19ECE457 Wavelets and Application
Pooja Muralidharan
Department of Electronics and Communication Engg.
Amrita School of Engineering, Coimbatore
Amrita Vishwa Vidyapeetham
Syllabus
Unit 1
Introduction to wavelets, Vector space, Functions and function spaces, Continuous time Fourier transforms, Short
time Fourier transforms, The uncertainty principle and time-frequency tiling, Discrete wavelet transforms, Scaling
and Wavelet functions, Filter banks.
Unit 2
Legendre polynomials, Recurrence formula, Laplace’s integral formula, Design of orthogonal wavelet systems,
Bi-orthogonal wavelet, Introduction to lifting scheme, Dealing with signal boundaries, Multi wavelet, Frequency
domain approach, Design of wavelet.
Unit 3
Wavelet in Image Processing, Biomedical applications, Data compression, EZW algorithm, Denoising, Edge
detection, Object isolation, Audio coding, Communication applications, Channel coding, Speckle removal, Image
fusion, Filter design, Signal analysis, Image compression, PDEs, Wavelet transforms on complex geometrical shapes.
2
3
Introduction to Wavelets
● Limitations of Fourier transform and short-time Fourier transform
● Introduction to wavelet theory
● Wavelet transform
○ Continuous wavelet transform (CWT)
○ Discrete wavelet transform (DWT)
○ Filter banks theory
● Types of wavelets
4
The Fourier Transform and its Applications
For CTS: For DTS:
● Fourier transform (FT) decomposes the signal f(t) in infinite number of sine/cosine
waves (harmonics).
● No means of identifying exactly where an event occurs.
● In FT, f(⍵): only frequency vs magnitude is present (Fourier Spectrum) - time
information is missing.
● Does not cope well with discontinuous bursts of signals - video, music, etc.
5
Can you identify the time
information in the Fourier
Spectrum of the original
signal?
6
Fourier Transform Not Suitable for Non-Stationary Signals
● Stationary signals:
○ Signals with frequency content unchanged in time.
○ All frequency components exist at all times.
● Non-stationary signals:
○ Frequency component changes in time.
○ Chirp signal (frequency is continuously increasing).
7
Fourier Spectrum for Stationary and Non-Stationary Signals
No clear visible information of frequency contents! 8
9
Short-Time Fourier Transform
Introduction
● Dennis Gabor (1946) used STFT to analyse only a small section of the signal at a time.
● A technique called windowing the signal is used.
10
Short-Time Fourier Transform
CT-STFT and DT-STFT
● Signal is multiplied with the window and then take a normal Fourier transform.
● There are three axes: time, frequency and magnitude.
● Magnitude is perpendicular to the screen (here in spectrogram indicated as colour
variations).
11
Short-Time Fourier Transform
Time Resolution and Frequency Resolution
Disadvantages of STFT:
- Once the window size if fixed, cannot be changed throughout the analysis.
- Dilemma of resolution:
- Narrow window (good time resolution) →Poor frequency resolution.
- Wide window (poor time resolution) → Good frequency resolution.
12
Short-Time Fourier Transform
Time Resolution and Frequency Resolution
13
Wavelets
A wavelet is a waveform of effectively limited duration that has an average value of zero.
It is defined as:
● a = scale parameter (dilation) → stretch/compress the wavelet in size.
● b = translation parameter (position) → wavelet will shift along the time axis.
14
Wavelets
15
Continuous Wavelet Transform (CWT)
16
Discrete Wavelet Transform (DWT)
● In CWT, calculating wavelet coefficients at every possible scale is a fair amount of
work, and it generates an awful lot of data.
● If scales (a) and positions (b) are chosen to be discrete then analysis will be much
easier and will not generate huge data.
● This idea of choosing discrete values of dilation (a) and translation (b) parameters is
implemented in:
○ Redundant wavelet transform (frames).
○ Orthonormal bases for wavelets or Multiresolution analysis (MRA).
17
Discrete Wavelet Transform - Redundant Wavelet Transform (Frames)
a is chosen to be an integer powers of one fixed dilation parameter a0 > 1, i.e.,
a = a0 m
The different values of m (compress/stretch) correspond to wavelets of different widths.
b = nb0 a0m
where b0 > 0 is fixed. b is discretized and n Ɛ Z
● Narrow wavelets are translated by small steps.
● Wider wavelets are translated by larger steps.
18
Discrete Wavelet Transform - Multiresolution Analysis (MRA)
19
Difference between CWT and DWT
20
Discrete Wavelet Transform - Filter Banks
Signal Analysis
The wavelet decomposition of a signal s(t) based on the MRA theory given by S. Mallat
and Meyer can be done using digital FIR filters as shown in figure:
21
Discrete Wavelet Transform - Filter Banks
Signal Analysis
Multi-level Decomposition:
Here,
S: signal
CD: detailed coefficients
CA: approximation coefficients
If we want to decompose the signal further,the
approximation coefficients are further decomposed in the same manner.
This is called Wavelet Decomposition!
Maximum number of decomposition levels: log2N
22
Discrete Wavelet Transform - Filter Banks
Signal Synthesis
IDWT: Inverse Discrete Wavelet Transform - One level Reconstruction
23
Discrete Wavelet Transform - Filter Banks
Analysis and Synthesis
Single Step Decomposition and Reconstruction:
24
Discrete Wavelet Transform - Filter Banks
Analysis and Synthesis
Multi Step Decomposition and Reconstruction:
25
Discrete Wavelet Transform - Filter Banks
Example of Multi-level Decomposition
26
Various Types of Wavelets
27
Various Wavelet Families Based on their Properties
28
29
