Wavelets &

Applications

N. Malmurugan
[email protected]
 Fourier analysis

Decompose image/signal into sinusoids

+
F (w) =
f ( x ) [ cos( 2 w x ) + j sin( 2 w x ) dx ]
 Fourier Transform

F
Fourier
Transform
Frequency
Time
 Case -1 Stationary signal

1 250

0.8
0.6 200

0.4
0.2 150

0
-0.2 100

-0.4
-0.6 50
-0.8
-1 0
0 0.2 0.4 0.6 0.8 1 0 20 40 60 80 100

time frequency
 Fourier Transform (FT)

Time-domain Frequency-domain
 Fourier transform and non-
stationary signals

The Fourier transform does not provide information concerning

the time interval(s) where a given frequency component exist.

Not a problem with stationary signals

(The frequency components of the signal then exist at all times )

Not suited for non-stationary signals
Neither suited for signals with discontinuities and sharp peaks
(like EKG or EEG signals)
 1

0.8

0.6

0.4

0.2

-0.2

Stationary -0.4

signal -0.6

-0.8

-1
0 100 200 300 400 500 600 700 800 900 1000

Non -
stationary
signal
The obtained Fourier spectra
 Case - 2 Signal varying with time

Ambiguity exists in the analysis

1 1.5
0.8
0.6
0.4 1
0.2
0
-0.2
-0.4 0.5
-0.6
-0.8
-1 0
0 50 100 150 200 250 300 350 400 0 50 100 150 200

time frequency

What about
instantaneous
frequency ?
 Problem with FT

FT contains only frequency information

No Time information is retained

Works fine for stationary signals

Non-stationary or changing signals cause problems

FT shows frequencies occurring at all times instead of
specific times
 Solution: STFT

How can we still use FT, but handle non-stationary

signals?

How can we include time?

Idea: Break up the signal into discrete windows

Each signal within a window is a stationary signal

Take FT over each part
 Short Time Fourier Transform ( STFT )

- Divide the signal into sections using windows

- Calculate the Fourier transform each section
- Display the results as a joint function of the frequency and time

-j2 ft
The STFT is defined as x ( t ) w( t ) e dt
X(f, ) =

where x(t) is the signal to be analysed , w(t) is the window function,

the process is illustrated in the next figure.

STFT maps a 1 -D time domain signal to a 2 -D plane (t , f )

Analysis depends on the type of window chosen
 Windowed Fourier transform

Sliding time window is used

Heisenberg uncertainty relations are playing!

One cannot know the exact time-frequency representation of a signal.
 STFT Contd..

( w)
STFT
X (t , f ) = [x(t ) w * (t t ' )] e j 2ft
dt
t
 Heisenburg Uncertainty Principle

Originally applied to the momentum and location

of moving signals

Can be applied to the time-frequency information
of a signal

Meaning

Cannot know what spectral components exist at what
instances in time
 Example of STFT

Perform it in the same non-stationary signal as
before
 Example of STFT (continued)

Gaussian Window with different amounts of support

 STFT with a = 0.01 (narrowest
window)

Well separated in
time

Peaks correspond
to a band of
frequencies
 STFT with a = 0.0001
(3rd widest window)

Time resolution is
becoming poorer

Frequency
resolution is
improving
 STFT with a = 0.00001
(widest window)

Poor time
resolution

Excellent
frequency
resolution
 Time-Frequency Tiles: STFT

Single window used for all

frequencies

Resolution analysis is the
same at all locations
 Conclusions: STFT

Developed as an alternative to the FT

STFT is application dependant

Better time or frequency resolution?

Choice of window is crucial
 Wheres the problem?

How do you pick an appropriate window?

Too small = poor frequency resolution

Too large may result in violation of stationary
condition

Different resolutions at different frequencies?
Alternative basis functions

Anyone heard about

wavelets?
 What are wavelets?

Waves : Waves are oscillating functions of

time or space or both.

Wavelets : Wavelets are small waves.

A wavelet has oscillating wave-like characteristics.

Its energy is concentrated in time over relatively small
intervals.
A wavelet can look like..

0.1

0.08

0.06

0.04

0.02

-0.02

-0.04

-0.06
0 500 1000 1500 2000 2500
 The Continuous Wavelet Transform
(CWT)

CWT decomposes f(t) into a set of basis wavelet

functions:
( s, ) = f (t ) dt
*
s ,

Wavelets are generated from a mother wavelet (t )

1 t
s , =
s s
 Wavelet properties

Admissibility condition:
IfIfsatisfied
satisfiedthen
then
the
the waveletscan
wavelets can
2 be used
be used toto
( ) analyze
analyzeand

and
d < then
thenreconstruct
aasignal
reconstruct
signalwithout
without
loss of
loss of
information
information

Implication:

( ) =0 = 0 (t )dt = 0
Oscillatory
Oscillatory! !
Must
Mustbebeaawave
wave
Some Common Wavelets
 Wavelet Transform

Sine Wave Mother Wavelet (db10)

Signal Signal

Sine wave at different freq Translated and Dilated

Mother Wavelet
scale decreases frequency
increases

Local outlook
Multi resolution analysis
Analysis between scales
Time-scale analysis
Evaluating the Continuous Wavelet Transform

signal

step -1

step -2

*
*
step - N
 Each wavelet (t) is associated with a scaling function (t)

The scaling function provides the scaling characterestic to the

wavelet and the choice of the scaling function depends on the
wavelet.
1
1
(t) HAAR Wavelet
-1
a. wavelet function (t) = 1; 0 < t < 1/2

1 = -1; 1/2 < t < 1

(t)
(t) = 1; 0 < t < 1
1

b. scaling function

other wavelet families include Daubechies, Morlet , Coiflets , Symlets , Biorthogonal wavelets
Chirped signal
and CWT
 CWT Example

Non-stationary Signal
 CWT Example

Poor scale
resolution at
high scales

Interpreted as:
good frequency
resolution at
lower scales
Time-frequency Tiles: CWT

Image recaps
everything we have
gone through

Areas of the boxes are
also determined by
Heisenburgs
Uncertainty Principle
 Sunspot activity for 500 months
250

200

150
Activity

100

50

0
0 50 100 150 200 250 300 350 400 450 500
time
 250

CWT transform
200

150

100

50

0
0 50 100 150 200 250 300 350 400 450 500
 Problems with the CWT

Redundancy - The scaled wavelets need not

be an orthogonal basis

An infinite number of wavelets in the
transform

For most functions no analytical expressions
exist for the wavelet transform

Fast algorithms are needed to exploit the
power of the transform
 Discrete wavelets
The
Thewavelet
wavelet
itself
itselfisisnot
j ,k (t ) = ( s t k 0 )
j not
discrete
discrete butititisis
but
0
scaled
scaledand and
translated
translatedin in
discrete steps
discrete steps

Usual choices:
s0 = 2 and 0 = 1 (dyadic sampling)
 Orthogonal basis

Wavelets that are orthogonal to their own

dilations and translations demands special
choices of the mother wavelets
 Multiresolutional Analysis

Analyzing according to scale

Designed to give:

Good time resolution and poor frequency resolution at
high frequencies

Good frequency resolution and poor time resolution at
low frequencies
 How many scales are needed?

Compression in time stretches the spectrum and shifts it

upwards:
1
F { f (at )} = F
a a

Using dilated wavelets we can cover the signal spectrum

 The scaling function

An infinite number of wavelets is needed to cover

the spectrum down to zero

Solution: Use a scaling function as cork plug
 Iterated filter bank

Iteratively split the signal in two

a low-pass and a high-pass signal

The wavelet transform is equivalent with a

subband coding scheme
 Closer look at the filter scheme

Twice as many data as we started!

x0 x0 + x7 x0 x7
x lowpass x + x highpass x0 x1
1 0 1
x2 x1 + x2 x1 x2

x 1 x2 + x3 1 x2 x3
X = 3 l X = Hl X = h X = Hh X =
x4 2 x3 + x4 2 x3 x4

4x + x5 4x x5
x5
x6 x5 + x6 x5 x6

x6 + x7 x6 x7
x7

Reconstruction: X = lX +hX
 Downsampling

Can the signal still

be reconstructed ?

x0 + x7 x0 x7
x + x
1 1 x2 x1
X l = DH l X = 2 1 X h = DH h X =
2 x4 x3
2 x4 + x3

x + x x
6 x5
6 5
 Reconstruction
Upsampling:

x0 + x7 x0 x7
0 0

x2 + x1 x2 x1

1 0 1 0
UX l = UX =
2 x4 + x3 h
2 x4 x3

0 0
x6 + x5 x6 x5
x0 + x7 x7 x0
0 0
lowpass x + x highpass x x
2 1 1 2
x2 + x1 x2 x1

1 x4 + x3 1 x3 x4
l * UX l =
2 x4 + x3
h UX h =
2 x4 x3
S = l UX l + h UX h

6x + x5
x
5 6 x
x + x x x
6 5
6 5

x0 + x7 0
x x 7
 Quadrature mirror filters
Analysing filters:
1 0 0 0 1
0 0 0 1 0 0 0 0 0 0 1
1 1 0 0 0
0 0 0 1 1 0 0 0 0 0 0

0 1 1 0 0
0 0 0 0 1 1 0 0 0 0 0

1 0 0 1 1 0
0 0 0 1 0 0 1 1 0 0 0 0
Hl = Hh =
2 0 0 0 1 0
1 0 0 2 0 0 0 1 1 0 0 0

0 0 0 0 1
0 1 0 0 0 0 0 1 1 0 0
0 0 0 0 0
0 1 1 0 0 0 0 0 1 1 0

0 0 0 0 0 0 1 1 0 0 0 0 0 0 1 1

Synthesis filters:
1 1 0 0 0 0 0 0 1 1 0 0 0 0 0 0
0 1 1 0 0 0 0 0 0 1 1 0 0 0 0 0

0 0 1 1 0 0 0 0 0 0 1 1 0 0 0 0

t 1 0 0 0 1 1 0 0 0 t 1 0 0 0 1 1 0 0 0
Hl = Hh =
2 0 0 0 0 1 1 0 0 2 0 0 0 0 1 1 0 0

0 0 0 0 0 1 1 0 0 0 0 0 0 1 1 0
0 0 0 0 0 0 1 1 0 0 0 0 0 0 1 1

1 0 0 0 0 0 0 1 1 0 0 0 0 0 0 1
 The discrete wavelet transform

Often the signal will be discrete due to sampling

Will a digital filter bank do the job?

Two-scale relation:

(2 t ) = h j +1 (k ) (2 t k )
j j +1

Scaling function at one scale can be expressed from translated

scaling functions at the next smaller scale!
 Haar Wavelets

The Haar wavelet system

 Haar Wavelets
The Haar wavelet system
 Haar Wavelets

The Haar wavelet system (example1)

 Haar Wavelets

The Haar wavelet system (example)
Signals : 37, 35, 28, 28, 58, 18, 21, 15
 Haar Wavelets

The Haar wavelet system

Signals : 37, 35, 28, 28, 58, 18, 21, 15
Decomposition (analysis) :
37 35 28 28 58 18 21 15
36 28 38 18 1 0 20 3
32 28 4 10 1 0 20 3
30 2 4 10 1 0 20 3

Averaging Differencing
(37+35)/2=36, (37-35)/2=1
(28+28)/2=28, (28-28)/2=0
(58+18)/2=38, (58-18)/2=20
(21+18)/2=15 (21-18)/2=3
 Haar Wavelets

The Haar wavelet system

Signals : 37, 35, 28, 28, 58, 18, 21, 15
Reconstruction (synthesis) :
30 2 4 10 1 0 20 3
32 28 4 10 1 0 20 3
36 28 38 18 1 0 20 3
37 35 28 28 58 18 21 15

30 + 2 = 32, 30 2 = 28
 Haar Wavelets

The Haar wavelet system

Signals : 37, 35, 28, 28, 58, 18, 21, 15

Threshold = 2 :
37 35 28 28 58 18 21 15
30 2 4 10 1 0 20 3
 Haar Wavelets

The Haar wavelet system

Signals : 37, 35, 28, 28, 58, 18, 21, 15
Threshold = 2 :
37 35 28 28 58 18 21 15
30 2 4 10 1 0 20 3
30 0 4 10 0 0 20 3
30 30 4 10 0 0 20 3
34 26 40 20 0 0 20 3
34 34 26 26 60 20 23 17
Truncate!
 Haar Wavelets

The Haar wavelet system (example)
Threshold = 2 :
37 35 28 28 58 18 21 15
34 34 26 26 60 20 23 17
 Application of Wavelets

Data Compression

Wavelets have good energy concentration properties.

Most DWT coefficients usually are very small.

They can be discarded without incorporating a
significant error in the reconstruction stage.
 Application of Wavelets

Denoising

Donoho & Johnstone have devised the wavelet
shrinkage denoising(WSD).

Noise will show itself at finer scales.

Discarding the coefficients that falls below a
certain threshold at these scales will remove the
noise.
 Application of Wavelets
Denoising
 Denoising
Raw OCT image
(false color coded)

Denoising using
wavelet transform
and soft thresholding
 Soft Skin on the Palm

Depth

Averaging over
One scan per position
ten scans per position

Speckle noise is not

removed here!
Denoising using
wavelet decomposition

Hidden structured
low high regions are extracted !
 Different Denoising Scale Levels

low high
Compression using DWT

Compression ratio: 26:1

Demonstration the Quality of a
Compressed Image

Original: left side

Compression Ratio of
100:1 no degradation is
visible
Demonstration the Quality of a
Compressed Image (2)

Comparison of
compression tools

Wavelet compressor
80:1; left inset

JPEG 80:1; right inset
 Applications -Medical Imaging

Tumor detection, Microcalcification detection

 Two Dimensional Wavelets

High Pass and Low Pass Combinations

High and Low Pass in Horizontal and Vertical Directions of the
Image Data

Low Pass Vertical High Pass Vertical In Each Quadrant

Low Pass Horizontal Low Pass Horizontal - One-Quarter Original Image
- Down-sampled X/2 by X/2

Low Pass Vertical High Pass Vertical

High Pass Horizontal High Pass Horizontal
 Current Researchers Organizations

Lawrence Livermore National Laboratory

BellLabs

Rice University

Schools such as

Yale

Berkley

Dartmouth

Georgia Tech
 Individual Researchers

Ingrid Daubechies still exploring the area of

wavelets

Chris Brislawn current developer for the wavelet
compression used by the FBI

Ali N. Akansu working at the NJCMR working on
wavelets and subband coding
 For Further Information

http://engineering.rowan.edu/~polikar/WAVELETS/
WTtutorial.html (Wavelet tutorial)

www.wavelet.org (wavelet digest)

www.dsp.rice.edu (publications in the area of signal
processing)

http://www.mt.mevis.de/products/MT-WICE/ (for
high quality wavelet image compression software)

http://www.cosy.sbg.ac.at/~uhl/wav.html (excellent
link for references, papers, java applets)

http://www.amara.com/current/wavelet.html
(excellent resource page of links)

http://perso.wanadoo.fr/polyvalens/clemens/wavelets/
wavelets.html - A really friendly guide to wavelets
 Wavelet Books
Wavelets and Subband Coding by Martin Vetterli and
Jelena Kovacevic (Prentice Hall,1995)

Wavelets and Filter Banks by Gilbert Strang and Truong

Nguyen (Wellesley-Cambridge Press, 1996)
Wavelets in Medicine and Biology edited by Akram Aldroubi
and Michael Unser (CRC Press, 1996)
A Wavelet Tour of Signal Processing by Stphane Mallat
(AcademicPress,1998)
 Introduction to Wavelets and Wavelets
Transforms
by C. Sidney Burrus (Contributor), et al
(Prentice-Hall 1997)
A Friendly Guide to Wavelets
by Gerald Kaiser (Springer Verlag June 1994)

Fundamentals of Wavelets : Theory,

Algorithms, and Applications
by Jaideva C. Goswami & Andrew K. Chan
(Wiley-Interscience April 1999)
 Courtesy

Dr.Murali Krishna. P
[email protected]
Dr. Robi Polikar
[email protected]
Mr.Min Wang
[email protected]
 That is all, folks
Thank you for your patience!
Ja
ck
ie
THANK YOU