Beruflich Dokumente
Kultur Dokumente
Abstract: The task of decomposing image into its simpler components is great interest for many applications.
Such decomposition method is important for applications in image coding, image analysis and synthesis. This
paper presents a method for decomposing image into its number of high and low frequency image components.
The proposed approach is based on empirical wavelet transform method. The basic idea promoted in this paper
is to presents another methodology to build adaptive wavelets for both color and grayscale image. In EWT each
decomposed mode represents a particular type of image content (either high or low frequency image
components). The new wavelet transform utilized for adaptive decomposition of the color image. Different
experiments are performed to adaptive decompose the color images and all experimental results validate the
algorithm's performance.
Keywords: Wavelet, Empirical wavelet transform for images, Adaptive filtering
I.
Introduction
www.iosrjournals.org
36 | Page
Existing Approaches
f t =
fs t 1
s=0
An IMF is an amplitude modulated-frequency modulated function which can be written in the form
fs t = Fs (t)cos
(s (t))Fs (t), s (t)>0 t
(2)
The main assumption is that Fs and s vary much slower thans . The IMF fs behaves as a harmonic
component. Originally, the method of Huang et al. [9] to extract such IMFs is a pure algorithmic method.
Candidates for an IMF are extracted by first computing the upper,f(t) and lowerf(t), envelopes via a cubic
spline interpolation from the maxima and minima of f . Then the mean envelope is obtained by computing
m(t) = (f(t) + f(t))/2 and finally the candidate by r1 (t) = f t m(t). Generally, r1 t does not fulfill the
properties of an IMF. A good candidate can be reached by iterating the same process to r1 and the
subsequentrs . The final retained IMF isf1 (t) = rn (t). Then the next IMF is obtained by the same algorithm
applied onf(t) f1 (t). The remaining IMFs can be computed by repeating this algorithm on the successive
residues.
Consequently it is difficult to really understand what the EMD provides. For example, some problems
appear when some noise is present in the signal. To deal with this problem, an Ensemble EMD (EEMD) was
proposed in [17]. The authors propose to compute several EMD decompositions of the original signal corrupted
by different artificial noises. Then the final EEMD is the average of each EMD. This approach seems to
stabilize the obtained decomposition but it increases the computational cost. Another EMD approach is
proposed in [8]. The authors proposed to minimize a functional which looks for a sparse representation of, f in a
dictionary of IMFs. This variational method provides similar results as the original EMD algorithm.
B. Wavelets approaches
The mathematical tool wavelet transform and the theoretical framework for wavelets, has been widely
developed and used in a variety of areas of sciences. Wavelet based methods, e.g. signal and image processing
such as noise reduction, compression and image decomposition have been proven to be extremely useful for a
number of applications, see for example [3], [10],[12],and [14]. The Fourier transform and its inverse are
denoted f andf, respectively. In the temporal domain, a wavelet dictionary {u,s } is defined as the dilated, with a
parameter s>0, and translated by u R of a mother wavelet (of zero-mean) as
u,s t =
1
s
tu
.
s
(3)
Then the wavelet transform of f is obtained by computing the inner productsWf u, s = f, u,s . If s is
a continuous variable then Wf u, s is called the continuous wavelet transform while if s = aj then Wf u, s =
Wf u, j is called the discrete wavelet transform. A useful property of the wavelet transform is that it can be
viewed as the application of a filter bank (each filter corresponds to one scale). In practice, the most used case is
the dyadic case, s = 2j analysing a signal with the Fourier transform leads to information about the frequency
spectrum. However, the standard form to analyse the Fourier transform is to study the behaviour of |f ()|, which
does not refer to at what time the phenomenon that cause the frequency peak(s) occur. Also, small transient
outbursts give almost no noticeable contribution to the frequency spectra since it is the average of the
frequencies that is measured.
Even though the wavelet packets are useful in many applications, they use a constant prescribed ratio in
the subdivision scheme, which limits their adaptability. Another approach, called the Malvar-Wilson wavelets
[10], [13], tries to build an adaptive representation by segmenting the temporal signal itself in order to separate
the time intervals containing different spectral information. While the original idea is interesting, it turns out that
DOI: 10.9790/4200-05133644
www.iosrjournals.org
37 | Page
III.
Empirical Wavelets
A. Definition
We propose a method to build a family of wavelets adapted to the processed signal. If we take the
Fourier point of view, this construction is equivalent to building a set of band pass filters. One way to reach the
adaptability is to consider that the filters supports depend on where the information in the spectrum of the
analysed signal is located. Indeed, the IMF properties are equivalent to say that the spectrum of an IMF is of
compact support and centered around a specific frequency (signal dependent). For clarity, we only consider real
signals (their spectrum is symmetric with respect to the frequency = 0) but the following reasoning can be
easily extended to complex signal by building different filters in the positive and negative frequencies,
respectively. We also consider a normalized Fourier axis which has a 2 periodicity, in order to respect the
Shannon criteria, and we restrict our discussion to [0, ]. Let us start by assuming that the Fourier
support[0 ] is segmented into N contiguous segments. Let n to be the limits between each segments. Each
segment is denoted n = [n1 , n ] .Centered around each n we define a transition phase Tn of width, 2n .
The empirical wavelets are defined as bandpass filters on each n .Thus define the empirical scaling function
and the empirical wavelets by expressions of (4) and (5), respectively
1
if n n .
1
n = cos
n + n
2 2n
0
otherwise.
1
cos
(4)
ifn n n + n .
n+1 + n+1
2
2n+1
n =
sin
n + n
2 2n
0
f n t d = f n
(7)
and the approximation coefficients are represented byWf 0, t . The reconstructed signal is
DOI: 10.9790/4200-05133644
www.iosrjournals.org
38 | Page
IV.
The color images basically consists the basic colors like red, blue and green .To decompose a color
image using EWT must apply the algorithm to each color Image denoted by f and then reconstruct it using the
algorithm. The steps for color image decomposition and reconstruction illustrate as
i) Take the 1D FFT of each rows i of;f(i, ); and compute the mean row spectrum magnitude
1
N row
:Frows = N
i=0 f i, .
row
ii) Take the 1D FFT of each column j of;f(, j);and compute the mean column spectrum magnitude:
1
N column
Fcolumn = N
f , j .
i=0
column
iii) Perform the boundaries detection on Frows and build the corresponding filter bank{1row , row
n
NR
n=1 }
iv) Perform the boundaries detection on Fcolumn and build the corresponding filter bank{1column , column
n
v)Filter f along the rows with {1row , row
n
NR
n=1 }
Nc
n=1
vi) Filter each previous output image along the columns with{1column , column
n
end(NR + 1)(Nc + 1) subband images.
Nc
n=1
vii) Apply the above steps for each color component images to get the decomposed image for each color bands.
viii) The inverse transform can be obtained by applying the inverse 1D EWT with respect to the columns first
and then with respect to the rows for each color band image.
A grayscale digital image is an image in which the value of each pixel is a single sample, it carries
only information about intensity. The intensity of a pixel is expressed within a given range between a minimum
and a maximum value of the pixels.
The steps for grayscale image decomposition and reconstruction illustrate as
i) Take the 1D FFT of each rows i of;f(i, );and compute the mean row spectrum magnitude:
1
N row
Frows = N
i=0 f i, .
row
ii) Take the 1D FFT of each column j of;f(, j);and compute the mean column spectrum magnitude:
1
N column
Fcolumn = N
f , j .
i=0
column
iii) Perform the boundaries detection on Frows and build the corresponding filter bank{1row , row
n
DOI: 10.9790/4200-05133644
www.iosrjournals.org
NR
n=1 }.
39 | Page
NR
n=1 }
Nc
n=1
}.
vi) Filter each previous output image along the columns with{1column , column
n
end(NR + 1)(Nc + 1) subband images.
Nc
n=1
vii) The inverse transform can be obtained by applying the inverse 1D EWT with respect to the columns first
and then with respect to the rows.
V.
The analysis of the proposed method is tested on three different images from standard database[19], [20],
i) The grayscale image boat.
ii) The color image baboon.
iii) The medical image Liver.
For the testing of our algorithm, the selection of the maximum number of band per direction is two.
The resulting subbands of color, grayscale and medical image are shown in Fig1.2,Fig2.2, Fig3.2and the
corresponding empirical tilling of the Fourier domain in Fig.1.3,Fig.2.3andFig.3.3 respectively.
i) Grayscale Images: The proposed Algorithm is tested on the boat image in Fig 1.1.The decomposition of the
original image using EWT consists of number of subbands in each rows and column. Each subband is in the
form of matrix of size of 13x17.
www.iosrjournals.org
40 | Page
DOI: 10.9790/4200-05133644
www.iosrjournals.org
41 | Page
Fig.2.3 (a) Empirical tilling of the Fourier domain for red image component.(b) Empirical tilling of the Fourier
domain for green image component. (c) Empirical tilling of the Fourier domain for blue image component
iii) Medical Image: Proposed Algorithm is tested on the color image of medical image of Liver shown in
Fig 3.1.The decomposition of the original image using EWT consists of number of subbands in each rows and
column for each color band i.e., red, green, blue respectively.
Fig.3.1OriginalImage of Liver
DOI: 10.9790/4200-05133644
www.iosrjournals.org
42 | Page
Fig.3.3 (a) Empirical tilling of the Fourier domain for red image component. (b) Empirical tilling of the Fourier
domain for green image component. (c) Empirical tilling of the Fourier domain for blue image component.
The subband of red is in the form of matrix of size of 7x5. The size of the matrix of subbands is depend
upon the number of modes used for the analysis of an red band image which are automatic detect the modes and
in each subbands consists of a images of same size as original image.The above process is apply for each color
bands and get the decomposition of the color image is formed. The EWT components of image by selecting
number of rows and columns are four.it is clear that the cell containing each filtered output subband ewtc {1, 1}
{1, 1} i.e., empirical wavelet transform first coefficient is the lowpass subband and the other coefficients are
the bandpass filtered images subbands is denoted by ewtc{r} {c} where r corresponds to the scales along the
rows and c to the scales along the columns shown in Fig3.2.
VI.
Conclusion
In this paper, we proposed a 2D extension based on the tensor product idea on color and grayscale
images. In this case, the 2D EWT provides an adaptive tilling of the 2D Fourier plane. The Idea of color image
decomposition using EWT is very important for the future study. The future work is to compress and denoise
the color and grayscale images using EWT.
References
[1].
P. Auscher and G.L. Weiss and M.V. Wickerhauser, Local Sine and Cosine Bases of Coifman and Meyer and the Construction of
Smooth Wavelets,WaveletSA Tutorial in Theory and Applications, Academic Press, pp. 237256, 1992.
DOI: 10.9790/4200-05133644
www.iosrjournals.org
43 | Page
[10].
[11].
[12].
[13].
[14].
[15].
[16].
[17].
[18].
[19].
[20].
R.R. Coifmanand M.V. Wickerhauser, Entropy Based Algorithms for Best Basis Selection, IEEE Transactions on Information
Theory, vol. 32, pp. 712718, 1992.
I. Daubechies, Ten Lectures on Wavelets, Society for Industrial and Applied Mathematics, CbmsNsf Regional Conference Series in
Applied Mathematics, 1992.
I. Daubechies and J. Lu and H-T. Wu,Synchrosqueezed wavelet transforms: An empirical Mode decomposition-like tool, Journal of
Applied and Computational Harmonic Analysis, vol. 30, No.2,pp. 243261, 2011.
P. Flandrin and P. Goncalves, Empirical mode decompositions as datadriven wavelet-like expansions, International Journal of
Wavelets, Multiresolution and Information Processing, vol. 2, No. 4, pp. 477496, 2004.
P. Flandrin and G. Rilling and P. Goncalves, Empirical mode decomposition as a filter bank, IEEE Signal Processing Letters, vol.
11, No. 2, pp.112114, 2004.
P. Flandrin and P. Goncalves and G. Rilling, EMD Equivalent Filter Banks, from Interpretation to Applications, Chapter 3 in
Hilbert-Huang Transform and Its Applications, pp. 5774, World Scientific, 2005.
T.Y. Hou and Z. Shi, Adaptive data analysis via sparse time-frequency representation, Advances in Adaptive Data Analysis, vol. 2,
pp. 128, 2011.
N.E. Huang and Z. Shen and S.R. Long and M.C. Wu and H.H. Shih and Q. Zheng and N-C. Yen and C.C. Tung and H.H. Liu,
The empirical mode decomposition and the Hilbert spectrum for nonlinear and nonstationary time series analysis, Proc. Royal
Society London A., vol. 454, pp. 903995, 1998.
S. Jaffard and Y. Meyer and R.D. Ryan, Wavelets: Tools for Science and Technology, SIAM, 2001.
F.W. King, Hilbert Transforms, Cambridge, University Press, Encyclopedia of Mathematics and its Applications, 2009.
S. Mallat, A Wavelet Tour of Signal Processing A sparse way, Third Edition Elsevier,Academic Press, 2009.
H.S. Malvar, Lapped transforms for efficient transform/subband coding, IEEE Transactions on Acoustics, Speech and Signal
Processing, vol. 38, pp. 969978, 1990.
Y. Meyer, Wavelets, Vibrations and Scalings, American Mathematical Society, Centre de RecherchesMathematiques, 1997.
F.G. Meyer and R.R. Coifman, Brushlets: A tool for directional image analysis and image compression, Journal of Applied and
Computational Harmonic Analysis, vol. 4, pp. 147187,1997.
G. Thakur and E. Brevdo and N.S. Fukar and H-T. Wu, The Synchrosqueezing algorithm for time-varying spectral analysis:
robustness properties and new paleoclimate applications, submitted, 2012, available at http://arxiv.org/pdf/1105.0010.pdf.
M.E. Torres and M.A. Colominas and G. Schlotthauer and P. Flandrin, A complete ensemble empirical mode decomposition with
adaptive noise, IEEE International Conference on Acoustic,speechand Signal Processing (ICASSP),Pragues, pp. 41444147, May
2011.
J. Gilles, Empirical wavelet transform, IEEE Transactions in Signal Processing, Vol. 61, No. 16,pp. 3999-4010, 2013.
http://www.imageprocessingplace.com/root files_V3/image_databases.html
http://decsai.ugr.es/~javier/denoise/testimages.
Author Profile
Atul Kumar Verma received B.Tech degree in Electronics & Communication Engineering
from UPTU, Lucknow in 2012. Currently he is pursing M.Tech in ECE fromDr. B. R.
Ambedkar National Institute of Technology (NIT) Jalandhar. His research interests include
speech signal processing, and image processing.
DOI: 10.9790/4200-05133644
www.iosrjournals.org
44 | Page