Scientific Research and Essays Vol. 6(26), pp. 5523-5533, 9 November, 2011
DOI: 10.5897/SRE11.856
Available online at http://www.academicjournals.org/SRE
ISSN 1992-2248 © 2011 Academic Journals
Full Length Research Paper
An efficient implementation of switching median filter
with boundary discriminative noise detection for image
corrupted by impulse noise
Haidi Ibrahim*, Theam Foo Ng and Sin Hong Teoh
School of Electrical and Electronic Engineering, Engineering Campus, Universiti Sains Malaysia, 14300 Nibong Tebal,
Penang, Malaysia.
Accepted 18 October, 2011
Switching Median Filter with Boundary Discriminative Noise Detection (BDND) is one of the useful
methods that are capable to restore digital images which have been extremely corrupted by universal
impulse noise. Following the fundamental framework of the switching median filter, the construction of
BDND can be divided into two stages. The first stage classifies the pixels into either “noise” or “noise-
free” pixels, while the second stage restores the image by changing only the intensity values of the
“noise” pixels. Unfortunately, the originally proposed BDND employs sorting operations in both of its
stages. This condition makes the originally proposed BDND computationally expensive. Therefore, in
this paper, an implementation of BDND with reduced computational time is suggested. This reduction is
achieved mainly by manipulating the local histograms’ properties. Experimental results show that the
proposed implementation successfully produces the same results as the originally proposed BDND, but
with much shorter processing time.
Key words: Digital image processing, impulse noise, median filter.
INTRODUCTION
One of the noises that are commonly corrupting digital
images is the impulse noise. This type of noise can be
contributed by many factors such as by faulty imaging
sensors and from malfunction memory cells in storage
devices. Besides, impulse noise also can be created
during the transmission of the signal through noisy
channel (Chan et al., 2005). In general, the impulse noise
can be considered as an additive noise. This noise
changes the value of some pixels at random locations
into either relatively high or relatively low intensity value.
Noise pixels with high intensity values appear as white
dots on the image (that is salt), while noise pixels with
*Corresponding author. E-mail: haidi_ibrahim@ieee.org.
low intensity values appear as black dots on the image
(that is pepper). Therefore, impulse noise is also named
as the salt-and-pepper noise (Petrou and Bosdogianni,
2000). As the impulse pixels are having a relatively high
contrast toward their surrounding, even at low percentage
of corruption, the impulse noise can degrade the
appearance of the image significantly (Ibrahim et al.,
2008). Therefore, it is crucial for us to remove the
impulse noise before any subsequent image processing
operations such as image segmentation and pattern
recognition. Commonly, median filter which is a nonlinear
filter is employed to reduce the impulse noise in digital
images due to its sensitivity towards outliers (Eng and
Ma, 2001). The standard median filter (SM) operates by
defining a contextual region by using a sliding window of
size W W. It replaces the intensity value of the centre