Sei sulla pagina 1di 4

2010 International Conference on Measuring Technology and Mechatronics Automation

A New Threshold Function for Signal Denoising Based on

Wavelet Transform

Yating Lin, Jianli Cai,


Department of Automation, Xiamen University, Department of Automation, Xiamen University
Xiamen, 361005, China Xiamen, 361005, China
e-mail:linyating008@163.com, e-mail:cjl701@sina.com,

1
Abstract -A novel thresholding function is presented based also obtained the prodigious development for several
on the wavelet thresholding denoising method put forward by years,which symbolizes the emergence of a kind of new
D. L. Donoho and I. M.Johnstone.This new thresholding signal filtering method.
function has many advantages over soft- and hard- threshold The method of wavelet threshold value filtering is a
function.It is simple in expression and as continuous as the kind of the simplest realization, the smallest calculation, so
soft-thresholding function.It also overcomes the shortcoming it has been widely applied. But its threshold value's
that there is an invariable dispersion between the estimated selection is quite difficult, although Donoho has proved
wavelet coefficients and the decomposed wavelet coefficients of theoretically and has found the general threshold value, the
the soft- thresholding method. At the same time , the new results are unsatisfactory in the practical application.And
thresholding function is more elastic than the soft- and hard- that there are two aspects in the improvement of threshold
thresholding,function.MATLAB Simulation results indicate value method: First, it proposes the new threshold value
that the denoising method adopting the new thresholding method based on wavelet transformation and it will be
function suppresses the Pseudo-Gibbs phenomena near the applied to the signal filtering and the image compression;
singularities of the signal effectively , and the numerical Second, it mainly research on the parameter selection
results also show the new method gives better SNR gains than algorithm of semi-soft threshold function in the wavelet
DJps hard- and soft-thresholding methods. threshold value function. Combined with the grey incidence
Key Words -Wavelet transform; MATLAB; threshold- degree theory and fuzzy theory, a new parameters
function;SNR algorithm will be proposed, and it can reduce the
complexity of algorithm, finally, it will be applied to the
I .INTRODUCTION different signals, and the results are satisfactory.The result
The appearance of wavelet has aroused the widespread for that further demonst ratesthe superiority and
interest in the field of mathematics, its theory and effectiveness of this method.
application obtained the unprecedented development, and it II.DENOISINGBASED ON WAVELET
has become a main research direction of the signal and TRANSFORM
image processing. The wavelet theory and application as an A. WaveletThresholding Algorithm
important branch, the wavelet filter's theory and application In the traditional Fourier-based signal processing, the
out-of-band noise can be removed by applying a linear
1
Jianli Cai is a corresponding author

978-0-7695-3962-1/10 $26.00 2010 IEEE 200


DOI 10.1109/ICMTMA.2010.347
time-invariant filtering approach; however, it cannot be The key steps of signal denoising based on wavelet
removed from the portions where it overlaps the signal transform are the threshold and shrinkage function
spectrum. The denoising technique used in wavelet analysis selection.If the thresholding is too low, noise will be
is based on the assumption that the amplitude rather than remained and if the thresholding is too high, useful signal
the location of the spectrum of the signal is different from will be lost.The universal threshold, which is proposed by
the noise. The localising property of the wavelet aids D. L.Donoho[4], is the most popular in all kinds of
thresholding and shrinking the wavelet coefficients, which threshold. It can be expressed as follows:
helps to separate the signal from noise.
= 2 ln( N )
Denoising by wavelet is quite different from
traditionalfiltering approaches because it is nonlinear
Median( c )
owing to a thresholding step [1]. =
Suppose that the clean signal s(k) of length N is
0.6745
corrupted by white noise e(k), e(k)with standard normal Where N is the sampling length of noisy signal, is
distribution N(0, 2 ) ,to give the noisy signal f(k) which is the standard deviation of zero-mean additive white
given by Donoho in [2] , shown as (1): Gaussian noise(WGN) estimated by Donoho and Johnston.
c is the detail coefficients of wavelet transform.
f(k)=s(k)+e(k) 1 k N (1) C Thresholding Function
Decomposition of the signal f(k) is performed by After several decades years development the
levels J. If the noise energy is obviously less than signal expressions of shrinkage function have lots of kinds. Such
energy, the corresponding noise wavelet coefficients will as soft shrinkage function, hard shrinkage function [5], firm
obviously smaller than the signal wavelet coefficients, shrinkage function[6], hyper-trim shrinkage function [7],
select a suitable thresholding to set the wavelet coefficients multi-parameter best basis thresholding shrinkage function
which below the thresholding to zero and remain or shrink [8], Yasser shrinkage function [9]. All these thresholds and
the wavelet coefficients which higher than the thresholding. shrinkage functions promoted the application of wavelets in
The Donoho s wavelet denoising algorithm (i.e., the signal denoising extremely.the soft- and hard-thresholding
SURE shrinkage) is summarized as follows [3]: schemes are defined by [3]:
I. Transform f (k) to the wavelet domain and obtain Hard-thresholding:

wavelet coefficients wj,k , with scales j =1 ~ J , and time w j ,k , w j ,k


w j,k =
indices k =1~ N ; 0, w j ,k <
2) Apply soft- or hard- thresholding to wj,k using (2)
Softthresholding:
threshold = 2 ln( N ) and obtain processed wavelet
sign( w j ,k )i( w j , k ), w j ,k
coefficients wj,k ; w j,k =
0, w j ,k <

3) Perform the inverse wavelet transform on , wj,k to (3)

obtain the denoised signal f k .


Where sgn(*) is a sign function :
B. Value of Thresholding

201
1, n > 0 ( w j ,k )2
sgn(n) =
1, n < 0 = exp
k

k is positive number .
W
From (4),we know when wj,k , 0 ,

then wj,k = wj,k is a asymptote of the improved


function wj,k wj,k
,and overcomes the soft shrinkage
function's disadvantages;
(Hard thresholding )
When
W w j ,k 1
, , wj,k 0 ,

then wj,k is continuous at wj,k = ,and wj,k


makes the signal more smooth than hard thredhoding
function.
EXPREIMENT AND ANALYSIS :
(Soft thresholding) In order to embody the advantages of the new wavelet
The derivation of standard soft shrinkage function is threshold denoising method more fully ,we select the
not continuous. The soft shrinkage estimates tend to have a HeaviSine signal from MATLAB toolbox which is popular
bigger bias, due to the shrinkage of large coefficients. Due library signal utilized in ,The length of test signal was,
to the discontinuities of the shrinkage function, hard respectively, N = 1024 and . The cubic spline wavelet was
shrinkage estimates tend to have a bigger variance and may chosen and the largest decomposition scale was J = 3. We
exhibit Gibbs phenomena in the neighborhood of used the median value method to estimate the standard
discontinuities [10].In addition, its performance often deviation of the noise
deteriorates substantially as the signal-to-noise ratio (SNR)
Median( c )
of the input signal decreases. =
0.6745 ,
To overcome the drawbacks of hard and soft threshold
function, a new threshold function was introduced to show
= 2 ln( N )
the advantage of non -negative dead zone thresholding
The test data is shown in table 1,and the Havisine
algorithm ,where wj,k is modified as:
signal denoising contrast are shown in Figure 1.

1 2 sgn( w ) w + w , w
j ,k j ,k j ,k j ,k
=
w j ,k 0, w j ,k <

(4)
Where

202
Table 1. SNR results in dB for different denoising algorithms SNR=11.8348db;(c)hardthresholdmethod;(d)softthresholdme
thod;(e)compromise method;(f)~(h)improved threshold
method with m=4,m=6,m=8.

from figure 1,we can find that the proposed


method obtains superior visual quality.
Conclusions
Although the standard wavelet threshold denoising
method is simple and practical, it does not work well for
signals with a low SNR; besides, it uses coordinate-wise
processing without taking the structural information of the
wavelet coefficients into account. Our new denoising
From Table 1,it can be observed that the new algorithm method based on wavelet transform and sparse
achieves higher SNR gains than the hard-thersholding,soft- representation overcomes these problems. It is suitable for
thresholding and compromise- thersholding methods. many signals including those with a low SNR.
References
[1] Hu Chuihai, Liu Xuefei. Speech recognition for noisy conditions
based on discrete wavelet transform and parallel modelcombination,
icemi2005,Volume: 1, 408-411
[2] Donoho,D.L., Johnston,I.M.. Denoising by soft-
thresholding.IEEE Trans.Inf. Theoryl, 1995Vol.41,No. 3, pp.77-79.
[3] D.L. Donoho, De-noising by Soft-thresholding, IEEETrans.
Inform. Theory, Vol.41, No.3, 1995, pp.613~627
[4] Donoho, D.L., Johnston, I.M.,Ideal spatial adaptive via wavelet
shrinkage, Biometrika, vol.81, 1994, pp425455.
[5] Donoho, D.L., De-noising by soft-thresholding, IEEE Trans.
Inf.Theory vol.41 (3), 1995, pp613627.
[6] A.G. Bruce, H.-Y. Gao, WaveShrink with firm shrinkage,
Statistica Sinica Research Report, No. 7, 1997, pp. 855874.
[7] S. Poornachandra, N. Kumaravel, Hyper-trim shrinkage for
denoising of ECG signal, Digital Signal Processing vol.15, 2005, pp317
327.
[8] Adelino R., Ferreira da Silva, Wavelet denoising with
evolutionary algorithms, Digital Signal Processing, vol. 15, 2005, pp382
399.
[9] Yasser Ghanbari, Mohammad Reza Karami-Mollaei, A new
approach for speech enhancement based on the adaptive thresholding of
the wavelet packets, Speech Communication, vol. 48, 2006, pp927940.
[10] S. E. Kelly, Gibbs phenomenon for wavelets, Applied and
Computational Harmonic Analysis, Vol.3, No.1, 1996,pp.7281
Figure 1:Simulation example using Heavisine signal with different
denoising mehtods. (a)original signal ;(b)noisy signal with

203

Potrebbero piacerti anche