blurred. 3 by 3 is the optimum kernel size.
@article{zubair2018robustness,
abstract = {Noises in images are caused by many sources. Image de-noising has remained an active area of research. Results of numerical experiments on the robustness of median filter for suppression of Salt and Pepper Noise (SPN) and Random Valued Impulse Noise (RVIN) of varying noise densities are presented and discussed. Varying densities of SPN and RVIN were simulated and used to corrupt five selected test images which have different image frequencies. The corrupted images were filtered with Median Filters which has 3 by 3 kernel size. The effects of larger kernels were also examined. The performance metrics are the Peak Signal to Noise Ratio (PSNR) and Gain. SPN is found to have more adverse effects on images than RVIN. However, the Median filter is found to achieve a higher degree of noise suppression with SPN than RVIN. Effects of SPN and RVIN increase with an increase in noise density. Median filtering of SPN and RVIN corrupted images are found to be satisfactory with 3 by 3 kernel for noise densities up to the maximum of 60% and 40% noise densities respectively. Median filter Gain is found to increase with noise density up 40% and then reduce with further increase in noise density. To some extent, there is some correlation between Median filter gain and test image frequency. Using 5 by 5 kernel may improve noise suppression but the resulting filter image is blurred. 3 by 3 is the optimum kernel size.},
added-at = {2018-12-12T07:22:39.000+0100},
author = {Zubair, Abdul Rasak and Busari, Hammed Oyebamiji},
biburl = {https://www.bibsonomy.org/bibtex/298c50e017079537f2c86a2263af47ac2/cscjournals},
interhash = {66b21e14a877497a8adeb506008ed274},
intrahash = {98c50e017079537f2c86a2263af47ac2},
issn = {1985-2304},
journal = {International Journal of Image Processing (IJIP)},
keywords = {Density, Filter, Frequency, Image Median Noise Noise, Peak Ratio Signal To},
month = {April},
number = 1,
pages = {12-27},
timestamp = {2018-12-12T07:22:39.000+0100},
title = {Robustness of Median Filter For Suppression of Salt and Pepper Noise (SPN) and Random Valued Impulse Noise (RVIN)},
url = {http://www.cscjournals.org/library/manuscriptinfo.php?mc=IJIP-1143},
volume = 12,
year = 2018
}