Adaptive Optimum Notch Filter for Periodic Noise

Reduction in Digital Images

Payman Moallemi * and Majid Behnampourii

ABSTRACT
Periodic noises are unwished and spurious signals that create repetitive pattern on images and decreased
the visual quality. Firstly, this paper investigates various methods for reducing the effects of the periodic
noise in digital images. Then an adaptive optimum notch filter is proposed. In the proposed method, the
regions of noise frequencies are determined by analyzing the spectral of noisy image. Then, the repetitive
pattern of the periodic noise is produced by applying the corresponding notch pass filter. Finally, an output
image with reduced periodic noise is restored by an optimum notch filter method. The results of the proposed
adaptive optimum notch filter are compared by the mean and the median filtering techniques in frequency
domain. The results show that the proposed filter has higher performances, visually and statistically, and has
lower computational cost. In spite of the other compared methods, the proposed filter does not need to tune
any parameters.
KEYWORDS
Periodic Noise, Optimum Notch Filter, Detection of Noisy Frequencies, Region Growing

1. INTRODUCTION these parameters spatially vary [3]. In a multi-sensor

Additive periodic noise is a repetitive signal which is imaging system, unequal sensitivity of detectors and
added to the main signal. This periodic noise in a digital corresponding electronic circuits, causes the third type of
image is repetitive spatial pattern which effectively periodic noise called stripping. The number of available
degrades the image quality [1]. detectors in spatial imaging scanners determines the
There are some different sources for creation of period of stripes [4].
periodic noises in a digital image. Electrical or
electromechanical inferences in imaging systems,
electrical inference in image receiver systems, and
unequal sensitivity of detectors are the main sources. For
example, periodic noises can be seen when an imager
system is mounted on vibrated holder (for example in a
(a) (b) (c)
non-stabilized aerial imaging), due to mechanical
Figure 1: Image of body scan by X-Ray contaminated by
inference, or in a TV receiver when the receiving signal a)global b)local and c)stripping periodic noises.
is weak, due to electrical inference between receiving
signal and another periodic signal (for example inter- Nowadays, due to image quality importance, it has
harmonics of power supply frequency) [2]. been a lot attention to the periodic noise reduction
Additive periodic noises are usually modeled by algorithms. According to the location of sensors in a
summing several sinusoidal functions with different multi-sensor system, the location of stripping bands are
amplitudes and frequencies; therefore, in the frequency predetermined, then it is possible to reduce the noise
domain the noisy image appears like stars with high effectively using some spatial simple methods [4]. On the
amplitude. The periodic noises in the digital images are other hand, since the global and local periodic noises can
divided into three main categories, including global, local not be simply separated from the main image in the
and stripping [3-4], which are shown in Figure (1) in a spatial domain, the frequency-domain approaches are
typical host image. usually applied [1]. However, there are some reports for
In the global periodic noise, the noise parameters using spatial-domain approaches like soft morphology [5-
including corresponding amplitudes and frequencies are 6] in this subject.
spatial-independent, while in the local periodic noise The implementation of frequency-domain approaches

*
i Corresponding Author, P. Moallem is with the Department of Electrical Engineering, University of Isfahan, Isfahan, Iran (e-mail:
p_moalem@ui.ac.ir).
ii M. Behnampour is with Iran Aircraft Manufacturing (HESA), Shain Shar, Isfahan, Iran (e-mail: majidbehnam20@gmail.com).

1
Amirkabir / Electrical & Electronics Engineering / Vol . 42 / No.1 / Spring 2010
for reducing the periodic noise effect is either fully in error.
frequency-domain (like band reject filter), or used some
extracted information from the image in frequency-
domain (like optimum notch filter); therefore, the
frequency-domain analysis of the noisy image and
applying the filter in the frequency-domain are important
in the periodic noise reduction algorithms.
In the next section, an overview on some traditional
frequency-domain approaches in the periodic noise
reduction subject is studied. Then the proposed optimum
notch filter which is adaptively used some extracted
information from the noisy image in frequency-domain is
presented, and finally the results are compared with some
other similar methods. Regarding to the other methods,
the results of the proposed method show higher
improvements in the output images, qualitatively and
quantitatively. Again, the computational cost of the
proposed method significantly reduces which is
important in a real-time application.

2. AN OVERVIEW ON EXISTING METHODS

Figure 2: Different step to reduce the periodic noise effect from
In this section, a brief overview on existing methods a sample joker image. a)Input image contaminated by additive
in frequency-domain for periodic noise reduction two periodic sinusoidal signals. b) 2-D Fourier transform of the
algorithms including band reject filter, notch filter, input image. c) Removing periodic noise frequencies by the
optimum notch filter, frequency-domain median filter and proper squared type notch-reject filter and d) Restored image
after applying the notch-reject filter.
frequency-domain masked mean filter is studied. It is
supposed that f(x,y) and g(x,y) are the pixel values of C. Optimum Notch Filters
noise-less and noisy image in the (x,y) coordinate, and In a real system contaminated by periodic noise, the
F(u,v) and G(u,v) are the 2-D Fourier transform of noise- output image tends to contain 2-D periodic structure
less and noisy image in the (u,v) frequency, respectively. superimposed on the input image with patterns more
A. Band-Reject Filters complex than several simple sinusoidal signals. In this
condition, two mentioned methods are not always
Since band reject filters attenuate a band of frequency
acceptable because they may remove much image
about the origin of the 2-D Fourier transform, these filter
information in filtering process.
can be uses in periodic noise reduction applications
Optimum notch filter tries to minimize local variance
where the general location of the noise components is
of the restored image [1]. At the first stage, principal
approximately known in the frequency domain [1].
contribution of the inference repetitive pattern is
Unfortunately, when the distances of the periodic noise
extracted from the noisy image, and then the output
components about the origin of the 2-D Fourier transform
image is restored by subtracting a variable weighted
are different, there is necessary to use either a wide band-
portion of the repetitive pattern from the contaminated
reject filter, or several narrow band-reject filters. In both
image. The extractions of the repetitive pattern is
cases, the restored image may lose some important image
implemented in frequency-domain by applying a proper
information.
notch-pass filter on every periodic noise frequency, and
B. Notch-Reject Filters then by applying inverse 2-D Fourier transform to restore
A notch-reject filter attenuates frequencies in the repetitive pattern in spatial-domain.
predefined neighborhoods about a center frequency of the The 2-D Fourier transform of the inference repetitive
2-D Fourier transform. Therefore, this filter can be used pattern, N(u,v), is given by Eq. (1),
to reduce the periodic noise effects where the main N (u , v ) = H np (u , v )G (u , v ) (1)
frequencies of the periodic noises are known [7].
Automatic determining of both the periodic noise main where Hnp(u,v) is superimposed response of all necessary
frequencies and the corresponding band-width are notch-pass filters and G(u,v) is the 2-D Fourier transform
challenge problems in this filter. Fig. 2 shows the of the contaminated image. The proper selecting of
different steps for reducing the periodic noise containing Hnp(u,v) is challenge problem in the optimum notch filter.
two main frequencies in a sample image. In this example, Then, the corresponding repetitive pattern in the
the periodic noise frequencies are determined by trail and spatial-domain, η(x,y), is obtained by Eq. (2),

2
Amirkabir / Electrical & Electronics Engineering / Vol . 42 / No.1 / Spring 2010
 η ( x, y ) = ℑ −1{N (u, v)} = ℑ −1{H np (u, v)G (u, v)} (2) ⎧ X (u , v)
>θ
⎪Med( X (u , v)) if
Med( X (u , v))
where F -1 is a symbolic representation of inverse 2-D ⎪
Y (u , v) = ⎨ and (u, v) ≠ (0,0) (5)
Fourier transform. ⎪ X (u , v) otherwise
For an additive noise model, if η(x,y) is known ⎪
⎩
perfectly, subtracting the repetitive pattern from the noisy
where X(u,v) is |G(u,v)|, Med(X(u,v)) is the median value
image, g(x,y), obtains noise-less input image, f(x,y). On
of X(u,v) over neighborhood frequencies of the (u,v)
the other hand, the filtering procedure by applying
central frequency, θ is a predefined fixed threshold value
Hnp(u,v) usually yields only an approximation of the true
and Y(u,v) is the estimate of |F(u,v)|. Since X(0,0) is the
repetitive pattern. In order to minimize the effects of
DC component of the contaminated image and its value is
components not present in the estimate of real true
usually very large, the frequency-domain median filter
repetitive pattern, subtracting a variable weighted portion
should not apply on the frequency of (0,0).
of η(x,y) from g(x,y) is done to obtain an estimate of
The window size of 5×5, 7×7, 9×9 and 11×11 are
f(x,y).
proposed for the frequency-domain median filters. In
fˆ ( x , y ) = g ( x , y ) − w ( x , y )η ( x , y ) (3) fact, the window size is related to band-width of periodic
noise, and should be selected by trial and error.
Where fˆ ( x, y ) is the estimate of f(x,y) and w(x,y), the Meanwhile, the predefined fixed threshold value of 3 and
weighting function, is to be determined such that the 6 are proposed for the window sizes of 5×5 and 7×7,
variance of the output is minimized over a specific
respectively [8].
neighborhood of every point (x,y). Usually, the noisy
image is partitioned to several non-overlapped E. Frequency-Domain Masked Mean Filters
neighborhood of size (2a+1)×(2b+1) about points (x,y), The basic idea of this type of periodic noise reduction
and corresponding w(x,y) in each neighborhood is
considered constant. For each neighbor hood, the filters is similar to the frequency-domain median filter.
constant weighting function w(x,y) is obtained by Eq. (4) The frequency-domain masked mean filter uses the
which tries to minimize local variance of the restored masked mean values instead of the median ones [9]. The
image, over corresponding neighborhood [1]. masked mean value is defined over an N×N masked
window. All values in the N×N masked window are ‘1’,
g ( x, y )η ( x, y ) − g ( x, y )η ( x, y )
w = w( x, y ) = 2 (4) except the center which is considered ‘0’. It means that
η 2 ( x, y ) − η ( x, y ) the center of the N×N local window is omitted in mean
value computations.
Where g ( x, y ) and η ( x, y ) are the local mean of the
Suppose that S(u,v) is the masked mean value of
corrupted image and the noise pattern, respectively, X(u,v) over N×N masked window, the (u,v) frequency is
η 2 ( x, y ) is the local mean of square of the noise pattern, detected as noisy frequency when Eq. (6) is satisfied.
and g ( x, y )η ( x, y ) is the local mean of the corrupted X (u , v )
>θ (6)
image multiply by the noise pattern, all of them over S (u , v )
corresponding neighborhood.
In order to apply the optimum notch filter, firstly the where θ is a predefined fixed threshold value which is
noise pattern is extracted by Eq. (2), then for each empirically set, and it depends on noise power of the
neighborhood, the constant weighting function w(x,y) is periodic noise and the window size of the masked
obtained by Eq. (4) and finally, the restored image is window. For a 3×3 masked window, θ =4 is proposed
obtained by Eq. (3). [9].
D. Frequency-Domain Median Filters The realization of the frequency-domain masked mean
filter for the frequency of (u,v) is summarized by Eq. (7),
The frequency-domain median filter which is used to
⎧ S (u, v)
⎪ X (u , v) / δ >θ
reduce the periodic noise effect contains two basic steps. if
In the first step, the median value of amplitudes of each X (u , v)
⎪
frequency is computed over a predefined window. Then ⎪ and (u , v) ≠ (0,0)
Y (u , v) = ⎨ (7)
frequencies of the periodic noise are detected by ⎪
comparing the median value of each frequency with its ⎪ X (u , v) otherwise
⎪
corresponding amplitude. In the second step, amplitude ⎩
of each noisy frequency is replaced by it corresponding where δ is selected based on the periodic noise reduction
median value. The realization of the frequency-domain power of this filter. Like the previous filter, the frequency
median filter for the frequency of (u,v) is summarized by of (0,0) should be unchanged [9].
Eq. (5),

3
Amirkabir / Electrical & Electronics Engineering / Vol . 42 / No.1 / Spring 2010
F. Discussion about Existing Methods A. Detecting of the Main Frequencies of Periodic Noises
For band reject, notch and optimum notch filters, it As mentioned previously, the amplitude of frequencies
supposed that the periodic noise frequencies and of the periodic noise is locally greater than its
corresponding bands are predetermined, where the neighborhood. Therefore, they can be detected by
optimum notch filters show the best results [1]. Detection applying a proper threshold in the spectral of image. On
of the periodic noise frequencies is slightly solved in the the other hand, the low frequency region of ordinary
frequency-domain mean and median filters, but there are images contains most of image information image and
some extra parameters in these filters including window corresponding amplitude is greater than other
size and threshold that degrade the output quality if they frequencies. It means that it is necessary to mask the low
choose incorrectly. The frequency-domain median filter frequency region before applying the threshold. We
show higher quality than frequency-domain masked consider the radius of this mask as Rmask.
mean filter, but the corresponding computational cost is In order to determine the Rmask at the first step,
higher. suppose that there are non-overlapped concentric circles
at frequency center, with radius of R and width of w, in
3. THE PROPOSED METHOD the image spectra as shown in Figure 4.
The proposed method which is based on the optimum
notch filter, the accurate periodic pattern in the spatial
domain is adaptively detected. In the proposed method,
the spectral of the contaminated image is supposed as a
gray level image. Figure (2-b) shows the spectral of a
sample image contaminated by the periodic noises which
frequencies are highlighted as bright points. It means that
the corresponding main frequencies of periodic noises
can be detected by applying a proper thresholding after
masking the low frequency of the input image, as shown
in Figure (3). Figure 4: The non-overlapped concentric circles at frequency
center, with radius of R and width of w.
For spectra of Figure (2-a) which is shown in Figure
(2-b), the plot of mean of sum of amplitudes in each
circle versus radius of circles is shown in Figure (5). The
Rmask can be computed by detecting the first local
minimum of the plot.

Figure 3: The low frequency region of spectral of the input

image is firstly masked which is shown by the black circle in
the center then by applying a proper thresholding, the
corresponding frequencies of the periodic noises can be
detected which are highlighted by the white circles.

In order to accurately detect the corresponding pattern

of the periodic noise in spatial domain, it is necessary to
consider all contaminated frequencies around the main
frequencies of the periodic noise. In the proposed Figure 5: Plot of mean of sum of amplitudes of circles for
method, these contaminated frequencies are detected by a Figure (2-b) in term of R. The corresponding radius of the first
local minimum is computed as radius of mask, Rmask.
simple proposed region growing around each main
frequency of the periodic noise. After providing the
After masking the low frequency region, it is
pattern of the periodic noise in spatial domain by
necessary to compute the proper threshold which is
applying the inverse
determined by Eq. (8) for detecting the main frequency of
2-D Fourier transform on all frequencies of the periodic
each periodic noise.
noises, the input image is restored by the conventional
optimum notch filter. AThr = ( Amax + Amean ) / 2 (8)

4
Amirkabir / Electrical & Electronics Engineering / Vol . 42 / No.1 / Spring 2010
AThr is the proposed threshold, Amax and Amean are the growing algorithm is repeated to find all contaminated
maximum and mean of amplitudes of spectra, frequencies by periodic noises.
respectively, after masking the low frequency region.
This simple threshold has been already used for
effectively detecting the high temperature point targets in
infrared image [10].
B. Detecting Other Frequencies of Periodic Noise
Applying the computed threshold by Eq. (8) detects all
main frequencies of periodic noises, but for accurate
extraction of the periodic noise pattern in spatial domain, Figure 7: A part of spectra contaminated by periodic noise in
the other frequencies of the periodic noise should be the right side, the main frequency of periodic noise is
highlighted by white. The region of detected frequencies of
detected and considered. In this step, a simple region
periodic noise is masked by black in the left side.
growing algorithm [11] which considers a 3×3 window
around each main frequency as primary region is C. Applying the Optimum Notch Filter
proposed in the spectral domain. At the first step of The output of applying the proposed thresholding and
region growing, a surrounding 5×5 window is considered region growing algorithm on the spectra of the input
and the proposed region growing algorithm tries to find noisy image is a proper notch pass filter, Hnp(u,v), which
which frequencies in perimeter of surrounding 5×5 can be applied on spectra of the input noisy image to
window are contaminated by the periodic noise by extract the spatial pattern of the periodic noise based on
comparing the amplitude of neighborhood frequencies as Eq. (2). Then the restored image can be computed by
shown in Figure (6). using Eqs. (3) and (4).
4. EXPERIMENTAL RESULTS
The proposed adaptive optimum notch filter as well as
the frequency-domain median and masked mean filters
are implemented under Matlab environment [12]. The
results on different images and periodic noises are
quantitatively and qualitatively compared. In order to
compare the results quantitatively, Mean of Absolute
Error (MAE) between the noiseless image and restored
image is presented. Moreover, to compare the output
images qualitatively, the histogram equalization
Figure 6: The star sign shows the main frequency of periodic algorithm is applied on the restored image to visually
noise. The 3×3 window around the main frequency is also highlight the differences. Two different experiments are
considered as corresponding frequencies of periodic noise
reported, at the first one, a sinusoidal periodic noise with
which is shown by multiply sign. The corresponding neighbors
of each frequencies in surrounding 5×5 window are shown by
variable amplitude is applied and the results are
arrows. quantitatively and qualitatively compared. At the second
Each frequency in perimeter of 5×5 window is experiment, all algorithms are applied on a sample TV
considered as periodic noise, if the corresponding image contaminated by a complex real periodic noise and
amplitude is less than the amplitude of neighborhood the results are qualitatively compared. It is necessary to
frequencies in 3×3 window. If the number of frequencies tune the parameters of the mean and median filters by
contaminated by periodic noise in perimeter of 5×5 trail and error to achieve the best results. In this
window is greater than the predefined threshold, NThr, the experiment, the window sizes and the thresholds for
region growing process repeats for a surrounding 7×7 mean and median filters are set to 3×3 and 15, and 9×9
window. In the proposed region growing algorithm, the and 6, respectively.
NThr is set to half of pixels in perimeter of surrounding A. Sinusoidal Periodic Noise with Variable Amplitude
window which is (2n-2) where n is surrounding window The two dimensional sinusoidal periodic noise, η(x,y)
size and n is always odd number. After stopping the
which is given by Eq. (9), is added to a sample 256×256
proposed region growing algorithm, all frequencies of the
gray level image.
grown region are considered as frequencies of the
periodic noise. Figure (7) shows the output of applying η ( x, y ) = 1 + a.sin( x / 2, y / 1.5) (9)
the proposed region growing algorithm on a sample part At the first part of this experiment, a is fixed to 0.1.
of spectra which is contaminated by a periodic noise. Figure (8) shows the noisy image and the restores images
Again, for each detected main frequency of periodic by the mean and median filters in the frequency domain
noise in the thresholding stage, the proposed region as well as the proposed adaptive optimum notch filter. In

5
Amirkabir / Electrical & Electronics Engineering / Vol . 42 / No.1 / Spring 2010
order to present a better comparison between methods, Figure (9) shows that the MAE of the proposed
the histogram equalization is applied on all restored adaptive optimum notch filter is always higher than MAE
images. As shown in Figure (8), the improvement of the of the other compared methods. Moreover, when the
proposed method is higher than the other compared amplitude of the sinusoidal periodic noise increases, the
method, qualitatively. quantitative difference between the proposed method and
two other compared ones increases, because the proposed
adaptive method could adapt itself in different noise
amplitudes.
B. Complex Real Periodic Noise
At the second experiment which results are shown in
Figure (10), all compared methods are applied on a
sample 640×580 TV image contaminated by a complex
real periodic noise. As shown in Figure (10), the visual
quality of the restored image by the proposed adaptive
optimum notch filter is a little better than two other
compared methods.
C. Execution Times
Complexity is another aspect of a restoration
algorithm. The implementation cost of low complexity
algorithm is lower than more complex algorithm. In order
to compare the complexity of the compared algorithm,
the execution times of the compared algorithms are
computed. Since the execution time of the masked mean
Figure 8: The visually comparison between the output of filter in the frequency domain is lower than the median
different restoration methods, after applying the histogram one, the execution time of the masked mean filter is
equalization. The image contaminated by the periodic noise of considered as execution time unit. Table (1) reports the
Eq. (8) with a=0.1 (a), and the restored images by the masked execution times of all compared restoration algorithms
mean filter (b), the median filter (c) and finally the proposed for Figures (8) and (10).
adaptive optimum notch filter (d).
TABLE 1
EXECUTION TIMES OF THE COMPARED RESTORATION METHODS
At the second part of this experiment, a in Eq. (9) is Restoration Filters Figure (8) Figure (10)
varied between to 0.0 to 1.0, and three methods including
Mean 1.0 1.0
mean and median filters both in frequency domain and
Median 20.0 94.5
the proposed adaptive optimum notch filter are used to
Adaptive Optimum Notch 0.57 0.73
restore the noiseless image. The parameters of the mean
and median filters in the frequency domain are selected
Table (1) shows that the execution time of the
as the corresponding parameters in the previous
proposed adaptive optimum notch filter is lower than the
experiments. In order to quantitatively compare the
masked mean filter which execution time is very lower
results, MAE of each method is computed and plotted in
than the median one.
Figure (9).
5. CONCLUSION
In this paper, an adaptive optimum notch filter is
proposed for reducing the effect of periodic noise in
digital image. Firstly, the proposed algorithm determines
the main frequency of the periodic noise by applying the
proposed thresholding algorithm on the spectra of noisy
image. Then, by applying the proposed region growing
on each main frequency of the periodic noise, related all
contaminated frequencies by periodic noise are separately
determined. Finally the periodic noise pattern is
computed by applying inverse 2-D Fourier transform, and
then the restored image is obtained by applying the
Figure 9: The plot of variation of MAE in term of amplitude of optimum notch filter method.
periodic noise, a, in Eq. (9).

6
Amirkabir / Electrical & Electronics Engineering / Vol . 42 / No.1 / Spring 2010
Figure 10: The visually comparison between the output of different restoration methods for restoring the TV image contaminated by
a real complex periodic noise. a) Noisy image, and the restored images by b) The masked mean filter, c) The median filter and
finally d) The proposed adaptive optimum notch filter.

The proposed adaptive optimum notch filter, the mean of restored image by the proposed method is higher than
and median filters in the frequency domain are used to two other compared methods, but also its execution time
reduce the effects of different periodic noises in different is lower. In spite of the other compared methods, the
sample images. The results show that not only the quality proposed filter does not need to tune any parameters.

6. REFERENCES
[1] R.C. Gonzalez, R.E. Woods, Digital Image Processing, 3rd Edition, [8] I. Aizenberg, C. Butakoff, "Frequency Domain Median-like Filter
Prentice Hall, 2007. for Periodic and Quasi-Periodic Noise Removal," SPIE Proceeding,
[2] R. Drigger, P. Cox, T. Edwards, Introduction to Infrared and Vol. 4667, 181-191, 2002.
Electro-Optical System, Artech House Publishers, 1999. [9] I. Aizenberg, C. Butakoff, J. Astola, K. Egiazarian, "Nonlinear
[3] R.A. Schowengerdt, Remote Sensing: Models and Methods for Frequency Domain Filter for the Quasi-Periodic Noise Removal,"
Image processing, 2nd Edition, Academic Press, 1997. in Proc. 2002 International TICSP Workshop on Spectral Methods
[4] V. Castelli, L.D. Bergman, Image Databases, Search and Retrieval and Multirate Signal Processing, pp. 147-153.
of Digital Imagery, John Wiley & Sons, 2002. [10] R. Venkateswarlu, K.V. Sujata, B. Venkateswara Rao, "Centroid
[5] T.Y. Ji, M.S. Li, Z. Lu, O.H. Wu, "Optimal morphological filter Tracker and Point Selection," SPIE Proceeding, Vol. 1697, pp.
design using a bacterial swarming algorithm," in Proc. 2008 IEEE 520-529, 1992.
Congress on Evolutionary Computation, pp. 452-458. [11] R. Szeliski, Computer Vision: Algorithms and Applications,
[6] T.Y. Ji, Z. Lu, O.H. Wu, "Optimal soft morphological filter for Springer, 2010.
periodic noise removal using a particle swarm optimiser with [12] R.C. Gonzalez, R.E. Woods, S.L Eddins, Digital Image Processing
passive congregation," Signal Processing, Vol. 87, Issue 11, pp. using MATLAB, Gatesmark Publishing, 2nd Edition, 2009.
2799-2809, 2007.
[7] I. Aizenburg, C. Butakoff, "A windowed Gaussian notch filter for
quasi-periodic noise removal," Image and Vision Computing, Vol.
26, Issue 10, pp. 1347-1353.

7
Amirkabir / Electrical & Electronics Engineering / Vol . 42 / No.1 / Spring 2010