2009 Second International Conference on Computer and Electrical Engineering

Image Denoising using Contourlet Transform

R.Sivakumar, G.Balaji,R.S.J.Ravikiran,R.Karikalan,S.Saraswathi Janaki

ECE Department
RMK Engineering College
Kavaraipettai, Tamilnadu, India
e-mail: [email protected]

Abstract— In image processing, considering fields like remote Low pass output is again passed through the Laplacian
sensing and medical applications, Speckle (Multiplicative) Pyramid [4] to obtain more coefficients and this is done till
noise dominates which affects the valuable features and the fine details of the image are obtained.
important information of the image. To denoise an image,
various transforms are used. But they are not efficient in case
of preserving the edges which is the important factor in image
processing for Image denoising. In this paper we have used the
new algorithm based on the transformation named ‘Contourlet
Transform’. This algorithm is more efficient than the wavelet
algorithm in Image Denoising particularly for the removal of
speckle noise. The parameters considered for comparing the
wavelet and Contourlet Transforms are SNR and IEF. The
results show that this proposed algorithm outperforms the
wavelet in terms of SNR, IEF values and visual perspective as
well.

Keywords- Denoising, Contourlet Transform, SNR.

I. INTRODUCTION
In Remote sensing applications, the major problem
arises with the Speckle noise. Speckle noise degrades the
quality of the image and affects the performance of
important image processing techniques such as detection,
segmentation, and classification. Till now it’s believed that Figure 1.1 Illustration of Contourlet Transform
Wavelet Transform is suited for denoising the Speckle
noise. But using this proposed algorithm based on The Contourlet Transform [2] of a signal ‘x’ is calculated
Contourlet Transform, we can achieve better results by passing it through a series of low pass and band pass
comparatively. filters. The output of these two filters is required to calculate
the Contourlet coefficients.
II. CONTOURLET TRAMSFORM
The Contourlet transform [2] is a directional transform ∞
which is capable of capturing contour and fine details in a Y low [n ] = ∑ x [k ]g [2 n − k ]
image. The approach in this transformation starts with the
k =∞−∞
discrete domain construction and then sparse expansion in
the continuous domain. The main difference between Y band [n ] = ∑ x [k ]h [2 n − k ]
Contourlet and other transformations is that, in this new k = −∞
transformation Laplacian pyramid [4,5] along with the
Directional Filter Banks [1, 3] are used. As a result, this not The above equations are the outputs of the filter and are
only detects the edge discontinuities, but also converts all the contourlet coefficients. The Low pass and Band pass
these discontinuities into continuous domain. The figure filters are used for Contourlet decomposition.
below illustrates the Contourlet Transformation, in which the
input image consists of frequency components like LL (Low III. DENOISING ALGORITHM
Low), LH (Low High), HL (High Low), and HH (High
Low). The Laplacian Pyramid at each level generates a Low A common approach for image denoising is to convert the
pass output (LL) and a Band pass output (LH, HL, and HH). noisy image into a transform domain such as the wavelet
The Band pass output is then passed into Directional Filter and Contourlet domain, and then compare the transform
Bank [1, 3] which results in Contourlet coefficients. The coefficients with a fixed threshold. We propose an algorithm

978-0-7695-3925-6/09 $26.00 © 2009 IEEE 22

DOI 10.1109/ICCEE.2009.70
which defines a new threshold value to eliminate the NOISY IMAGE
corrupted pixels.
CONTOURLET DECOMPOSITION

IV. ALGORITHM DESCRIPTION

A. The process starts with the noisy image. VARIANCE ESTIMATION

B. This image is converted into Contourlet Transform

domain, using the decomposition process. In Wavelet, THRESHOLDING

we determine the coefficients using scaling and a

wavelet filter. But, in Contourlet, we construct discrete-
FEATURE NOISE
domain multiresolution and multi direction using non- PRESERVATION SUPPRESSION
separable filter banks, in the same way wavelets are
obtained from the filter banks. This construction results
in flexible multiresolution, local and directional image CONTOURLET
expansion using contour sectors and so named RECONSTRUCTION

Contourlet transform.
C. Thus from the decomposition process the coefficients DENOISED IMAGE
are determined.
D. Then for each noisy image pixels, the variance is
estimated.
E. The resultant values are then compared with a threshold Figure 2.1 Threshold Algorithm
value to determine whether the pixel is corrupted or not.
F. If the pixels are corrupted, they are suppressed or There are many other algorithms available for denoising the
modified. Otherwise the pixels are preserved for further image particularly for the speckle corrupted images in
process. remote sensing applications. But this algorithm is very
G. Then all the resultant coefficients are reconstructed simpler and effective compared to other algorithms. If the
which results in denoised image. algorithm is simpler, then the time consumed for complete
denoising of the image will be less and the hardware
V. THRESHOLDING implementation will also be feasible with high memory
Generally for denoising, the coefficients of the noisy VLSI technologies.
image are compared with the threshold value. These
VI. RESULTS
threshold values are either obtained by trial and error method.
Since human eyes are very sensitive to intensity of The Standard test images considered for denoising are
neighboring pixel values, in image denoising techniques, the Lena, Pepper, Barbara and Satellite, Medical images. The
variance in homogeneous regions must be less. Considering evaluation parameters used for comparing Wavelet and the
the threshold values depending on the variance, the noise proposed algorithm using Contourlet Transform are SNR,
level in the corrupted image still decreases. In this algorithm, IEF and Visual quality assessment. The table 5.1 below
a threshold value is set based upon the variance of the shows the comparative results between wavelet and proposed
corrupted image. Based upon the results from various algorithm using Contourlet Transform.
variance levels (nvar), the threshold is fixed. The intensity of IEF (Image Enhancement Factor) is used to determine
the noise being added to the image (th) and the standard the edge preserving capabilities of the considered
deviation of the noise less image (sigma) are also the transformation. IEF can be defined as the ratio between the
deciding factors in fixing the threshold values. Based upon square of difference between the noisy image and the
the various results obtained, we deduce that the threshold original image to the square of difference between the
values must be fixed depending upon the high noise level and enhanced image and the reference image. If the IEF (Image
low noise level. In Speckle noise, the default variance level is Enhancement Factor) value is high, it means that more
0.04, so considering Speckle noise variance (nvar) above edges are preserved and the information in those edge
0.05 as high noise level and below 0.05 as low noise level, regions also can be extracted. For different Speckle noise
we introduced two threshold values separately. The results level densities we have obtained various SNR and IEF
also prove that this two separate threshold values improve the values of Wavelet and the proposed algorithm using
denoising ability of the algorithm. Now to reconstruct the Contourlet Transform, that are tabulated. From these
image, the coefficients above the threshold values are quantitative results we infer that the new proposed
retained for Contourlet reconstruction and the coefficients algorithm using Contourlet Transform outperforms Wavelet.
below the threshold values are suppressed. The retained
coefficients are reconstructed to obtain the Denoised image.
This process is shown in the figure 2.1.

23
 Table5.1 Comparative results between wavelet and proposed algorithm
using Contourlet

Images SNR(dB) IEF

Wavelet Proposed Wavelet Proposed
algorithm algorithm
using using
Contourlet Contourlet
NOISE LEVEL=0.03dB
a) Original image b) noisy image
LENA 9.44 11.77 2.01 3.45
PEPPER 10.73 11.36 2.10 2.43
SATELLITE 8.22 10.32 1.85 3.00
MEDICAL 13.36 15.73 3.02 5.25
BARBARA 10.04 9.81 1.65 1.57
NOISE LEVEL=0.04dB
LENA 7.94 10.95 1.87 3.75
PEPPER 9.21 10.68 1.97 2.75
c) Wavelet denoising d) Proposed algorithm
SATELLITE 6.65 9.34 1.69 3.16
SNR = 3.15 dB SNR = 8.20 dB
MEDICAL 11.55 14.81 2.66 5.62 IEF = 1.47 IEF = 4.75
BARBARA 8.84 9.39 1.66 1.88
Figure 5.1 Results of filter(for Lena image with speckle
NOISE LEVEL=0.06dB
variance= 0.1)
LENA 5.75 10.01 1.66 4.44
PEPPER 7.02 9.65 1.75 3.19
Another test image considered for experimenting is the
‘Satellite image’, shown in 5.2a. The image is corrupted
SATELLITE 4.47 8.40 1.52 3.76 with the Speckle noise variance of 0.04 and the corrupted
MEDICAL 8.84 14.56 2.13 8.03 image is shown in fig 5.2b. The fig 5.2c show the Wavelet
BARBARA 6.99 8.67 1.58 2.33
denoised image and the fig 5.2d shows the proposed
algorithm based on Contourlet transformation. In remote
NOISE LEVEL=0.1dB sensing applications, extracting information from the
LENA 3.15 8.20 1.47 4.75 contours is crucial. With high IEF value it’s apparent in the
PEPPER 4.44 8.56 1.55 3.97
proposed algorithm that it contains more information to be
extracted
SATELLITE 1.95 6.48 1.37 3.88
MEDICAL 5.65 12.00 1.71 7.36
BARBARA 4.56 7.66 1.45 2.96

The test image considered for testing is the ‘Lena’ image,

shown in fig 5.1a. The image is corrupted with the Speckle
noise variance of 0.1 and the corrupted image is shown in fig
5.1b. The fig 5.1c show the Wavelet denoised image and the
fig 5.1d shows the proposed algorithm based on Contourlet
transformation. There is a significant improvement in SNR
and IEF values in the proposed algorithm compared to the
Wavelet transform. Visually too, it’s evident that the a) Original image b) noisy image
Proposed algorithm has more information comparatively.

24
 is concluded based on considering test images like Lena,
Barbara, Peppers along with Satellite images and Medical
images after corrupting with Speckle noise which is a
multiplicative noise of various noise levels like 0.03dB,
0.04dB, 0.06dB and 0.1dB. Thus the obtained results in
qualitative and quantitative analysis shows that this proposed
algorithm outperforms the wavelet in terms of SNR, IEF
values and visual perspective as well. The algorithms are
implemented using MATLAB 7.5 R2007b. This can be
implemented using hardware also and the feasible hardware
is using VLSI with high memory.
c) Wavelet denoising d) Proposed algorithm
SNR = 6.65 dB SNR = 9.34 dB
IEF = 1.69 IEF = 3.16 REFERENCES

So it is evident with the results that this proposed [1] W.Y. Chan, N.F. Law, W.C. Siu, “Multiscale feature analysis using
algorithm based on Contourlet transform is best suit for directional filter bank,” Proceedings of the 2003 Joint Conference of
the Fourth International Conference on Information, Communications
Satellite image applications. and Signal Processing, December 2003, pp. 822-826.
[2] M. N. Do and M. Vetterli, “Contourlets: a Directional Multiresolution
image representation,” Proceedings of 2002 IEEE International
VII. CONCLUSION Conference on Image Processing, vol. 1,2002, pp 357-360.
In this paper, the removal of speckle noise from [3] M. N. Do and M. Vetterli, “Pyramidal directional filter banks and
images has been discussed. Hence the new proposed curvelets,” Proc. IEEE Int. Conf. on Image Proc., vol. 3, 2001, pp.
158-161.
algorithm based on the Contourlet transformation is found to
be more efficient than the wavelet algorithm in Image [4] P. J. Burt and E. H. Adelson, “The Laplacian pyramid as a compact
image code,” IEEE Trans. Commun., vol. 31, no. 4, 1983, pp. 532–
Denoising particularly for the removal of speckle noise. This 540.
[5] Zhiling Longa, b and Nicolas H. Younana “Denoising of images with
Multiplicative Noise Corruption”, 13th Europian Signal Processing
Conference, 2005,,a1755.

25