You are on page 1of 24

# Application of Wavelets in Signal Processing

## Special Emphasis : De-noising of 1 & 2 Dimensional Signals

Outline of Presentation
Fourier Transform & STFT Wavelets History DWT vs. STFT Continuous & Discrete Wavelets Applications Conclusion

Signal Analysis
Most of the signals encountered in practice are expressed in time ( space) domain.
Many features of a signal are not explicit in time domain

## Analysis of these signals are carried in frequency domain

Using Fourier Transform

## The Fourier Transform

A Mathematical prism
breaks a function into its constituent frequencies as a prism breaks up light into colours

It transforms a function which depends on time ( or space) into a new function which depends on frequency. A function and its Fourier transform are two faces of the same coin.
The function displays the time ( or space) information and hides information about frequencies The FT displays information about frequencies and hides the information about time (or space) in phases.

## Important Features of Fourier Transform

Tells how much of each frequency a signal contains, but is secretive about when these frequencies were emitted

Note: FT hides information about time, but does not destroy time information; otherwise we could not reconstruct the signal from the transform. Time information is buried deep within the phases.

## Important Features of Fourier Transform

Computing time information from phases with enough precision is impossible. Major drawback:
Information about one instant of a signal is dispersed throughout all the frequencies of the entire transform A local characteristic of the signal becomes a global characteristic of the transform.
Example: a discontinuity is represented by a superposition of all possible frequencies.

## Searching for Lost Time : Short Time Fourier Transform

Used for analyzing a signal both in time and frequency
Divides the signal into different (fixed size) time segments Study the frequencies of a signal segment by segment. When one segment of the signal has been analyzed, slide the window along the signal to analyze another segment.

## Painful Compromises in STFT

Smaller the widow size, better to locate sudden changes, such as peak or discontinuities But blinder to the lower frequency components of the signal Bigger the window size, Worse at localization in time; difficult to locate sudden changes or discontinuities See more of lower frequencies.

## Wavelet : A Mathematical Microscope

Automatically adapt to the different components of a signal ( Multiresolution)
uses small window to look at brief high frequency components Uses large window to look at long-lived low frequency components

## This procedure is called multiresolution;

signal is studied at coarse resolution to get an overall picture studied at higher and higher resolutions to see increasingly fine details.

History
First mention in appendix of the thesis of A. Haar (1909) Cochlear transform, Zweig (1975) Continuous Wavelet Transform (CWT), Grossman and Morlet (1982) Geophysics Discrete Wavelet Transform (DWT), Strmberg (1983) Daubechies' orthogonal wavelets with compact support (1988) Mallat's mult-iresolution framework (1989)

## DWT vs. STFT Multi-Resolution

STFT DWT

Haar Wavelet
Change in scale and time

## Common Features of Different Types of Wavelet

Wavelet: a wave-like oscillation with an amplitude that starts out at zero, increases, and then decreases back to zero Wavelet-theory: wavelet with zero-integral and finite energy Wavelet transform: projection on the sub-spaces associated with each wavelet

## Ex:-3 Modified Haar wavelet

M orlet wavelet 1 0.8 0.8 0.6 0.4 0.2 0.4 0 -0.2 -0.4 -0.6 -0.2 -0.8 -1 -4 -0.4 -3 -2 -1 0 1 2 3 4 0 5 10 15 20 25 30 35 40 0.2 0.6 1

## Discrete Wavelets and Filter Banks

The two main conditions for obtaining a perfect reconstruction and an alias-free filter bank respectively are :

10 0

-10

-20 G ain, dB

-30 -40

-50

-60

-70 0 0. 1 0. 2 0. 3 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 1

[/ T

## Example of Haar Transform for 1- Dimensional signal

Approximates Coefficients

## [5/ 2,3/ 2| 3/ 2,3/ 2]

Details Coefficients

4v / 2( v / 2)!7/ 2 1 3 1

0 v1 /

2 !0

## Designing New Wavelet

To generate new wavelet from a given biorthogonal quadruplet, a finite sequences of primal or dual elementary lifting steps (ELS) should be applied. For a wavelet has four filters ,  the primal ELS can be defined as:

Where
 The dual ELS can be defined as:

## Principle of Wavelet-based De-noising

Step-1: Obtain wavelet coefficients both (approximates and details )
Approximates are the lowpass and downsampled output coefficients Details are the highpass and downsampled output coefficients

Step-2: Determine the threshold type ( e.g. soft and hard threshold) and value of the threshold Step-3 :Eliminate details coefficients Step-4: Retain only those detail coefficients which satisfy the threshold conditions and approximates

## Wavelet-based De-noising Procedure

Three main steps
 Compute wavelet transform of the signal (1D) or image (2D) by passing them through Lowpass and Highpass filters  Threshold the coefficients using soft threshold and SURE to determine threshold value
th e o rig in a l s ig n a l 1 0.8 0.6 0.4 0.2 0 -0 . 2 -0 . 4 -0 . 6 -0 . 8 -1 H a rd th re s h o ld o f th e s ig n a l 1 0.8 0.6 0.4 0.2 0 -0 . 2 -0 . 4 -0 . 6 -0 . 8 -1 s o ft th re s h o ld o f th e s ig n a l 0.5 0.4 0.3 0.2 0.1 0 -0 . 1 -0 . 2 -0 . 3 -0 . 4 -0 . 5

50

100

50

100

50

100

 Compute the inverse wavelet transform or performing reconstruction of the original signal by passing the threshold coefficients through the synthesis filters to obtain the original approximation coefficients.

 It s possible to remove the noise with little loss of details.  The idea of wavelet de-noising is based on the assumption that the amplitude, rather than the location, of the spectra of the signal to be as different as possible for that of the noise.

## Example1: -Duffing s oscillator

x 10

-3

O rig in a l s ig n a l 10

x 10

-3

N o is y s ig n a l

5 0 0

-5

0 x 10
-3

200

400

600

-5

200

400

600

D e -N o is e d s ig n a

-5

200

400

600

Example-2:-Biomedical Data
D e ta il c D 1 2 2 D e ta il c D 2

## C orrelation of R esudual 1 0 -1 -1000 400 200 0

-2 5

200 400 D e ta il c D 3

600

-2 5

100 200 D e ta il c D 4

300

-800

-600

-400

-200

200

400

600

800

1000

H istogram O f R esudual

-5

50

100

150

-5

20

40

60

80

10

## C um ulative H istogram O f R esudual 2000 1000 0

10

20

30

40

50

60

70

80

10

t he 2 0

o r ig in a l

s ig n a l

1 5

1 0

2 0 0 t he

4 0 0 d e - n o is e d o r ig in a l

6 0 0 s ig n a l u s in g

8 0 0 wa v e l e t

1 0 0 0 ba s e d

1 2 0 0

2 0

1 5

1 0

2 0 0

4 0 0

6 0 0

8 0 0

1 0 0 0

1 2 0 0

## Example 3:- De-noising Images

O rig in a l Im a g e

N o is y Im a g e

## 50 100 150 200 250 50 100 150 P S NR= 10 200 250

D e -N o is e d Im a g e

D e -N o is e d Im a g e u s in g A ve ra g e F ilte r 5 X5

## 50 100 150 200 250 50 100 150 P S NR= 20 200 250

Original, Noisy (10db), Wavelet-based De-noised (27db), FT-based 5X5 Averaging Filtered (20db)

## Example 4:- De-noising of Seismic Data

1) For analysing seismic traces, both oscillations and the time they occur are important.
0 0 0

2) WT can be used to zoom in on the short bursts and zoom out to detect long oscillation
A)The Noisy S ynthetic S eism ogram 0 B)De-noised S eism ic by M W 0 C)De-noised seism ic M W using PCA 0

50

50

50

50

50

50

100

100

100

150

150

150

150

150

150

200

200

200

200

200

200

250 0 5 10

250 0 5 10

250 0 5 10

250

250

250

10

10

10