IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 19, NO. 9, SEPTEMBER 2010 2307
Switching Bilateral Filter With a Texture/Noise
Detector for Universal Noise Removal
Chih-Hsing Lin, Jia-Shiuan Tsai, and Ching-Te Chiu
Abstract—In this paper, we propose a switching bilateral filter
(SBF) with a texture and noise detector for universal noise removal.
Operation was carried out in two stages: detection followed by
filtering. For detection, we propose the sorted quadrant median
vector (SQMV) scheme, which includes important features such as
edge or texture information. This information is utilized to allo-
cate a reference median from SQMV, which is in turn compared
with a current pixel to classify it as impulse noise, Gaussian noise,
or noise-free. The SBF removes both Gaussian and impulse noise
without adding another weighting function. The range filter in-
side the bilateral filter switches between the Gaussian and impulse
modes depending upon the noise classification result. Simulation
results show that our noise detector has a high noise detection rate
as well as a high classification rate for salt-and-pepper, uniform
impulse noise and mixed impulse noise. Unlike most other impulse
noise filters, the proposed SBF achieves high peak signal-to-noise
ratio and great image quality by efficiently removing both types
of mixed noise, salt-and-pepper with uniform noise and salt-and-
pepper with Gaussian noise. In addition, the computational com-
plexity of SBF is significantly less than that of other mixed noise
filters.
Index Terms—Gaussian noise, image restoration, impulse noise,
mixed noise, nonlinear filters, switch bilateral filter, switching
scheme.
I. INTRODUCTION
N
OISE is introduced into images during acquisition, signal
amplification and transmission [6], [27]–[31]. An impor-
tant problem of image processing is to effectively remove noise
from an image while keeping its features. Noise removal is a dif-
ficult task because images may be corrupted by different types
of noise, such as additive, impulse or signal dependent noise
[27]–[29]. The solution depends upon the type of noise added
to the image [28]–[30]. Linear filtering possesses mathematical
simplicity and offers satisfactory performance on images with
additive Gaussian noise [29]–[31]. However, linear techniques
blur edges and fail for non-Gaussian and/or impulse noise. This
Manuscript received September 02, 2009; revised March 02, 2010; accepted
March 20, 2010. First published April 08, 2010; current version published Au-
gust 18, 2010. The associate editor coordinating the review of this manuscript
and approving it for publication was Prof. Kiyoharu Aizawa.
C.-H. Lin is with the Institute of Communications Engineering, National
Tsing-Hua University, Hsinchu, Taiwan. R.O.C. (e-mail: chihhsinglin@gmail.
com).
J.-S. Tsai is with the Department of Computer Science, National Tsing-Hua
University, Hsinchu, Taiwan, R.O.C. (e-mail: thymine666@gmail.com).
C.-T. Chiu is with the Department of Computer Science and Institute
of Communications Engineering, National Tsing-Hua University, Hsinchu,
Taiwan, R.O.C. (e-mail: ctchiu@cs.nthu.edu.tw).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TIP.2010.2047906
disadvantage leads to the use of nonlinear filtering in image pro-
cessing [29]–[31].
In this paper, we propose a filtering scheme that can remove
both the additive Gaussian noise and the impulse noise. Addi-
tive Gaussian noise is characterized by adding to each image
pixel a value with a zero-mean Gaussian distribution [6], [27].
Such noise is usually introduced during image acquisition [27].
The zero-mean distribution property allows such noise to be
removed by averaging pixel values locally [27]. Traditional
linear filters can remove noise effectively but with the side
effect of blurring edges and details significantly [6], [31],
[32]. The more advanced methods for noise removal aim at
preserving edges and details in images while removing the
noise [25], [32]. Tomasi and Manduchi propose a bilateral filter
that uses weights based upon spatial and radiometric similarity
[1]. The bilateral filter has good results in removing noise while
preserving edges in images [1], [32]. In addition, this method
is noniterative, local and simple [1], [32].
Impulse noise is characterized by replacing a portion of
an image pixels with noise values, leaving the remainder un-
changed [29]. Such noise is introduced due to acquisition or
transmission errors [27], [29], [30]. Nonlinear filters have been
developed for removing impulse noise such as the traditional
median filter [29]. Extensions of the median filter [2], [5],
[14]–[21], [23], [24], [26] are proposed to meet various criteria,
e.g., robustness, preservation of edge, or preservation of details.
The learning algorithm [7] and the switching noise filters
[14], [16], [17], [19], [22]–[24], [26] are also proposed. The
genetic programming (GP) filter [7] is based upon the learning
algorithm that is used to build two detectors, and this method
requires a training procedure to arrive at an optimal classifi-
cation based upon the measure of pixels and their neighbors.
Methods that require training are less easily controlled and
more unpredictable than traditional methods. The switching
scheme detects impulse noise pixels before filtering and re-
places them with estimated values while leaving the remaining
pixels unchanged [14], [16], [17], [19], [22]–[24], [26].
Filters that can remove Gaussian or impulse noise, or any
mixture thereof, have also been proposed [3], [4], [6], [8]–[11],
[13]. The median-based signal-dependent rank ordered mean
(SDROM) filter can remove impulse noise rather effectively, but
when applied to images with Gaussian or mixed noise, it often
produces a visually disappointing output [8]. This is because
the rank-ordered mean gets corrupted in a high noise inten-
sity window. Another median-based filter, the adaptive center-
weighted median filter (ACWMF), uses a comparison of the
center weighted medians and adaptive thresholds for detection
[9]. When applied to Gaussian or mixed noise images, it cre-
ates blur and removes the details. The directional weighted me-
1057-7149/$26.00 © 2010 IEEE