IOSR Journal of VLSI and Signal Processing (IOSR-JVSP)

Volume 5, Issue 1, Ver. III (Jan - Feb. 2015), PP 36-44

e-ISSN: 2319 4200, p-ISSN No. : 2319 4197
www.iosrjournals.org

Image Decomposition Using Adaptive Empirical Wavelet

Transform
Atul Kumar Verma1, Dharmendra Gurve2
1
2

(ECE Department, Dr. B. R. Ambedkar National Institute of Technology Jalandhar, India)

(ECE Department, Dr. B. R. Ambedkar National Institute of Technology Jalandhar, India)

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

An Adaptive method to analyse a signal is of

extraordinary enthusiasm to discover spare
representations in the context of compressive sensing. "Rigid" strategies, in the same way as the Fourier or
wavelets transforms, compare to the utilization of some frame designed independently of the processed signal.
The point of adaptive strategies is to build such a frame straight forwardly focused on the information contained
in the signal. A well-known approach to build adaptive representations is the basis pursuit methodology which
is utilized as a part of the wavelet packets transform. Even though the wavelet packets have shown interesting
results for practical applications, they still are based on a prescribed subdivision scheme. A completely different
approach to build an adaptive representation is the algorithm called Empirical Mode Decomposition (EMD)
proposed by Huang et al. [9]. The purpose of this method is to detect the principal modes which represent the
signal; a mode corresponds to a signal which has a compactly supported Fourier spectrum. This method has
gained a lot of interest in signal analysis this last decade, mainly because it is able to separate stationary and
non-stationary components from a signal. However, the main issue of the EMD approach is its lack of
mathematical theory.The main drawback of the EMD technique is that it is a nonlinear algorithmic procedure
and the obtained representation is implementation dependent.Moreover due to the nonlinear aspects, no
theoretical background supports this method. Indeed, it is an algorithmic approach and, due to its nonlinearity, is
difficult to model. Nevertheless, some experiments [5][7] show that EMD behaves like an adaptive filter bank.
Some recent works attempt to model EMD in a variational framework.
In [4], the authors proposed to model a mode as an amplitude modulated-frequency modulated signal
and then use the properties of such signals to build a functional to represent the entire signal. To minimize these
functional we retrieve the image in different modes. Another proposed variational approach is the work of Hou
et al. [8] where the authors also use the AM-FM formalism. They propose to minimize a functional which is
built on some regularity assumptions about the different components and uses higher-order total variation priors.
We will show that it is possible to adapt the wavelet formalism by considering distinct Fourier supports and then
build a set of functions which form an orthonormal basis.In [18], a straightforward tensor extension was
proposed to do some image analysis. In this paper, we propose to investigate the possibility of generalizing this
empirical approach to different existing 2D wavelet transforms. Based on this construction, we propose an
empirical wavelet transform and its inverse to analyse an image. The remainder of the paper is organized as
follows. Section II has two distinct subsections: in II-A, we recall the principle of the EMD algorithm; while in
II-B, we recall some wavelet formalism which will be useful in our own construction and we discuss some of
the existing adaptive wavelet methods. In section III has three distinct subsections: in III-A, we build the
proposed empirical wavelets technique; and in III-B, we explain the empirical wavelet transform and its inverse
for signals. In III-C, 2-D EWT (empirical wavelet transform) for images is described. The proposed method for
grayscale and color image decomposition described in Section IV. Section V deals with the results tested on
different images. Finally, we conclude and give the future scope for our method in section VI.
DOI: 10.9790/4200-05133644

www.iosrjournals.org

36 | Page

Image Decomposition Using Adaptive Empirical Wavelet Transform

II.

Existing Approaches

A. Empirical Mode Decomposition

Empirical mode decomposition method is one of the existing method, the key part of the method is it
can decomposed any complicated data set into a finite and often small number of `intrinsic mode functions' that
admit well-behaved Hilbert transforms. This decomposition technique is highly efficient andadaptive. Since, the
decomposition is based on the local characteristic time scale of the data; it is applicable to nonlinear and nonstationary processes. In 1998, Huang et al. [9] proposed an original method called Empirical Mode
Decomposition (EMD) to decompose a signal into specific modes.
In this paper, as we will use the Fourier formalism in section III, we adopt the description used in [4]
which is slightly different from the original used in [9].Let us assume analysed signal is f(t) then EMD aims to
decompose a signal as a finite sum of N + 1 Intrinsic Mode Functions (IMF) fs (t) such that
M

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

Image Decomposition Using Adaptive Empirical Wavelet Transform

the temporal segmentation is a difficult task to perform efficiently. In [15], the authors propose a method, called
the brushlets, which aims to build an adaptive filter bank directly in the Fourier domain. Basically, it uses the
idea of the Malvar- Wilson wavelets but segments the Fourier spectrum of the signal, instead of the signal itself.
Conceptually the ideas in this work are really interesting; however the proposed construction is quite
complicated and is also based on prescribed subdivisions. The last work we want to mention is a recent work of
Daubechies et al. [4]entitled synchrosqueezed wavelets. This approach combines a classic wavelet analysis
and a reallocation method of the time-frequency plane information. Wavelet analysis, on the other hand, gives
the information about both time and frequency, although both parameters cannot be exactly determined
simultaneous sly due to the Heisenberg uncertainty relation.
This algorithm permits to obtain a more accurate time-frequency representation and consecutively it is
possible to extract specific modes by choosing the appropriate information to keep. All the above methods use
either a prescribed scale subdivision schemes or a smart utilization of the output of a classic wavelet analysis.
As far as we know, no work exists which aims to build a full adaptive wavelet transform. The remaining of the
paper will address such construction.

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 .

ifn + n n+1 n+1 .

n+1 + n+1
2
2n+1

n =
sin

n + n
2 2n
0

ifn+1 n+1 n+1 + n+1

(5)
ifn n n + n
otherwise

The function x is an arbitrary C k ([0,1])function such that

0
ifx 0
x =
(6)
1
ifx 1
and x + 1 x = 1
x 0,1 .
B. Empirical wavelet transform
In this section, the Empirical Wavelet Transform (EWT) [18]Wf (n, t) is defined, in the same way as
for the classic wavelet transform. The detail coefficients are given by the inner products with the empirical
wavelets:
Wf n, t = f, n =

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

Image Decomposition Using Adaptive Empirical Wavelet Transform

f t = Wf 0, t 1 t + Nn=1 Wf n, t n t (8)
Hence the approximation and details coefficients of EWT is defined a
f0 t = Wf 0, t 1 t
fk t = Wf k, t k t
(9)
C. Empirical wavelet transform for images
In this section, we address the opportunity to extend the previous EWT to 2D signals (images). The
basic idea to build such 2D-EWT is to use a tensor product as for classic wavelets, which means process the
rows and then the columns of the input image by using the 1D-EWT defined previously. If this idea seems
obvious to adopt, it turns out that two main issues arise: there is no guarantee that two different rows (or
columns) will have the same number of Fourier supports; even if we have the same number of bands for each
rows (or columns), the supports corresponding to the same band number can be very far away from each other.
The direct consequence is that in the output image corresponding to a specific band, very different frequency
modes can be mixed providing a strange representation. In order to avoid these issues, a solution consists to
use the same filters for each rows (or columns). A simple way to build such solution is to consider a mean
spectrum for the rows (or the columns), then perform the detection of the Fourier supports based on this mean
spectrum and finally use the same filters for all rows (columns).

IV.

Proposed Algorithm For Image Decomposition

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

which provides NR + 1 output images.

vi) Filter each previous output image along the columns with{1column , column
n
end(NR + 1)(Nc + 1) subband images.

Nc

n=1

}, this provides at the

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

Image Decomposition Using Adaptive Empirical Wavelet Transform

iv) Perform the boundaries detection on Fcolumn and build the corresponding filter bank{1column , column
n
v)Filterf along the rows with{1row , row
n

NR
n=1 }

Nc
n=1

}.

which provides NR + 1 output images.

vi) Filter each previous output image along the columns with{1column , column
n
end(NR + 1)(Nc + 1) subband images.

Nc
n=1

}, this provides at the

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.

Results And Discussion

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.

Fig.1.1 Original Image of boat.

Fig.1.2EWT components of the Boat imageNR = 4; NC = 4.

DOI: 10.9790/4200-05133644

www.iosrjournals.org

40 | Page

Image Decomposition Using Adaptive Empirical Wavelet Transform

The size of the matrix of subbands is depend upon the number of modes used for the analysis of an image which
are automatic and each subband consists of images of same size as original image. The EWT components of
image are obtained by selecting number of rows and columns has 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 inFig1.2.The boundaries
of the resulting output partitioning image with the empirical tilling of the Fourier domain for the image of boat
shown in Fig.1.3.

Fig.1.3Empirical tilling of the Fourier domain for boat image.

ii) Color Image: The Proposed Algorithm is tested on the color image of baboon shown in Fig2.1.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. The subband of red is in the form of matrix of size of 9x5.
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 each subbands consists of a images of same size as original
image. The above process is apply for each color bands and the decomposition of the color image is formed. The
EWT components of image obtained by selecting number of rows and columns has 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 image subbands is denoted by ewtc{r}
{c} where r corresponds to the scales along the rows and c to the scales along the columnsshown inFig2.2.

Fig.2.1Original Image of baboon.

DOI: 10.9790/4200-05133644

www.iosrjournals.org

41 | Page

Image Decomposition Using Adaptive Empirical Wavelet Transform

Fig.2.2EWT components of the Baboon image NR = 4;NC = 4.

The boundaries of the resulting output partitioning image for each color bands with the empirical tilling of the
Fourier domain respectively shown in Fig2.3.

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

Image Decomposition Using Adaptive Empirical Wavelet Transform

Fig.3.2EWT components of the medical imageNR = 4;NC = 4.

The boundaries of the resulting output partitioning image for each color bands with the empirical tilling of the
Fourier domain respectively shown in Fig3.3.

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

Image Decomposition Using Adaptive Empirical Wavelet Transform

[2].
[3].
[4].
[5].
[6].
[7].
[8].
[9].

[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.

DharmendraGurve received B.E. degree in Electronics & Communication Engineering from

RGPV, Bhopal 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, cellular communication and biomedical signal processing.

DOI: 10.9790/4200-05133644

www.iosrjournals.org

44 | Page