Wavelet Introduction

Wavelet Transform
Eva Hostalkova
WAVELET TRANSFORM
Eva Hostalkova
Dept of Computing and Control Engineering

Institute of Chemical Technology, Prague
ATHENS Nov 2009
Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 1 / 36

Wavelet Transform
Eva Hostalkova
1 Introduction to Wavelet Transform
2 Fourier Transform
Continuous Fourier Transform
Short-Time Fourier Transform (STFT)
3 Wavelet Transform
Multi-Resolution Analysis (MRA)
Continuous Wavelet Transform (CWT)
Wavelet Functions Properties
4 Discrete Wavelet Transform
Discrete and Continuous Wavelet Transform
Subband Coding Algorithm
Matrix Interpretation
5 Wavelet Transform Applications
Discontinuity Detection in the ECG Signal
Image Compression
Image Segmentation
Noise Reduction by Wavelet Shrinkage

Wavelet Transform
Eva Hostalkova
2 Fourier Transform
3 Wavelet Transform
Image Compression
Image Segmentation

Wavelet Transform
Eva Hostalkova
2 Fourier Transform
3 Wavelet Transform
Image Compression
Image Segmentation

Wavelet Transform
Eva Hostalkova
2 Fourier Transform
3 Wavelet Transform
Image Compression
Image Segmentation

Wavelet Transform
Eva Hostalkova
2 Fourier Transform
3 Wavelet Transform
Image Compression
Image Segmentation

Introduction
Wavelet Transform
Eva Hostalkova
Introduction 2 Fourier Transform
Fourier Transform Continuous Fourier Transform
STFT Short-Time Fourier Transform (STFT)
Wavelet Transform
MRA
3 Wavelet Transform
CWT
Wavelets Properties
DWT
DWT and CWT Wavelet Functions Properties
Subband Coding
WT Applications
Discontinuity Detection Discrete and Continuous Wavelet Transform
Image Compression
Image Segmentation
Noise Reduction Matrix Interpretation
References
Image Compression
Image Segmentation

Introduction
Wavelet Transform
Eva Hostalkova
Wavelet Transform (WT) History
Introduction
19th cent. - Jean B. Fourier: frequency analysis
Fourier Transform 1909 - Alfred Haar: Haar function (not yet WT)
STFT since 1990s - WT related research and applications
Wavelet Transform
MRA
CWT
Wavelets Properties Wavelet Functions
DWT
DWT and CWT
Wavelet = a small wave, i.e. an oscillatory function - zero
Subband Coding outside a bounded interval (having compact support)
WT Applications
Some real, some complex
Discontinuity Detection
Image Compression
Designed as much smooth and symmetric as possible
Image Segmentation
Noise Reduction
Seldom analytic expression, mostly parametric (e.g.
References Morlet, Haar or Daubechies fcns)

Introduction
Wavelet Transform
Eva Hostalkova
Wavelet Transform (WT) History
Introduction
19th cent. - Jean B. Fourier: frequency analysis
Fourier Transform 1909 - Alfred Haar: Haar function (not yet WT)
STFT since 1990s - WT related research and applications
Wavelet Transform
MRA
CWT
Wavelets Properties Wavelet Functions
DWT
DWT and CWT
Wavelet = a small wave, i.e. an oscillatory function - zero
Subband Coding outside a bounded interval (having compact support)
WT Applications
Some real, some complex
Image Compression
Designed as much smooth and symmetric as possible
Image Segmentation
Noise Reduction
Seldom analytic expression, mostly parametric (e.g.
References Morlet, Haar or Daubechies fcns)
DAUBECHIES 2 WAVELET
0.1
0.05
−0.05
10 20 30 40 50 60 70
time

Introduction
Wavelet Transform
Eva Hostalkova
Introduction
WT Analysis
Fourier Transform
STFT
For non-stationary signals (with time-varying frequency
Wavelet Transform
content)
MRA
CWT
Wavelets Properties
WT Applications in Signal Processing
DWT
DWT and CWT Noise reduction
Subband Coding
Matrix Interpretation Detection of trends and discontinuities in higher
WT Applications
derivatives
Image Compression
Image Segmentation
Compression (JPEG2000, FBI fingerprints database)
Noise Reduction
Image edge detection
References
Watermarking
Features extraction for image segmentation

Introduction
Wavelet Transform
Eva Hostalkova
Introduction
WT Analysis
Fourier Transform
STFT
For non-stationary signals (with time-varying frequency
Wavelet Transform
content)
MRA
CWT
Wavelets Properties
WT Applications in Signal Processing
DWT
DWT and CWT Noise reduction
Subband Coding
Matrix Interpretation Detection of trends and discontinuities in higher
WT Applications
derivatives
Image Compression
Image Segmentation
Compression (JPEG2000, FBI fingerprints database)
Noise Reduction
Image edge detection
References
Watermarking
Features extraction for image segmentation

Fourier Transform
Wavelet Transform
Eva Hostalkova
Wavelet Transform
MRA
3 Wavelet Transform
CWT
Wavelets Properties
DWT
Subband Coding
WT Applications
Image Compression
Image Segmentation
References
Image Compression
Image Segmentation

Wavelet Transform
Eva Hostalkova
Continuous Fourier Transform (FT)
Introduction
Fourier Transform
Decomposes signals (real or complex) into the orthogonal
Continuous Fourier Transform basis of complex exponentials e −j2πft of frequencies f
STFT
Wavelet Transform
FT for the angular frequency ω = 2πf and time t:
MRA
Z +∞
CWT X (ω) = x (t) e −jωt dt (1)
Wavelets Properties
−∞
DWT
DWT and CWT
Subband Coding
where X (ω) is the Fourier transform of the function x (t)
Converges for a piece-wise smooth x (t) or finite energy:
WT Applications Z ∞
|x (t)|2 dt < ∞
Image Compression (2)
Image Segmentation −∞
Noise Reduction
References
Inverse FT
Z +∞
1
x (t) = X (ω) e jωt dω (3)
2π −∞

Wavelet Transform
Eva Hostalkova
Continuous Fourier Transform (FT)
Introduction
Fourier Transform
Decomposes signals (real or complex) into the orthogonal
Continuous Fourier Transform basis of complex exponentials e −j2πft of frequencies f
STFT
Wavelet Transform
FT for the angular frequency ω = 2πf and time t:
MRA
Z +∞
CWT X (ω) = x (t) e −jωt dt (1)
Wavelets Properties
−∞
DWT
DWT and CWT
Subband Coding
where X (ω) is the Fourier transform of the function x (t)
Converges for a piece-wise smooth x (t) or finite energy:
WT Applications Z ∞
|x (t)|2 dt < ∞
Image Compression (2)
Image Segmentation −∞
Noise Reduction
References
Inverse FT
Z +∞
1
x (t) = X (ω) e jωt dω (3)
2π −∞

Wavelet Transform
FT Analysis of Non-Stationary Signals
Eva Hostalkova
Introduction
Discrete FT for the discrete freq. index k = 0, 1, . . . , N −1
Fourier Transform and discrete time n (for the unit sampling frequency)
STFT
N−1
X
Wavelet Transform X (k) = x (n) e −j2πkn/N (4)
MRA
n=0
CWT
(a) SIGNAL 1 (b) MAGNITUDE SPECTRUM 1
Wavelets Properties 250
3
DWT
2 200
DWT and CWT
1
Subband Coding 150
Matrix Interpretation 0
100
−1
WT Applications
Discontinuity Detection −2 50
Image Compression −3
Image Segmentation 0 100 200 300 400 500 0 0.1 0.2 0.3 0.4
Noise Reduction
(c) SIGNAL 2 (d) MAGNITUDE SPECTRUM 2
1
References
60
0.5 50
40
0
30
20
−0.5
10
−1
0 100 200 300 400 500 0 0.1 0.2 0.3 0.4
time frequency

Fourier Transform STFT
Wavelet Transform
Discrete Short-Time Fourier Transform (STFT)
Eva Hostalkova
Introduction
Non-stationary signal analysis (localizes frequency
Fourier Transform
components within time intervals)
STFT
STFT of a discrete signal x (n) for k = 0, 1, . . . , N −1:
Wavelet Transform
MRA N−1
X
CWT
Wavelets Properties
X (u, k) = x (n) w (n − u)e −j2πkn/N (5)
DWT n=0
DWT and CWT
Subband Coding where u denotes the window position
WT Applications
Image Compression
Uncertainty (Heisenberg) Principle
Image Segmentation
Noise Reduction
∆T ∆f = 1 (6)
References
∆f = Fs /N . . . frequency resolution (distance between

adjacent spectral samples)
∆T = N Ts . . . time resolution (window length)
N . . . number of samples per window
Fs = 1/Ts . . . sampling frequency
Wavelet Transform
Discrete Short-Time Fourier Transform (STFT)
Eva Hostalkova
Introduction
Non-stationary signal analysis (localizes frequency
Fourier Transform
components within time intervals)
STFT
STFT of a discrete signal x (n) for k = 0, 1, . . . , N −1:
Wavelet Transform
MRA N−1
X
CWT
Wavelets Properties
X (u, k) = x (n) w (n − u)e −j2πkn/N (5)
DWT n=0
DWT and CWT
Subband Coding where u denotes the window position
WT Applications
Image Compression
Uncertainty (Heisenberg) Principle
Image Segmentation
Noise Reduction
∆T ∆f = 1 (6)
References
∆f = Fs /N . . . frequency resolution (distance between

adjacent spectral samples)
∆T = N Ts . . . time resolution (window length)
N . . . number of samples per window
Fs = 1/Ts . . . sampling frequency
Wavelet Transform
STFT - Uncertainty Tradeoff
Eva Hostalkova
Introduction (a) SIGNAL 2 (b) STFT: WINDOW L=32

1
Fourier Transform
0.5
STFT 6
Wavelet Transform 4
MRA 0
2
CWT
Wavelets Properties
−0.5
30
DWT 15
20
DWT and CWT 10
10
−1 5
Subband Coding 0 100 200 300 400 500 frequency ξ time u
Matrix Interpretation time
WT Applications (c) STFT: WINDOW L=128 (d) STFT: WINDOW L=256

Image Compression
Image Segmentation
30 30
Noise Reduction
20 20
References
10 10
120 250
100 4 200 2
80 150
60 3 100
40 1.5
20 2 50
frequency ξ 1 time u frequency ξ 1 time u
(Using the Hamming window w - gently dropping to zero)

Wavelet Transform
Wavelet Transform
Eva Hostalkova
Wavelet Transform
MRA
3 Wavelet Transform
CWT
Wavelets Properties
DWT
Subband Coding
WT Applications
Image Compression
Image Segmentation
References
Image Compression
Image Segmentation

Wavelet Transform MRA
Wavelet Transform
Eva Hostalkova
Introduction
Fourier Transform
Continuous Fourier Transform Multi-Resolution Analysis (MRA)
STFT
Wavelet Transform The wavelet transform deals with the uncertainty principle
MRA
CWT
by MRA
Wavelets Properties
Analyzing signals at different frequency bands with
DWT
DWT and CWT
different resolution:
Subband Coding Higher frequencies: good ∆T , poor ∆f
Lower frequencies: good ∆f , poor ∆T
WT Applications
Discontinuity Detection Convenient for most real-world signals composed of:
Image Compression
Image Segmentation Long-lasting lower frequencies (approximations)
Noise Reduction
Short-lasting higher frequencies (details - main
References
information)

Wavelet Transform CWT
Wavelet Transform
Eva Hostalkova
Introduction FT: correlation of a signal of finite energy with complex
Fourier Transform exponentials e j2πft of different frequencies f
STFT WT: correlation of a signal of finite energy with delated
Wavelet Transform and shifted versions ψu,s of the mother wavelet ψ:
MRA

CWT
1 t −u
Wavelets Properties ψu,s (t) = √ ψ (7)
DWT s s
DWT and CWT
Subband Coding
where u is the shift of ψ along the signal → time
WT Applications √s is scale (dilation of ψ → the inverse of frequency)
and
Discontinuity Detection 1/ s ensures energy normalization
Image Compression
Image Segmentation
Noise Reduction
References CWT of the signal x (t):

+∞
t −u
Z
1
Wxψ (s, u) = hx , ψu,s i = √ x (t) ψ ∗ dt (8)
s −∞ s
where * denotes the complex conjugate pair

and hi stands for the inner product

Wavelet Transform
Eva Hostalkova
Introduction FT: correlation of a signal of finite energy with complex
Fourier Transform exponentials e j2πft of different frequencies f
STFT WT: correlation of a signal of finite energy with delated
Wavelet Transform and shifted versions ψu,s of the mother wavelet ψ:
MRA

CWT
1 t −u
Wavelets Properties ψu,s (t) = √ ψ (7)
DWT s s
DWT and CWT
Subband Coding
where u is the shift of ψ along the signal → time
WT Applications √s is scale (dilation of ψ → the inverse of frequency)
and
Discontinuity Detection 1/ s ensures energy normalization
Image Compression
Image Segmentation
Noise Reduction
References CWT of the signal x (t):

+∞
t −u
Z
1
Wxψ (s, u) = hx , ψu,s i = √ x (t) ψ ∗ dt (8)
s −∞ s
where * denotes the complex conjugate pair

and hi stands for the inner product

Wavelet Transform
Eva Hostalkova
CWT Computation Procedure (s = 1)
Introduction
Fourier Transform
STFT
Wavelet Transform
MRA
CWT
Wavelets Properties
DWT
DWT and CWT
Subband Coding
WT Applications
Image Compression
Image Segmentation
Noise Reduction
References

Wavelet Transform
Eva Hostalkova
CWT Computation Procedure (s = 5)
Introduction
Fourier Transform
STFT
Wavelet Transform
MRA
CWT
Wavelets Properties
DWT
DWT and CWT
Subband Coding
WT Applications
Image Compression
Image Segmentation
Noise Reduction
References

Wavelet Transform
Eva Hostalkova
Introduction CWT Properties

Fourier Transform
Linearity: implied by the properties of the inner product
STFT
Wavelet Transform
MRA Shift invariance:
CWT
Input
Wavelets Properties A shift in the signal
Wavelets
DWT along the time axis Level 1
DWT and CWT

Subband Coding
7→ the equivalent Level 2
Matrix Interpretation shift of the wavelet Level 3
WT Applications coefficients without Level 4
Image Compression
any changes Scaling fn
Level 4
Image Segmentation Does not apply to (a) Dual Tree CWT (b) Real DWT
Noise Reduction
the discrete WT
References
(great drawback)
Figure: Comparison of DWT and nearly invariant complex WT

(CWT is even better - completely shift invariant)

Wavelet Transform
Eva Hostalkova
Orthogonality and orthonormality
Introduction
Fourier Transform Not necessary for wavelet analysis (biorthogonality for

STFT more freedom in wavelet filters construction)
Wavelet Transform
MRA
A set of shifted and dilated versions of the mother wavelet
CWT ψu,s , is orthonormal on the interval [a, b] when:
Wavelets Properties
DWT Z b
DWT and CWT 1 for u = v , s = r
Subband Coding hψu,s , ψv ,r i = ψu,s (t) ψv∗,r (t) dt =
a 0 otherwise
WT Applications (9)
Image Compression
Image Segmentation
Noise Reduction
A set of ψu,s , is orthogonal on the interval [a, b] when:
References
Z b
c for u = v , s = r
hψu,s , ψv ,r i = ψu,s (t) ψv∗,r (t) dt =
a 0 otherwise
(10)
∗
where c is a constant, denotes a complex conjugate

Wavelet Transform Wavelets Properties
Wavelet Transform
Eva Hostalkova Wavelet Functions Properties I

Introduction Admissibility condition
Z ∞
Fourier Transform
| Ψ(ω)|2
dω < +∞ (11)
STFT
−∞ | ω|
Wavelet Transform
MRA where Ψ(ω) is the FT of ψ(t)
CWT
Wavelets Properties
Required for no loss of information during the analysis nor
DWT
the synthesis
DWT and CWT The basis do not have to be orthogonal
Subband Coding
WT Applications
Consequences of Admissibility Condition
Image Compression
Image Segmentation Oscillatory function (zero mean):
Noise Reduction
Z ∞
References
ψ(t) dt = 0 (12)
−∞
Band-pass spectrum (Ψ(ω) vanishes at ω = 0):

| Ψ(ω)|2 | ω=0 = 0 (13)

Wavelet Transform
Eva Hostalkova Wavelet Functions Properties I

Introduction Admissibility condition
Z ∞
Fourier Transform
| Ψ(ω)|2
dω < +∞ (11)
STFT
−∞ | ω|
Wavelet Transform
MRA where Ψ(ω) is the FT of ψ(t)
CWT
Wavelets Properties
Required for no loss of information during the analysis nor
DWT
the synthesis
DWT and CWT The basis do not have to be orthogonal
Subband Coding
WT Applications
Consequences of Admissibility Condition
Image Compression
Image Segmentation Oscillatory function (zero mean):
Noise Reduction
Z ∞
References
ψ(t) dt = 0 (12)
−∞
Band-pass spectrum (Ψ(ω) vanishes at ω = 0):

| Ψ(ω)|2 | ω=0 = 0 (13)

Wavelet Transform
Wavelet Functions Properties II
Eva Hostalkova
Introduction Time dilation of ψ causes shift and compression of the

Fourier Transform magnitude frequency spectrum |Ψ|
STFT
Wavelet Transform
MRA DILATION OF THE WAVELET FT OF THE DILATED WAVELET
1 10
CWT
s=20
Wavelets Properties 0.5 s=2 0
8 s=21
DWT 0 s=22
6
DWT and CWT
−0.5
Subband Coding 4
−50 0 50
2
1
WT Applications
Discontinuity Detection 0.5 s = 21 0 0.1 0.2 0.3 0.4 0.5
Image Compression norm. frequency
0
Image Segmentation
Noise Reduction −0.5 NEWLAND WAVELET:
References −50 0 50 1 π
ψ(t) = π (ej π t − ej 2
t)
1 j 2 t
0.5 s = 22
WAVELET SCALING & DILATION:
0
µ ¶
1 t−u
−0.5 ψu,s (t) = √ ψ
s s
−50 0 50
time

Wavelet Transform
Eva Hostalkova
Introduction
Fourier Transform
STFT
Desired Wavelet Functions Properties
Wavelet Transform
MRA
CWT Compact support (non-zero only on a restricted interval) -
Wavelets Properties
not necessary
DWT
DWT and CWT Regularity:
Subband Coding
Matrix Interpretation Smoothness of ψ and vanishing of |Ψ(ω)| for large ω
WT Applications (small scales)
Vanishing moments concept (see literature)
Image Compression
Image Segmentation
Noise Reduction
Symmetry (liner phase)
References

DWT
Wavelet Transform
Eva Hostalkova
Wavelet Transform
MRA
3 Wavelet Transform
CWT
Wavelets Properties
DWT
Subband Coding
WT Applications
Image Compression
Image Segmentation
References
Image Compression
Image Segmentation

DWT DWT and CWT
Wavelet Transform
Eva Hostalkova Discrete Wavelet Transform (DWT)

Introduction CWT provides us with redundant signal representation
Fourier Transform
DWT - derive by critically sampling the CWT (great
STFT reduction in the number of dilations and shifts)
Wavelet Transform
!
t −k τ0 s0j
MRA
CWT 1 t −u 1
Wavelets Properties ψu,s (t) = √ ψ ⇒ ψj,k (t) = q ψ
DWT
s s
sj s0j
0
DWT and CWT
Subband Coding (14)
WT Applications
Discontinuity Detection Sampling on the Dyadic Grid
Image Compression
Image Segmentation
Noise Reduction
s0 = 2 ⇒ the scale s = 2j
References τ0 = 1 ⇒ the time translation u = k 2j
t denotes time, j, k are integers
t −k2j

1
ψj,k (t) = √ ψ (15)
2j 2j

DWT DWT and CWT
Wavelet Transform
Eva Hostalkova Discrete Wavelet Transform (DWT)

Introduction CWT provides us with redundant signal representation
Fourier Transform
DWT - derive by critically sampling the CWT (great
STFT reduction in the number of dilations and shifts)
Wavelet Transform
!
t −k τ0 s0j
MRA
CWT 1 t −u 1
Wavelets Properties ψu,s (t) = √ ψ ⇒ ψj,k (t) = q ψ
DWT
s s
sj s0j
0
DWT and CWT
Subband Coding (14)
WT Applications
Discontinuity Detection Sampling on the Dyadic Grid
Image Compression
Image Segmentation
Noise Reduction
s0 = 2 ⇒ the scale s = 2j
References τ0 = 1 ⇒ the time translation u = k 2j
t denotes time, j, k are integers
t −k2j

1
ψj,k (t) = √ ψ (15)
2j 2j

DWT DWT and CWT
Wavelet Transform
Eva Hostalkova
Discrete & Continuous Wavelet Transform
Introduction
(a) ECG SIGNAL
Fourier Transform
3500
STFT 3000
Wavelet Transform 2500

MRA
2000
CWT
Wavelets Properties 50 100 150 200 250 300 350 400 450 500
DWT (b) DISCRETE WAVELET TRANSF.: ABSOLUTE COEFFICIENTS
DWT and CWT 16 60
Subband Coding
8
Matrix Interpretation 40
scale
4
WT Applications
2 20
1
Image Compression
Image Segmentation
50 100 150 200 250 300 350 400 450 500
(c) CONTINUOUS WAVELET TRANSF.: ABSOLUTE COEFFICIENTS
Noise Reduction
16 60
15
14
References 13 50
12
11
10 40
scale
9
8 30
7
6
5 20
4
3 10
2
1
50 100 150 200 250 300 350 400 450 500
time

DWT Subband Coding
Wavelet Transform
Wavelet and Scaling Filters
Eva Hostalkova
Introduction
−3 (a) Db2 WAVELET FCN (b) Db2 WAVELET FCN: Freq. Response
x 10
Fourier Transform
6
Continuous Fourier Transform 1
4
Magnitude
STFT 0.8
2
0.6
Wavelet Transform 0
0.4
MRA −2
0.2
CWT −4
0
Wavelets Properties 200 400 600 0 0.1 0.2 0.3 0.4
time ω/2π
DWT −3
x 10 (c) Db2 SCALING FCN (d) Db2 SCALING FCN: Freq. Response
DWT and CWT
Subband Coding 1
4
Magnitude
Matrix Interpretation 0.8
2 0.6
WT Applications
0.4
0
Image Compression 0.2
Image Segmentation 0
200 400 600 0 0.1 0.2 0.3 0.4
Noise Reduction time ω/2π
References
The spectrum of the wavelet function corresponds to a

high-pass filter (or a band-pass filter for higher levels)
The spectrum of the scaling function corresponds to a
low-pass filter
DWT Subband Coding
Wavelet Transform
Wavelet and Scaling Filters
Eva Hostalkova
Introduction
−3 (a) Db2 WAVELET FCN (b) Db2 WAVELET FCN: Freq. Response
x 10
Fourier Transform
6
Continuous Fourier Transform 1
4
Magnitude
STFT 0.8
2
0.6
Wavelet Transform 0
0.4
MRA −2
0.2
CWT −4
0
Wavelets Properties 200 400 600 0 0.1 0.2 0.3 0.4
time ω/2π
DWT −3
x 10 (c) Db2 SCALING FCN (d) Db2 SCALING FCN: Freq. Response
DWT and CWT
Subband Coding 1
4
Magnitude
Matrix Interpretation 0.8
2 0.6
WT Applications
0.4
0
Image Compression 0.2
Image Segmentation 0
200 400 600 0 0.1 0.2 0.3 0.4
Noise Reduction time ω/2π
References
The spectrum of the wavelet function corresponds to a

high-pass filter (or a band-pass filter for higher levels)
The spectrum of the scaling function corresponds to a
low-pass filter
DWT Subband Coding
Wavelet Transform
Eva Hostalkova Subband Coding Algorithm

Introduction
Fourier Transform approximation

1D DWT
Continuous Fourier Transform coefficients
l [n] 2 A3
STFT d
Wavelet Transform l [n] 2

d
detail
MRA
coefficients
CWT h [n] 2 D3
l [n] 2 d
Wavelets Properties original d
signal detail
DWT x[n] coefficients
h [n] 2 D2
DWT and CWT d
Subband Coding detail

Matrix Interpretation coefficients
hd[n] 2 D1
WT Applications
Discontinuity Detection LEVEL 1 LEVEL 2 LEVEL 3
Image Compression
Image Segmentation
Noise Reduction
References
CWT versus DWT
Correlation Convolution
Dilating the Keeping filters the same and
wavelets down-sampling the previous level
output by 2

DWT Subband Coding
Wavelet Transform
Eva Hostalkova Subband Coding Algorithm

Introduction
Fourier Transform approximation

1D DWT
Continuous Fourier Transform coefficients
l [n] 2 A3
STFT d
Wavelet Transform l [n] 2

d
detail
MRA
coefficients
CWT h [n] 2 D3
l [n] 2 d
Wavelets Properties original d
signal detail
DWT x[n] coefficients
h [n] 2 D2
DWT and CWT d
Subband Coding detail

Matrix Interpretation coefficients
hd[n] 2 D1
WT Applications
Discontinuity Detection LEVEL 1 LEVEL 2 LEVEL 3
Image Compression
Image Segmentation
Noise Reduction
References
CWT versus DWT
Correlation Convolution
Dilating the Keeping filters the same and
wavelets down-sampling the previous level
output by 2

DWT Subband Coding
Wavelet Transform
Convolution and Downsampling by 2
Eva Hostalkova
The taps of the high-pass filter hd and the low-pass ld
Introduction
Fourier Transform
filter are derived from the wavelet and the scaling function,
Continuous Fourier Transform resp., of a chosen family (e.g. Daubechies, symlets, etc.)
STFT
Wavelet Transform
MRA Convolution and Downsampling by 2
CWT
Wavelets Properties Approximation coefficients of the first level
DWT
∞
X
DWT and CWT
A1 [n] = ld [k] x [2n − k] (16)
Subband Coding
Matrix Interpretation k=−∞
WT Applications
Detail coefficients of the first level
Image Compression
∞
X
Image Segmentation
D1 [n] = hd [k] x [2n − k] (17)
Noise Reduction
k=−∞
References
∞
The j-th level A [n] = X l [k] A [2n − k] (18)
j d j−1
k=−∞
∞
X
Dj [n] = hd [k] Aj−1 [2n − k] (19)
k=−∞
DWT Subband Coding
Wavelet Transform
Eva Hostalkova
Introduction
Fourier Transform
STFT
Wavelet Transform
CWT
DWT
∞
X
DWT and CWT
A1 [n] = ld [k] x [2n − k] (16)
Subband Coding
WT Applications
Image Compression
∞
X
Image Segmentation
D1 [n] = hd [k] x [2n − k] (17)
Noise Reduction
k=−∞
References
∞
j d j−1
k=−∞
∞
X
Dj [n] = hd [k] Aj−1 [2n − k] (19)
k=−∞
DWT Subband Coding
Wavelet Transform
Eva Hostalkova
Introduction
Fourier Transform
STFT
Wavelet Transform
CWT
DWT
∞
X
DWT and CWT
A1 [n] = ld [k] x [2n − k] (16)
Subband Coding
WT Applications
Image Compression
∞
X
Image Segmentation
D1 [n] = hd [k] x [2n − k] (17)
Noise Reduction
k=−∞
References
∞
j d j−1
k=−∞
∞
X
Dj [n] = hd [k] Aj−1 [2n − k] (19)
k=−∞
DWT Subband Coding
Wavelet Transform
Eva Hostalkova Subband Coding & the Frequency Spectrum

Introduction
Fourier Transform
STFT
Wavelet Transform
MRA
CWT
Wavelets Properties
DWT
DWT and CWT
Subband Coding
WT Applications
Image Compression
Image Segmentation
Noise Reduction
References
Dilated by 2 ⇒ the spectrum is compressed and shifted
Finite number of dilations ⇒ advantageous to use a
low-pass filter - derived from the scaling function φ(t)
(the counter part of the wavelet filter at each level -
creating a filter bank)

DWT Subband Coding
Wavelet Transform
Eva Hostalkova Subband Coding & the Frequency Spectrum

Introduction
Fourier Transform
STFT
Wavelet Transform
MRA
CWT
Wavelets Properties
DWT
DWT and CWT
Subband Coding
WT Applications
Image Compression
Image Segmentation
Noise Reduction
References
Dilated by 2 ⇒ the spectrum is compressed and shifted
Finite number of dilations ⇒ advantageous to use a
low-pass filter - derived from the scaling function φ(t)
(the counter part of the wavelet filter at each level -
creating a filter bank)

DWT Matrix Interpretation
Wavelet Transform
DWT Matrix
Eva Hostalkova
The Haar filters √
Introduction
ld = 1/ 2 · [1, 1] (20)
Fourier Transform
√
STFT hd = 1/ 2 · [1, −1] (21)
Wavelet Transform
MRA DWT matrix for the Haar filters
CWT  
Wavelets Properties 1 1 0 0 ... 0 0
DWT  −1 1 0 0 . . . 0 0 
DWT and CWT  
Subband Coding
 0 0 1 1 . . . 0 0 
1  
W = √ ·  0 0 −1 1 . . . 0 0
(22)
 
WT Applications

2  .. .. .. .. . . .. ..
Image Compression
 . .
 . . . . .

Image Segmentation  0 0 0 0 ... 1 1 
Noise Reduction
References
0 0 0 0 . . . −1 1
W includes both convolution and down-sampling by 2

(the filters are shifted by 2 samples)
W is orthonormal ⇒ W · WT = I
where I stands for the identity matrix
Wavelet Transform
DWT Matrix
Eva Hostalkova
The Haar filters √
Introduction
ld = 1/ 2 · [1, 1] (20)
Fourier Transform
√
STFT hd = 1/ 2 · [1, −1] (21)
Wavelet Transform
MRA DWT matrix for the Haar filters
CWT  
Wavelets Properties 1 1 0 0 ... 0 0
DWT  −1 1 0 0 . . . 0 0 
DWT and CWT  
Subband Coding
 0 0 1 1 . . . 0 0 
1  
W = √ ·  0 0 −1 1 . . . 0 0
(22)
 
WT Applications

2  .. .. .. .. . . .. ..
Image Compression
 . .
 . . . . .

Image Segmentation  0 0 0 0 ... 1 1 
Noise Reduction
References
0 0 0 0 . . . −1 1
W includes both convolution and down-sampling by 2

(the filters are shifted by 2 samples)
W is orthonormal ⇒ W · WT = I
where I stands for the identity matrix
Wavelet Transform
Eva Hostalkova
DWT of Signal x Using the Haar Filters
Introduction     
1 1 0 0 ... 0 0 x (0) A1 (0)
Fourier Transform
−1 1 0 0 ... 0 0   x (1)   D1 (0) 
 0 0 1 1 ... 0 0   x (2)   A1 (1) 
STFT
    
Wavelet Transform 1  0 0 −1 1 . . . 0 0  ·  x (3)  =  D1 (1) 
    
MRA
√ 
CWT  ...
2 .. .. .. . . .. ..  
  ..   .. 
Wavelets Properties  . . . . . .  .
 
  .


N
DWT
 0 0 0 0 ... 1 1  x (N −1)  A1 ( 2 −1) 
DWT and CWT 0 0 0 0 . . . −1 1 x (N) D1 ( N2 −1)
Subband Coding
WT Applications
Discontinuity Detection DWT Decomposition and Reconstruction
Image Compression
Image Segmentation
Signal decomposition
Noise Reduction
References W·x=w (23)

Signal reconstruction
x = W−1 · w (24)
Signal reconstruction for the orthonormal DWT
x = WT · w (25)

Wavelet Transform
Eva Hostalkova
DWT of Signal x Using the Haar Filters
Introduction     
1 1 0 0 ... 0 0 x (0) A1 (0)
Fourier Transform
−1 1 0 0 ... 0 0   x (1)   D1 (0) 
 0 0 1 1 ... 0 0   x (2)   A1 (1) 
STFT
    
Wavelet Transform 1  0 0 −1 1 . . . 0 0  ·  x (3)  =  D1 (1) 
    
MRA
√ 
CWT  ...
2 .. .. .. . . .. ..  
  ..   .. 
Wavelets Properties  . . . . . .  .
 
  .


N
DWT
 0 0 0 0 ... 1 1  x (N −1)  A1 ( 2 −1) 
DWT and CWT 0 0 0 0 . . . −1 1 x (N) D1 ( N2 −1)
Subband Coding
WT Applications
Discontinuity Detection DWT Decomposition and Reconstruction
Image Compression
Image Segmentation
Signal decomposition
Noise Reduction
References W·x=w (23)

Signal reconstruction
x = W−1 · w (24)
Signal reconstruction for the orthonormal DWT
x = WT · w (25)

WT Applications
Wavelet Transform
Eva Hostalkova
Wavelet Transform
MRA
3 Wavelet Transform
CWT
Wavelets Properties
DWT
Subband Coding
WT Applications
Image Compression
Image Segmentation
References
Image Compression
Image Segmentation

WT Applications Discontinuity Detection
Wavelet Transform
Eva Hostalkova Discontinuity Detection in the ECG Signal

Introduction
Fourier Transform
(a) 1−D DWT
STFT aproximace
ld[n] 2 A
Wavelet Transform 3
MRA ld[n] 2
nízkofrekvenèní
CWT propust detaily
Wavelets Properties
hd[n] 2 D
ld[n] 2 3
DWT signál
x[n] detaily
DWT and CWT hd[n] 2 D2
Subband Coding vysokofrekvenèní (c) DB4: ABS. DETAILNÍ KOEFICIENTY
propust
Matrix Interpretation detaily
hd[n] 2 D1
WT Applications
Image Compression (b) EKG SIGNÁL
Image Segmentation
Noise Reduction
100 200 300 400 500
References time
Nespojitost
0 100 200 300 400 500

time
Nespojitost

WT Applications Image Compression
Wavelet Transform
Eva Hostalkova
Image Compression
Introduction
Fourier Transform The Parseval theorem: the energy εx conveyed in the

STFT
signal x equals the energy of the coefficients w obtained
Wavelet Transform
through an orthonormal transform
MRA
CWT N−1
X N−1
X
Wavelets Properties
DWT
εx = |xn |2 = |wk |2 (26)
DWT and CWT n=0 k=0
Subband Coding
(a) ORIGINAL IMAGE (b) DECOMPOSED IMAGE (c) COMPRESSED IMAGE : 94.2%
WT Applications
Discontinuity Detection 94.2% 2% 0.8%
Image Compression
Image Segmentation
Noise Reduction
References 2% 0.4%
0.7% 0.1%

WT Applications Image Segmentation
Wavelet Transform
Magnetic Resonance Image Segmentation
Eva Hostalkova
Introduction
Fourier Transform
STFT
Wavelet Transform
MRA
CWT
Wavelets Properties
DWT
DWT and CWT
Subband Coding
WT Applications
Image Compression
Image Segmentation
Noise Reduction
References
1 Image preprocessing
2 Watershed transform
3 DWT features extraction
4 Features classification using a neural network
WT Applications Noise Reduction
Wavelet Transform
Eva Hostalkova
Introduction
Fourier Transform
STFT
Wavelet Transform
MRA
CWT
Wavelets Properties Noisy
DWT Image
DWT and CWT
Subband Coding
WT Applications
Image Compression
Image Segmentation
Noise Reduction
References

Wavelet Transform
Eva Hostalkova
Introduction
Fourier Transform
STFT
Wavelet Transform
MRA
CWT
Wavelets Properties Noisy
DWT Image
DWT and CWT
Subband Coding
WT Applications
Wavelet
Image Compression Analysis
Image Segmentation
Noise Reduction
LoLo2
References
LoHi2
LoLo1
HiLo2
LoHi1
Image HiHi2
HiLo1
HiHi1

Wavelet Transform
Eva Hostalkova
Introduction
Fourier Transform
STFT
Wavelet Transform
MRA Thresholding
CWT
Wavelets Properties Noisy wavelet
DWT Image coefficients
DWT and CWT
Subband Coding
WT Applications
Wavelet
Image Segmentation
Noise Reduction
LoLo2
References
LoHi2
LoLo1
HiLo2
LoHi1
Image HiHi2
HiLo1
HiHi1

Wavelet Transform
Eva Hostalkova
Introduction
Fourier Transform
STFT
Wavelet Transform
MRA Thresholding
CWT
Wavelets Properties Noisy wavelet
DWT Image coefficients
DWT and CWT
Subband Coding
WT Applications
Wavelet
Discontinuity Detection Wavelet Synthesis
Image Segmentation
Noise Reduction
LoLo2 LoLo1
References
LoHi2 LoHi1
LoLo1 LoLo2
HiLo2 HiLo1
LoHi1 LoHi2
Image HiHi2 HiHi1 Image
HiLo1 HiLo2
HiHi1 HiHi2

Wavelet Transform
Eva Hostalkova
Introduction
Fourier Transform
STFT
Wavelet Transform
MRA Thresholding
CWT
Wavelets Properties Noisy wavelet Denoised
DWT Image coefficients Image
DWT and CWT
Subband Coding
WT Applications
Wavelet
Image Segmentation
Noise Reduction
LoLo2 LoLo1
References
LoHi2 LoHi1
LoLo1 LoLo2
HiLo2 HiLo1
LoHi1 LoHi2
Image HiHi2 HiHi1 Image
HiLo1 HiLo2
HiHi1 HiHi2

Wavelet Transform
Eva Hostalkova
Introduction
Fourier Transform
STFT
Wavelet Transform
MRA Thresholding
CWT
DWT and CWT
Subband Coding
WT Applications
Wavelet
Image Segmentation
Noise Reduction
References
Wavelet function? Number of levels?

Wavelet Transform
Eva Hostalkova
Introduction
Fourier Transform Thresholding procedure type? Threshold level?
STFT
Wavelet Transform
MRA Thresholding
CWT
DWT and CWT
Subband Coding
WT Applications
Wavelet
Image Segmentation
Noise Reduction
References
Wavelet function? Number of levels?

Wavelet Transform
Eva Hostalkova Noise Reduction by Wavelet Shrinkage

Introduction
(a) ORIGINAL (b) WITH ADD. GAUSSIAN NOISE (c) DENOISED
Fourier Transform
STFT
Wavelet Transform
MRA
CWT
Wavelets Properties
DWT
DWT and CWT
Subband Coding
WT Applications
Image Compression
Image Segmentation
Noise Reduction
References

Wavelet Transform
Eva Hostalkova Noise Reduction by Wavelet Shrinkage

Introduction
(a) ORIGINAL (b) WITH ADD. GAUSSIAN NOISE (c) DENOISED
Fourier Transform
STFT
Wavelet Transform
MRA
CWT
Wavelets Properties
DWT
DWT and CWT
Subband Coding
WT Applications
Image Compression
Image Segmentation
Noise Reduction
References

References
Wavelet Transform
Eva Hostalkova
Introduction S. Mallat.A Wavelet Tour of Signal Processing.

Fourier Transform Academic Press, 1998.
STFT
G. Strang, T. Nguyen. Wavelets and Filter Banks.
Wavelet Transform
MRA
Wellesley-Cambridge Press, 1996.
CWT
Wavelets Properties M. Barni. Document and Image Compression.
DWT Taylor & Francis, 2001.
DWT and CWT
Subband Coding
R. Polikar. The Wavelet Tutorial.
WT Applications http://users.rowan.edu/ polikar/, Iowa State University, 2001.
Image Compression C. Valens. A Really Friendly Guide to Wavelets.
Image Segmentation
Noise Reduction http://pagesperso-orange.fr/polyvalens/, 2003.
References
www.wavelet.org
The MathWorks, Inc. Matlab Help.

www.mathworks.com 2009.

Wavelet Introduction

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Wavelet Introduction

Uploaded by

Copyright:

Available Formats

Wavelet Transform

Dept of Computing and Control Engineering

ATHENS Nov 2009

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 1 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 2 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 2 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 2 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 2 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 2 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 3 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 4 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 4 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 5 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 5 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 6 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 7 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 7 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 8 / 36

∆f = Fs /N . . . frequency resolution (distance between

∆f = Fs /N . . . frequency resolution (distance between

Introduction (a) SIGNAL 2 (b) STFT: WINDOW L=32

WT Applications (c) STFT: WINDOW L=128 (d) STFT: WINDOW L=256

(Using the Hamming window w - gently dropping to zero)

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 11 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 12 / 36

References CWT of the signal x (t):

where * denotes the complex conjugate pair

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 13 / 36

References CWT of the signal x (t):

where * denotes the complex conjugate pair

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 13 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 14 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 15 / 36

Introduction CWT Properties

DWT and CWT

Matrix Interpretation shift of the wavelet Level 3

WT Applications coefficients without Level 4

Figure: Comparison of DWT and nearly invariant complex WT

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 16 / 36

Fourier Transform Not necessary for wavelet analysis (biorthogonality for

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 17 / 36

Eva Hostalkova Wavelet Functions Properties I

Band-pass spectrum (Ψ(ω) vanishes at ω = 0):

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 18 / 36

Eva Hostalkova Wavelet Functions Properties I

Band-pass spectrum (Ψ(ω) vanishes at ω = 0):

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 18 / 36

Introduction Time dilation of ψ causes shift and compression of the

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 19 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 20 / 36

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 21 / 36

Eva Hostalkova Discrete Wavelet Transform (DWT)

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 22 / 36

Eva Hostalkova Discrete Wavelet Transform (DWT)

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 22 / 36

Wavelet Transform 2500

Eva Hostalkova (ICT, Prague) Wavelet Transform Athens 2009 23 / 36

The spectrum of the wavelet function corresponds to a

The spectrum of the wavelet function corresponds to a

Eva Hostalkova Subband Coding Algorithm

Fourier Transform approximation

Wavelet Transform l [n] 2