Introduction to Wavelets and Applications

Doris Hinestroza Gutierrez, Ph.D.

Departamento de Matematicas
Facultad de Ciencias Naturales y Exactas
Universidad del Valle

August 4, 2009

Doris H. (V) Fourier-Wavelet August 4, 2009 1 / 30

Outline

1 References

2 Historical Development

3 Introduction

4 Fourier analysis

5 Wavelet Analysis

Doris H. (V) Fourier-Wavelet August 4, 2009 2 / 30

References
Unser M. Sampling - 50 year after Shannon. Proceedings of the IEEE, Vol
88, 2000.
Bachman G, Narici L, Beckestein E. Fourier and Wavelets Analysis, Springer.
2000.
Benson D. Mathematics and Music, Aberdeen, 2006.
Brémaud P. Mathematical principles of signal processing, Springer. 2002.
Mallat S. A wavelet tour of signal processing, Academic Press. 1999.
Strang, G., Nguyen, T. (1996). Wavelets and Filter Banks.
Wellesley-Cambridge Press
Daubechies, I. (1992). Ten Lectures on Wavelets. Society for Industrial and
Applied Mathematics.
Holschneider, M. (1995). Wavelet An Analysis Tool. Clarendon Press.

Valens, C. (1995). A really friendly guide to wavelets.

http://perso.wanadoo.fr/polyvalens/clemens/wavelets/wavelets.html
Doris H. (V) Fourier-Wavelet August 4, 2009 3 / 30
Historical Development

1807 -Joseph Fourier: that every periodic function can be express as

infinity sum of sins and cosine in different frequencies.
1909 –Alfred Haar: Haar basis (latter will be the first wavelets)
1946 –Gabor: functions decompose in packets of time-frequency.
1984 –Alex Grossman, Morlet develops a model. “wavelet” appears for
first time.
1985 – Ives Meyer: first wavelet orthogonal.
1987 – Ingrid Daubechies: first orthogonal wavelets with compact
support.
1990 – David Donoho and Johnstone: Wavelets to eliminate the noise
of signal.
2004 .....– After the 90 wavelets can be an useful tool to analysis and
practical problems

Doris H. (V) Fourier-Wavelet August 4, 2009 4 / 30

Some Application of Wavelets

Mathematics:
Approximation Theory
Differential equations
Integral Equations
Fractional dimension
Signal and Images Processing
Geophysics:
Decomposition in time-frequency of
seismic signals, fractal analysis, elimination
of noise, Inversion, data processing
scaling
Medicine: Medical Images

Doris H. (V) Fourier-Wavelet August 4, 2009 5 / 30

Why to use Transform?

Ψ : X →Y
f → Ψ(f )
TRANSFORMS:
The transform function can give us information than the original
function.
Using transforms we can solve equation in a simpler form.
Using transforms we can require less space. (compression of data)
We can apply other simple operations to the transform function than
the original (Convolution), Derivatives, etc)
Doris H. (V) Fourier-Wavelet August 4, 2009 6 / 30
Hilbert Spaces
H : Hilbert Space: h, ·, i Inner product
x, y ∈ H, x⊥y ⇐⇒ hx, y i = 0.

V a subspace of H , define
V ⊥ = {h ∈ H : hh, v i = 0, for all v ∈ V }

V ⊥ : Orthogonal complement of V .

Theorem
(Projection Theorem) Let V a closed subspace of the Hilbert space H and

b ∈ H. Then There is a unique u ∈ V such that

kb − uk = inf {kb − v k : v ∈ A} , u = PrV b.

Doris H. (V) Fourier-Wavelet August 4, 2009 7 / 30
Fourier Series

Doris H. (V) Fourier-Wavelet August 4, 2009 8 / 30

Fourier Series
If f ∈ L2 [0, 2π] with period 2π, f has a representation
∞
X
f (t) = Cn e int ,
n=−∞

where Fourier coefficients Cn

Z 2π
1
Cn = f (t)e −int dt.
2π 0

The series converges f (t). If we define

N
X
fN (t) = Cn e int ,
n=−N

kf − fN k2 −→ 0, cuando N −→ ∞.

Doris H. (V) Fourier-Wavelet August 4, 2009 9 / 30

The functions Wn are orthonormal basis in L2 [0, 2π] .

Wn (t) = e int ,

Wn is generated by a unique function

W (t) = e it = cos t + i sin t

Wn (t) =W (nt)

Figura: Real and Imaginary part de Wo (t) − W1 (t) − W2 (t)

Doris H. (V) Fourier-Wavelet August 4, 2009 9 / 30
If L2 (R) is the space of square integrable functions
Z ∞
|f (t)|2 dt < ∞.
−∞

If f , g ∈ L2 (R) the inner product in L2 (R) is given by

Z ∞
hf , g i = f (t)g (t)dt,
−∞
kf k2 = hf , f i.

If f ∈ L2 (R) f decays to 0 as ±∞.

¿Is there a function ψ ∈ L2 (R) that generes all L2 (R)?

Let’s considerer translations of ψ (t − k), k ∈ Z

Let’s consider dilations of ψ, ψ(2j t − k) j, k ∈ Z

ψj,k (t) = 2j/2 ψ(2j . − k)

kψj,k k = kψk
Doris H. (V) Fourier-Wavelet August 4, 2009 10 / 30
Wavelet
Definition: A function ψ ∈ L2 (R) is called an orthogonal wavelet if the family
{ψj,k }j,k∈Z is an orthonormal basis for L2 (R), that is
hψj,k , ψl,m i =δjl δkm j, k, l, m ∈ Z,
donde
ψj,k =2j/2 ψ(2j t − k), j, k ∈ Z, j > 0.
, for each f ∈ L2 (R) is represented by the series
X
f (t) = cj,k ψj,k (t)
j,k∈Z

where
cj,k = hf , ψj,k i
Z ∞
= f (t)ψj,k (t)dt
−∞
Z ∞
= 2j/2 f (t)ψ(2j t − k)dt.
−∞
,→ Doris H. (V) Fourier-Wavelet August 4, 2009 10 / 30
Fourier Transform
The Fourier Transform of a function f ∈ L2 (R) is defined by
Z ∞
f (ω) =
b f (t) e −iωt dt
−∞

The function f is recovered by

Z ∞
1
f (t) = f (ω)e iωt dω.
b
2π −∞

Parseval Identity
If f ∈ L2 (R) then
Z Z
1
2
|f (t)| dt = |fˆ(w )|2 dw ,
R 2π R

or
1 ˆ 2
kf k22 = kf k2 .
2π
Doris H. (V) Fourier-Wavelet August 4, 2009 11 / 30
EXAMPLE

Example

f (t) = cos(2π20t) + cos(2π50t) + cos(2π100t) + cos(2π200t)

Doris H. (V) Fourier-Wavelet August 4, 2009 12 / 30

STATIONARY AND NON-STATIONARY SIGNALS

Example

Doris H. (V) Fourier-Wavelet August 4, 2009 13 / 30

FOURIER TRANSFORM OF THE SIGNAL1
Example

Doris H. (V) Fourier-Wavelet August 4, 2009 14 / 30

FOURIER TRANSFORM OF THE SIGNAL2

Example

Doris H. (V) Fourier-Wavelet August 4, 2009 15 / 30

FOURIER TRANSFORM OF THE SIGNAL3

Example

Doris H. (V) Fourier-Wavelet August 4, 2009 16 / 30

Short Time Fourier Transform

If a function f is continuous, f ∈ L2 (R) and g ∈ L2 (R) window function, we

define the Short Time Fourier Transform as:
Z ∞
fbg (τ, ω) = f (t)g (t − τ )e −iωt dt.
−∞

τ is used to translate the window.

Given a function f ∈ L2 (R), we can reconstruct the original function by using the
inversion formula:

Z ∞ Z ∞
f (t) = fbg (τ, ω)e iωt dωdτ.
−∞ −∞

Doris H. (V) Fourier-Wavelet August 4, 2009 17 / 30

STFT THE SIGNAL

Doris H. (V) Fourier-Wavelet August 4, 2009 18 / 30

STFT OF THE SIGNAL2

Doris H. (V) Fourier-Wavelet August 4, 2009 19 / 30

FOURIER TRANSFORM OF THE SIGNAL4

Doris H. (V) Fourier-Wavelet August 4, 2009 20 / 30

FOURIER TRANSFORM OF THE SIGNAL4

Doris H. (V) Fourier-Wavelet August 4, 2009 21 / 30

Continuous Wavelet Transform (CWT)
The Continuous Wavelet Transform Wψ of the function f ∈ L2 (R) with respect
to the waveletψ is defined as :
Z ∞
1 t −b
(Wψ f ) (a, b) = hf , ψa,b i = √ f (t)ψ( )dt
a −∞ a
wherea, b ∈ H = {(a, b) ∈ R 2 : a > 0} y
1 t −b
ψa,b (t) = √ ψ( )
a a
ψ(t) is know as a mother wavelet.
To construct wavelets two have two properties very important:
1 ψ has to have mean 0, that means
Z ∞
ψ(t)dt = 0
−∞
2 ψ has to satisfy the admissibility condition given by
Z ∞ b
1 |ψ(ω)|2
Cψ = √ dω < +∞.
2π 0 |ω|
Doris H. (V) Fourier-Wavelet August 4, 2009 22 / 30
Some properties of the CWT
1 Convolution: The CWT can be written as
Wψ f (a, b) = f ∗ ψ̄a (b)
where
1 −t
ψ̄a (t) = √ ψ( )
a a
2 Isometry: Given wavelet ψ, such that
 2
[)
Z ∞ ψ(w
Cψ = dw < ∞
0 |w |
then
kWψ f kL2 (H) = Cψ kf k.
kWψ f k, is the norm respect to the measure da a2 .
3 Parseval Formula: Let ψ ∈ L2 (R) admissible Wavelet, then f ∈ L2 (R) we
have
Z ∞
1 da
hf , g iL2 (R) = Wψ f (a, b)Wψ g (a, b) 2 db.
Cψ −∞ a
Doris H. (V) Fourier-Wavelet August 4, 2009 23 / 30
Inverse Wavelet Transform

Reconstruction Formula - Calderón, Grossman, Morlet

Let ψ ∈ L2 (R) admissible wavelet for every f ∈ L2 (R) we have
Z
1 da
f (t) = Wψ f (a, b)ψa,b (t) 2 db
Cψ H a

where H = (a, b) ∈ R2 : a > 0 .

Doris H. (V) Fourier-Wavelet August 4, 2009 23 / 30

How to built Wavelets

Two properties for the ψ to be a wavelet

1 ψ has mean 0, Z ∞
ψ(t)dt = 0. (1)
−∞
2 ψ has to satisfy the admissibility condition

∞ 2
|ψ(ω)|
Z b
Cψ = dω < +∞.
0 |ω|
If ψ satisfies the admissibility condition then ψ has mean 0.

Doris H. (V) Fourier-Wavelet August 4, 2009 24 / 30

Lemma 1 Wavelets

Lemma
Let ϕ ∈ L2 (R) a function such that ϕ(n) ∈ L2 (R). Then ψ = ϕ(n) satisfy
the admissibility condition and therefore ψ is a wavelet.

Proof.
|ψ(ω)|
b = |ω|n |ϕ(ω)
b

∞ 2 ∞ 2
|ψ(ω)| |ϕ(ω)|
Z b Z
|ω|2n
b
Cψ = dω = dω
−∞ |ω| −∞ |ω|
Z1 Z
2n−1 2
= |ω| |ϕ(ω)|
b dω + |ω|2n |ϕ(ω)|
b 2
dω
−1 |ω|≥1

≤ kϕk2 + kϕ(n) k2 < ∞.

Doris H. (V) Fourier-Wavelet August 4, 2009 25 / 30

Lemma 2 Wavelets

Lemma
For every ψ 6= 0, ψ ∈ L2 (R) with compact support, the following are
equivalent
a The function ψ is a wavelet.
b The function ψ satisfies the admissibility condition.

Doris H. (V) Fourier-Wavelet August 4, 2009 26 / 30

Lemma 3 Wavelets

Theorem
Let ψ a wavelet and ϕ a bounded integrable function then ψ ∗ ϕ is a
wavelet

Doris H. (V) Fourier-Wavelet August 4, 2009 27 / 30

Wavelet Examples

Example
Let ψ defined by

0 ≤ x < 12

 1
1
ψ(x) = −1 2 ≤x <1 .
0 In other case


Doris H. (V) Fourier-Wavelet August 4, 2009 28 / 30

Wavelet Examples

Example
( Mexican Hat Wavelet). Considerer the function
2 /2
ϕ(x) = e −x ∈ L2 (R).

we have that
d 2 ϕ(x) 2
2
= (1 − x 2 )e −x /2 ∈ L2 (R).
x
Therefore
2 /2
ψ = (1 − x 2 )e −x ,
is a wavelet

Doris H. (V) Fourier-Wavelet August 4, 2009 29 / 30

the Mexican Hat Wavelet

Doris H. (V) Fourier-Wavelet August 4, 2009 30 / 30