Digital Signal and Image

Processing using MATLAB

Gerard Blanchet
Maurice Charbit
This page intentionally left blank
Digital Signal and Image Processing using MATLAB
This page intentionally left blank
 Digital Signal and Image
Processing using MATLAB

Gerard Blanchet
Maurice Charbit
Part of this book adapted from "Signaux et images sous Matlab : méthodes, applications et
exercices corriges" published in France by Hermes Science Publications in 2001
First published in Great Britain and the United States in 2006 by ISTE Ltd
Translated by Antoine Hervier

Apart from any fair dealing for the purposes of research or private study, or criticism or
review, as permitted under the Copyright, Designs and Patents Act 1988, this publication may
only be reproduced, stored or transmitted, in any form or by any means, with the prior
permission in writing of the publishers, or in the case of reprographic reproduction in
accordance with the terms and licenses issued by the CLA. Enquiries concerning reproduction
outside these terms should be sent to the publishers at the undermentioned address:
ISTE Ltd ISTE USA
6 Fitzroy Square 4308 Patrice Road
London WIT 5DX Newport Beach, CA 92663
UK USA

www.iste.co.uk

© HERMES Science Europe Ltd, 2001

© ISTE Ltd, 2006

The rights of Gerard Blanchet and Maurice Charbit to be identified as the authors of this work
have been asserted by them in accordance with the Copyright, Designs and Patents Act 1988.

Library of Congress Cataloging-in-Publication Data

Blanchet, Gerard.
[Signaux et images sous Matlab. English]
Digital signal and image processing using Matlab / Gerard Blanchet, Maurice Charbit.
p. cm.
Translation of: Signaux et images sous Matlab.
Includes index.
ISBN-13: 978-1-905209-13-2
ISBN-10: 1-905209-13-4
1. Signal processing—Digital techniques—Data processing. 2. MATLAB. I.Charbit, Maurice.
II. Title.
TK5102.9.B545 2006
621.382'2-dc22
2006012690

British Library Cataloguing-in-Publication Data

A CIP record for this book is available from the British Library
ISBN 10: 1-905209-13-4
ISBN 13: 978-1-905209-13-2

Printed and bound in Great Britain by Antony Rowe Ltd, Chippenham, Wiltshire.
 MATLAB is a trademark of The MathWorks, Inc. and is used with per-
mission. The Math Works does not warrant the accuracy of the text or exer-
cises in this book. This book's use or discussion of MATLAB software does
not constitute endorsement or sponsorship by The Math Works of a particular
pedagogical approach or use of the MATLAB software.
This page intentionally left blank
 Contents

Preface 15

Notations and Abbreviations 19

Introduction to MATLAB 23
1 Variables 24
1.1 Vectors and matrices 24
1.2 Arrays 26
1.3 Cells and structures 27
2 Operations and functions 29
2.1 Matrix operations 29
2.2 Pointwise operations 30
2.3 Constants and initialization 31
2.4 Predefined matrices 31
2.5 Mathematical functions 32
2.6 Matrix functions 34
2.7 Other useful functions 34
2.8 Logical operators on boolean variables 35
2.9 Program loops 35
3 Graphically displaying results 36
4 Converting numbers to character strings 39
5 Input/output 39
6 Program writing 40

Part I Deterministic Signals 41

Chapter 1 Signal Fundamentals 43

1.1 The concept of signal 43
1.1.1 A few signals 44
1.1.2 Spectral representation of signals 46
1.2 The Concept of system 48
1.3 Summary 50
8 Digital Signal and Image Processing using MATLAB

Chapter 2 Discrete Time Signals and Sampling 51

2.1 The sampling theorem 52
2.1.1 Perfect reconstruction 52
2.1.2 Digital-to-analog conversion 64
2.2 Plotting a signal as a function of time 65
2.3 Spectral representation 67
2.3.1 Discrete-time Fourier transform ( D T F T ) 67
2.3.2 Discrete Fourier transform (DFT) 71
2.4 Fast Fourier transform 77

Chapter 3 Spectral Observation 81

3.1 Spectral accuracy and resolution 81
3.1.1 Observation of a complex exponential 81
3.1.2 Plotting accuracy of the D T F T 83
3.1.3 Frequency resolution 84
3.1.4 Effects of windowing on the resolution 87
3.2 Short term Fourier transform 90
3.3 Summing up 94
3.4 Application examples and exercises 95
3.4.1 Amplitude modulations 95
3.4.2 Frequency modulation 98

Chapter 4 Linear Filters 101

4.1 Definitions and properties 101
4.2 The z-transform 106
4.2.1 Definition and properties 106
4.2.2 A few examples 107
4.3 Transforms and linear filtering 109
4.4 Difference equations and rational T F filters Ill
4.4.1 Stability considerations 112
4.4.2 FIR and IIR filters 114
4.4.3 Causal solution and initial conditions 115
4.4.4 Calculating the responses 117
4.4.5 Stability and the Jury test 118
4.5 Connection between gain and poles/zeros 119
4.6 Minimum phase filters 129
4.7 Filter design methods 133
4.7.1 Going from the continuous-time filter to the discrete-
time filter 133
4.7.2 FIR filter design using the window method 137
4.7.3 IIR filter design 147
4.8 Oversampling and undersampling 150
4.8.1 Oversampling 151
 Contents 9

4.8.2 Undersampling 155

Chapter 5 Filter Implementation 159

5.1 Filter implementation 159
5.1.1 Examples of filter structures 159
5.1.2 Distributing the calculation load in an FIR filter . . . 164
5.1.3 FIR block filtering 165
5.1.4 FFT filtering 167
5.2 Filter banks 173
5.2.1 Decimation and expansion 174
5.2.2 Filter banks 177

Chapter 6 An Introduction to Image Processing 187

6.1 Introduction 187
6.1.1 Image display, color palette 187
6.1.2 Importing images 191
6.1.3 Arithmetical and logical operations 193
6.2 Geometric transformations of an image 196
6.2.1 The typical transformations 196
6.2.2 Aligning images 199
6.3 Frequential content of an image 203
6.4 Linear filtering 207
6.5 Other operations on images 217
6.5.1 Undersampling 217
6.5.2 Oversampling 217
6.5.3 Contour detection 220
6.5.4 Median filtering 226
6.5.5 M a x i m u m enhancement 227
6.5.6 Image binarization 229
6.5.7 Morphological filtering of binary images 234
6.6 J P E G lossy compression 236
6.6.1 Basic algorithm 236
6.6.2 Writing the compression function 237
6.6.3 Writing the decompression function 240
6.7 Watermarking 241
6.7.1 Spatial image watermarking 241
6.7.2 Spectral image watermarking 244

Part II Random Signals 245

Chapter 7 R a n d o m Variables 247

7.1 R a n d o m phenomena in signal processing 247
7.2 Basic concepts of random variables 248
10 Digital Signal and Image Processing using MATLAB

7.3 Common probability distributions 256

7.3.1 Uniform probability distribution on (a,b) 256
7.3.2 Real Gaussian random variable 257
7.3.3 Complex Gaussian random variable 258
7.3.4 Generating the common probability distributions . . . 259
7.3.5 Estimating the probability density 262
7.3.6 Gaussian random vectors 263
7.4 Generating an r.v. with any type of p.d 265
7.5 Uniform quantization 270

Chapter 8 R a n d o m Processes 273

8.1 Introduction 273
8.2 Wide-sense stationary processes 274
8.2.1 Definitions and properties of WSS processes 275
8.2.2 Spectral representation of a WSS process 278
8.2.3 Sampling a WSS process 285
8.3 Estimating the covariance 289
8.4 Filtering formulae for WSS random processes 296
8.5 MA, AR and A R M A time series 302
8.5.1 Q order MA [Moving Average) process 302
8.5.2 P order A R (Autoregressive) Process 305
8.5.3 The Levinson algorithm 312
8.5.4 A R M A (P,Q) process 315

Chapter 9 Continuous Spectra Estimation 317

9.1 Non-parametric estimation of the PSD 317
9.1.1 Estimation from the autocovariance function 317
9.1.2 Estimation based on the periodogram 320
9.2 Parametric estimation 329
9.2.1 AR estimation 329
9.2.2 Estimating the spectrum of an AR process 337
9.2.3 T h e Durbin m e t h o d of MA estimation 338

Chapter 10 Discrete Spectra Estimation 341

10.1 Estimating the amplitudes and the frequencies 341
10.1.1 The case of a single complex exponential 341
10.1.2 Real harmonic mixtures 343
10.1.3 Complex harmonic mixtures 345
10.2 Periodograms and the resolution limit 347
10.3 High resolution methods 358
10.3.1 Periodic signals and recursive equations 358
10.3.2 The Prony method 363
10.3.3 The MUSIC algorithm 366
10.3.4 Introduction to array processing 379
 Contents 11

Chapter 11 The Least Squares Method 389

11.1 The projection theorem 389
11.2 The least squares method 393
11.2.1 Formulating the problem 393
11.2.2 The linear model 394
11.2.3 The least squares estimator 395
11.2.4 The RLS algorithm (recursive least squares) 402
11.2.5 Identifying the impulse response of a channel 405
11.3 Linear predictions of the WSS processes 407
11.3.1 Yule-Walker equations 407
11.3.2 Predicting a WSS harmonic process 408
11.3.3 Predicting a causal A R - P process 411
11.3.4 Reflection coefficients and lattice filters 412
11.4 Wiener filtering 417
11.4.1 Finite impulse response solution 419
11.4.2 Gradient algorithm 420
11.4.3 Wiener equalization 427
11.5 The LMS (least mean square) algorithm 430
11.5.1 The constant step algorithm 430
11.5.2 The normalized LMS algorithm 439
11.5.3 Echo canceling 442
11.6 Application: the K a i m a n algorithm 446
11.6.1 The K a i m a n filter 446
11.6.2 The vector case 449

Chapter 12 Selected Topics 451

12.1 Simulation of continuous-time systems 451
12.1.1 Simulation by approximation 451
12.1.2 Exact model simulation 452
12.2 Dual Tone Multi-Frequency (DTMF) 455
12.3 Speech processing 461
12.3.1 A speech signal model 461
12.3.2 Compressing a speech signal 468
12.4 D T W 471
12.5 Modifying the duration of an audio signal 474
12.5.1 PSOLA 475
12.5.2 Phase vocoder 477
12.6 Quantization noise shaping 478
12.7 Elimination of the background noise in audio 482
12.8 Eliminating the impulse noise 484
12.8.1 The signal model 484
12.8.2 Click detection 485
12.8.3 Restoration 488
12 Digital Signal and Image Processing using MATLAB

12.9 Tracking the cardiac r h y t h m of the fetus 490

12.9.1 Objectives 490
12.9.2 Separating the EKG signals 491
12.9.3 Estimating cardiac rhythms 494
12.10 Extracting the contour of a coin 501
12.11 Principal component analysis (PCA) 503
12.11.1 Determining the principal components 503
12.11.2 2-Dimension P C A 507
12.11.3 Linear discriminant analysis (LDA) 509
12.12 Separating an instantaneous mixture 514
12.13 Matched filters in radar telemetry 516
12.14 Kaiman filtering 518
12.15 Compression 524
12.15.1 Scalar quantization 524
12.15.2 Vector quantization 526
12.16 Digital communications 538
12.16.1 Introduction 538
12.16.2 8-phase shift keying (PSK) 541
12.16.3 PAM modulation 543
12.16.4 Spectrum of a digital signal 545
12.16.5 The Nyquist criterion in digital communications . . . 549
12.16.6 The eye pattern 555
12.16.7 PAM modulation on the Nyquist channel 556
12.17 Linear equalization and the Viterbi algorithm 562
12.17.1 Linear equalization 564
12.17.2 The Viterbi algorithm 566

Part III Hints and Solutions 571

Chapter 13 Hints and Solutions 573

HI Signal fundamentals 573
H2 Discrete time signals and sampling 573
H3 Spectral observation 579
H4 Linear filters 590
H5 Filter implementation 610
H6 An Introduction to image processing 614
H7 R a n d o m variables 641
H8 R a n d o m processes 646
H9 Continuous spectra estimation 656
H10 Discrete spectra estimation 661
H l l The least squares method 668
H12 Selected topics 676
 Contents 13

Chapter 14 Appendix 727

Al Fourier transform 727
A2 Discrete time Fourier transform 728
A3 Discrete Fourier transform 729
A4 z-Transform 730
A5 Jury criterion 732
A6 F F T filtering algorithms revisited 734

Bibliography 739

Index 747
This page intentionally left blank
 Preface

A practical approach through simulation

Simulation is an essential tool in any field related to engineering techniques,
whether it is used for teaching purposes or in research and development.
When teaching technical subjects, lab works play an important role, as im-
portant as exercise sessions in helping students assimilate theory. The recent
introduction of simulation tools has created a new way to work, halfway be-
tween exercise sessions and lab works. This is particularly the case for digital
signal processing, for which the use of the MATLAB® Ian guage, or its clones,
has become inevitable. Easy to learn and to use, it makes it possible to quickly
illustrate a concept after introducing it in a course.
As for research and development, obtaining and displaying results often
means using simulation programs based on a precise "experimental protocol",
as it would be done for actual experiments in chemistry or physics.
These characteristics have led us, in a first step, to try to build a set of exer-
cises with solutions relying for the most part on simulation; we then a t t e m p t e d
to design an introductory course mostly based on such exercises. Although this
solution cannot replace the traditional combination of lectures and lab works,
we do wonder if it isn't just as effective when associated with exercise sessions
and a few lectures. There is of course no end in sight to the debate on educa-
tional methods, and the amount of experiments being conducted in universities
and engineering schools shows the tremendous diversity of ideas in the m a t t e r .

Basic concepts of D S P
The recent technical evolutions, along with their successions of technological
feats and price drops have allowed systems based on micro-controllers and
microprocessors to dominate the field of signal and image processing, at the
expense of analog processing. Reduced to its simplest form, signal processing
amounts to manipulating d a t a gathered by sampling analog signals. Digital
16 Digital Signal and Image Processing using MATLAB

Signal and Image Processing, or DSIP, can therefore be defined as the art of
working with sequences of numbers.

The sampling theorem

The sampling theorem is usually the first element found in a DSIP course, be-
cause it justifies the operation by which a continuous time signal is replaced
by a discrete sequence of values. It states t h a t a signal can be perfectly recon-
structed from the sequence of its samples if the sampling frequency is greater
t h a n a fundamental limit called the Nyquist frequency. If this is not the case,
it results in an undesired effect called spectrum aliasing.

Numerical Sequences and D T F T

The Discrete Time Fourier Transform, or D T F T , introduced together with

the sampling theorem, characterizes the spectral content of digital sequences.
The analogy between the D T F T and the continuous time Fourier transform is
considered, with a detailed description of its properties: linearity, translation,
modulation, convolution, the Parseval relation, the Gibbs phenomenon, ripples
caused by windowing, etc.
In practice, signals are only observed for a finite period of time. This
"time truncation" creates ripples in the spectrum and makes it more difficult
to the separate two close frequencies in the presence of noise. This leads to
the concept of frequency resolution. The D T F T is a simple way of separating
two frequencies, but only if the observation time is greater t h a n the inverse of
the difference between the two frequencies. The frequency resolution will allow
us to introduce the reader to weighting windows. However, a more complete
explanation of the concept of resolution can only be made if noise disturbing
the signal is taken into account, which is why it will be studied further when
random processes are considered.
The Discrete Fourier Transform, or D F T is the tool used for a numerical
computation of the D T F T . Because this calculation involves a finite number
of frequency values, the problem of precision has to be considered. There are
a few differences in properties between the D F T and the D T F T , particularly
regarding the indexing of temporal sequences t h a t are processed modulo N.
Some examples of this are the calculation of the D T F T and the D F T of a
sinusoid, or the relation between discrete convolution and the D F T . At this
point, the fast algorithm calculation of the D F T , also called F F T (Fast Fourier
Transform), will be described in detail.

F i l t e r i n g a n d E l e m e n t s of F i l t e r D e s i g n

Linear filtering was originally used to extract relevant signals from noise. The
basic tools will be introduced: the discrete convolution, the impulse response,
 Preface 17

the frequency response, the z-transform. We will then focus on the fundamen-
tal relation between linear filtering with rational transfer functions and linear
constant-coefficient recursive equations.
Filter design is described based on a few detailed examples, particularly the
window method and the bilinear transform. The concepts of over-sampling and
under-sampling are then introduced, some applications of which are frequency
change and the reduction of quantization noise. From a broader perspective,
multi-rate processing and filter banks which are described here, are two subjects
t h a t attract a lot of attention in the field of DSIP.

An introduction to images

Image processing is described in its own separate chapter. Many of the concepts
used in signal processing are also used in image processing. The only difference
is t h a t two indices are used instead of one. However images have particular
characteristics t h a t require specific processing: erosion, expansion, etc. The
computation time is usually much longer for images t h a n it is for signals. It is
nevertheless possible to conduct image processing with MATLAB or one of
its clones. This theme will be discussed using examples on 2D filtering, contour
detection, and other types of processing in cases where the 2D nature of the
images does not make t h e m too different from a I D signal. This chapter will
also be the opportunity to discuss image compression and entropie coding.

Random Processes

Up until now, the signals used as observation models have been described by
functions t h a t depend on a finite number of well known parameters and on
simple known basic functions: the sine function, the unit step function, the
impulse function... This type of signal is said to be deterministic.
There are other situations where deterministic functions cannot provide us
with a relevant apprehension of the variability of the phenomena. Signals must
then be described by characteristics of a probabilistic nature. This requires
the use of random processes, which are time-indexed sequences of random vari-
ables. Wide sense stationary processes, or WSSP, are an important category
of random processes. The study of these processes is mainly based on the es-
sential concept of power spectral density, or P S D . The PSD is the analog for
W S S P of the square module of the Fourier transform for deterministic signals.
The formulas for the linear filtering of W S S P are then laid down. Thus, we
infer t h a t WSSPs can also be described as the linear filtering of a white noise.
This result leads to a large class of stationary processes: the AR process, the
MA process, and the A R M A process.
18 Digital Signal and Image Processing using MATLAB

Spectral Estimation

One of the main problems DSIP is concerned with is evaluating the PSD of
WSSPs. In the case of continuous spectra, it can be solved by using non-
parametric approaches (smooth periodograms, average periodograms, etc.) or
parametric methods based on linear models (AR, MA, A R M A ) . As for line
spectra, the most commonly used methods are the periodogram and what are
called high resolution methods, which use the structures of the signal and the
noise: Prony, Pisarenko, MUSIC, ESPRIT, etc.

T h e least squares

This chapter discusses the use of the least squares method for solving problems.
This method is used in a number of problems, in fields such as spectral analysis,
modelling, linear prediction, communications... We will discuss such methods
as Wiener, RLS, LMS, K a i m a n . . .

Applications

This last chapter presents case studies t h a t go a little further in depth t h a n the
examples described earlier. The emphasis is set on audio signal processing, on
compression as well restoring and denoising for speech and music, and on mod-
ulation, demodulation and equalization issues for digital communications. This
chapter is also an opportunity to discover typical approaches and algorithms:
pitch detection, PSOLA, D T W , ACP, LBG, V i t e r b i . . .

As a Conclusion
One of the issues raised by many of those who use signal processing has to
do with the artificial aspect introduced by simulation. For example, we use
sampling frequencies equal to 1, and therefore frequencies with no dimension.
There is a risk t h a t the student may lose touch with the physical aspect of
the phenomena and, because of t h a t , fail to acquire the intuition of these
phenomena. T h a t is why we have tried, at least in the first chapters, to give
exercises t h a t used values with physical units: seconds, Hz, etc.
This work discusses important properties and theorems, but its objective
is not to be a book on mathematics. Its only claim, and certainly an excessive
one, is to show how interesting signal and image processing can be, by providing
themes of study we chose because they were good examples, because they were
simple, while trying not to be too trivial.
All of the subjects discussed far from cover the extent of knowledge required
in this field. However they seem to us to be a solid foundation for an engineer
who would happen to deal with DSIP problems.
 Notations and Abbreviations

0 E m p t y Set

( 1 when \t\ < T/2

rect^i)
I 0 otherwise
sin(7r;c)
πχ
, I l when x E A . . .
t(x E A) = < . (Indicator Function oí A)
0 otherwise
(a,b] = {x : a < x < b}

Í
Dirac Distribution when í £ R
S(t)
Kronecker Symbol when Í G Z
Re(z) Real P a r t of z
Im(z) Imaginary Part of z
i or j = v—1
x(t) ^± X(f) Fourier Transform
(x*y)(t) Continuous Time Convolution

= / x(u)y(t — u)du
JR
(x*y)(t) Discrete Time Convolution

= Σ x{u)y{t - w)
uez
20 Digital Signal and Image Processing using MATLAB

Ijv (N x 7V)-dimension Identity Matrix

A* Complex Conjugate of A
Ατ Transpose of A
H
A Transpose-Conjugate of A
A"1 Inverse Matrix of A

P{X e A) Probability t h a t X E A
Έ{Χ} Expectation Value of X
Xc = X -Έ{Χ} Zero-mean R a n d o m Variable
2
var(X)=E{|Xc|} Variance of X
Έ{Χ\Υ} Conditional Expectation of X given Y

ADC Analog to Digital Converter

ADPCM Adaptive Differential P C M
AMI Alternate Mark Inversion
AR Autoregressive
ARMA AR and MA
BER Bit Error Rate
bps Bits per second
cdf Cumulative distribution function
CF Clipping Factor
CZT Causai z-Transform
DAC Digital to Analog Converter
DCT Discrete Cosine Transform
d.e./de Difference equation
DFT Discrete Fourier Transform
DTFT Discrete Time Fourier Transform
DTMF Dual Tone Multi-Frequency
dsp Digital signal processing/processor
e.s.d./esd Energy spectral density
FIR Finite Impulse Response
FFT Fast Fourier Transform
FT Continuous Time Fourier Transform
 Notations and Abbreviations 21

HDB High Density Bipolar

IDFT Inverse Discrete Fourier Transform
i.i.d./iid Independent and Identically Distributed
IIR Infinite Impulse Response
ISI InterSymbol Interference
LDA Linear discriminant analysis
1ms Least mean squares
MA Moving Average
MAC Multiplication A c c u m u l a t i o n
OTF Optical Transfer Function
PAM Pulse Amplitude Modulation
PCA Principal Component Analysis
p.d. Probability Distribution
ppi Points per Inch
p.s.d./PSD Power Spectral Density
PSF Point Spread Function
PSK Phase Shift Keying
QAM Quadrature Amplitude Modulation
rls Recursive least squares
rms Root mean square
r.p./rp R a n d o m process
SNR Signal to Noise Ratio
r.v./rv R a n d o m variable
STFT Short Term Fourier Transform
TF Transfer Function
wss Wide (Weak) Sense Stationary (Second Order) Process
ZOH Zero-Order Hold
ZT z-Transform
This page intentionally left blank
 Introduction to MATLAB

In this book the name MATLAB} ® (short for Matrix Laboratory) will refer to:

— the program launched by using the c o m m a n d matlab in Dos or Unix

environments, or by clicking on its icon in a graphic environment such as
x l l , Windows, M a c O S . . . ,

— or the language defined by a vocabulary and syntax rules.

(R) · ·

MATLAB is an interpreter, t h a t is to say a program t h a t remains in

the computer's memory once it is launched. MATLAB® displays a com-
m a n d window used for interpreting commands. If they are considered correct,
MATLAB will execute them. This execution will itself lead to verifications.

E x a m p l e 1 ( D i r e c t i n t e r p r e t a t i o n ) Type a = 2 * l o g l 0 ( 5 ) then < r e t u r n > .

The result is shown in a P C environment (Figure 1).

■WWH.lJi.HUILI.LII
File Edit Options Windows Help
►a=2*logl0(5)
■ Command line

Result

Prompt

(R)
Figure 1 - The MATLAB command window on MS-Windows

C o m m a n d s can be gathered together in text files called matlab programs.

The user gives t h e m a name t h a t can be called from the prompt line. The
MATLAB documentation explains how to use an editor to create such files.
24 Digital Signal and Image Processing using MATLAB

This editor may either be integrated in the software or kept external (the
user's favorite editor). Program files use the extension .m. If a program is
called p r o g l . m , all the user has to do is type p r o g l in the MATLAB com-
(R)

m a n d window to have it executed. MATLAB then searches for the file in

the routine directory. If it doesn't find the file there, it looks for p r o g l . m in
the various files specified in the directory p a t h . The latter can be defined di-
rectly in the c o m m a n d prompt window, or by using a program and executing
commands such as path, addpath, rmpath, genpath, p a t h t o o l , s a v e p a t h (see
documentation, online help, or type h e l p path).

Eile Edit View Graphics Dej]ug Desktop Window Help

D GÌ | A ^ E Ό « I DP E Í I ? I Current Director!: 'usrflocaUmatlapJmatlab7r14sp2Jbin / | ... IS

Shortcuts Ξ HowtoAdd tu What's New

Command Window

<MATLAB>
CurrentDirectory| Workspace\ Copyright 19S4-2005 The MathWorks, Inc.
Vepsion 7.0.4.352 (R14) Service Pack 2
Command History
* - 4Í7ÍU5 4:34 PM -%
Γ** - 4J8Í05 3:29 AM -%
1-11 To get slatted, select MATLAB Help or Demos fpom the Help menu.

•»Sari |

(R)
F i g u r e 2 — The MATLAB window in an X-windows environment. The definition
of the routine folder can be done directly by clicking on the icon with ". . . " in the
top-right corner of the window. The definition of the directory path can be done by
selecting the item s e t path . . . in the menu f i l e

Clones of MATLAB are now available. Some belong to the public domain.
There also exists a compiler t h a t allows the user to translate MATLAB pro-
grams in machine language, making the execution quicker, and meaning t h a t
it is not required to own the interpreter.

1 Variables
1.1 Vectors and matrices

The MATLAB® Ian guage is dedicated to m a t r i x calculations and was opti-

mized in this perspective. The variables handled as a priority are real or com-
plex matrices. A scalar is a 1 x 1 matrix, a column vector is a m a t r i x with only
one column, and a Une vector a m a t r i x with only one line.
The notation (£ x c) indicates t h a t the considered variable has £ lines and
c columns.
 Introduction to MATLAB 25

E x a m p l e 2 ( A s s i g n m e n t of a real m a t r i x ) Type a = [ l 2 3 ; 4 5 6] at
the MATLAB prompt in the c o m m a n d window. The answer is shown in
Figure 3.

D « Command ^ ^ ^ ^ = ^ ^ ^ ^ ^ ¡EIE
»a=| 1 2 3;4 5 6]

■
Π
a = ^~"^- Assignment of matrix a

1
1
1 2 3 -* Result
4 5 6 (2 lines, 3 columns)

F i g u r e 3 — Assigning a matrix

Values are assigned to the elements of a m a t r i x by using brackets. A space

(or a comma) is a separator, and takes you to the next column, while the semi-
colon takes you to the next line. E l e m e n t s are i n d e x e d s t a r t i n g f r o m 1.
The first index is the line number, the second one is the column number. In
our example, a ( l , 1) = 1 and a ( 2 , 1 ) = 4 . The assignment a = [ l 2 ; 3 4 5] will of
course lead to an error message, since the number of columns is different for
the first and second lines.
Character strings can also be assigned to the elements of a matrix. However,
the string length must be compatible with the structure of the matrix. For
example, N = [ ' p a u l ' ; ' J o h n ' ] would be correct, whereas N = [ ' p a u l ' ; ' p e t e r ' ]
would cause an error.
When the vector's components form a sequence of values separated by reg-
ular intervals, it is easier to use what is called an "implicit" loop of the type
(indD : s t e p : indF). This expression refers to a list of values starting at indD
and going up to indF by increments of s t e p . Values cannot go beyond indF.
The increment value s t e p can be omitted if it is equal to 1.

E x a m p l e 3 ( I m p l i c i t e n u m e r a t i o n ) Type a=(0:l:10) a=(0:10).

MATLAB® returns:

2 3 4 5 6 7 8 9 10

E x a m p l e 4 ( I n c r e m e n t e d i m p l i c i t e n u m e r a t i o n ) Type a = ( 0 : 4 : 1 0 ) .
MATLAB® returns:

0 4 8
26 Digital Signal and Image Processing using MATLAB

The last element of a vector is indicated by the reserved word end. In the
previous example, a ( e n d ) indicates t h a t its value is 8.
It is possible to extend the size of a matrix. The interpreter takes care of
available space by dynamically allocating memory space during the analysis of
the typed phrase.

E x a m p l e 5 ( E x t e n s i o n of m a t r i x ) Type the following commands one after

the other:
» a = [ i 2 3; 4 5 6]
a =
1 2 3
4 5 6
» a = [ a a]
a =
1 2 3 1 2 3
4 5 6 4 5 6
» a = [ l 2 3; 4 5 6] ;
>>a=[a;a]
a =
1 2 3
4 5 6
1 2 3
4 5 6
COMMENTS:

— When defining variables and objects, the language takes into account
whether letters are capital or lowercase.

— Typing ";" at the end of a c o m m a n d line prevents the program from

displaying the results of an operation.

— The display format can be modified by using the f o r m a t command. Exe-

cuting f o r m a t l o n g , for example, changes the number of significant digits
from 5 to 15.

— The user must bear in mind t h a t MATLAB® dedicates memory space

every time a variable is used for the first time. All of the variables used
during a work session are stored in the computer's memory, which means
it is necessary to free space from time to time so as not to get the OUT
OF MEMORY error message (see the c l e a r c o m m a n d in the documentation
or type h e l p c l e a r ) .

1.2 Arrays

Multidimensional arrays (not supported by all versions) are an extension of the

normal two-dimensional matrix. One way to create such an array is to start
with a 2-dimension m a t r i x t h a t already exists and to extend it. Type:
 Introduction to MATLAB 27

A=[l:3;4:6]
» A
A=
1 2 3
4 5 6
» A(: , : ,2)=zeros(2,3) , '/„ or A ( : , : , 2 ) = 0
A(:,:,l) =
1 2 3
4 5 6
A(:,:,2) =
0 0 0
0 0 0

The r e p m a t and c a t functions are provided in order to build multidimen-

sional arrays.

1.3 Cells a n d s t r u c t u r e s

In the most recent versions of MATLAB®, there are two groups of d a t a t h a t

are more elaborate t h a n scalar arrays and character string arrays: the first one
is called a cell and the second a structure.

In an array of cells, the elements can be of any nature, numerical value,

character string, array, etc. Type:

langcell={'MATLAB', [ 6 . 5 ; 2 . 3 ] , 2 0 0 2 }
» langcell(2)
ans =
[2x1 double]
» langcell{2}
ans =
6.5000
2.3000
» langcell{2}(l)
ans =
6.5000
l a n g c e l l is made up of three elements: the first one is a character string,
the second one is a column vector, and the third one is a scalar. This
example shows the difference in syntax between an array and a cell, a left
brace ({) and a right brace (}) being used instead of a left square bracket
([) and a right square bracket (]). As for the content, l a n g c e l l ( 2 ) refers
to the vector [6 . 5000; 2 . 3] , l a n g c e l l { 2 } to the content of this vector, and
l a n g c e l l { 2 } ( l ) to the numerical value 6.5.

A structure is defined by the s t r u c t instruction. The following exam-

ple defines a structure, called l a n g s t r u c , comprising three fields: Language,
V e r s i o n , and Year. The instruction assigns the character string MATLAB to the
28 Digital Signal and Image Processing using MATLAB

first field, the character string 6 . 5 to the second field, and the numerical value
2002 to the third field:
» l a n g s t r u c = s t r u c t C L a n g u a g e ' , 'MATLAB' , 'Version' , ' 6 . 5 ' , 'Year' ,2002) ;
>>langstruc.Year
ans =
2002
»
The second instruction displays the content of l a n g s t r u c . Year, which is
2002. A 1 x 1 dimension structure is organized in the same way as an n x 1
dimension array of cells, where n is the number of fields of the structure. Cells
can therefore be compared to structures with unnamed fields.
The following example defines a structure named l a n g s t r u c , comprised of
two recordings. Each recording contains all three fields Language, V e r s i o n ,
and Year to which were respectively assigned the sequences of two character
strings MATLAB and C, of the two values 6.5 and 15.1, and of the two values
2002 and 2003:
» langstruc=struct('Langage',{{'MATLAB','C'}},...
'Version' , [6.5;15.1],'Year',[2002;2003] );
>> l a n g s t r u c
langstruc =
Language: {'MATLAB' 'C>
Version: [2x1 double]
Year: [2x1 double]
>> langstruc.Langage{l}
ans =
MATLAB
>> langstruc.Language(1)
ans =
'MATLAB'
»
These objects can be handled using certain functions: isstruct,
fieldnames, s e t f i e l d , rmfield, c e l l f u n , c e l l d i s p , num2cell, cell2mat,
c e l l 2 s t r u c t , s t r u c t 2 c e l l . . . An example of a conversion is as follows:
>> c l e a r a l l
» langcell={'MATLAB',[6.5;2.3],2002}
>> chps={'Langage','Version','Year'};
>> cell2structClangceli,chps,2)
ans =
Language: 'MATLAB'
Version: 6.5000
Year: 2002
»
The 2 t h a t is part of the instruction c e l l 2 s t r u c t C l a n g c e l i , chps , 2 ) indi-
cates the dimension of l a n g c e l l t h a t needs to be taken into account to define
the number of fields. Here, for example, s i z e ( l a n g c e l l , 2 ) means t h a t the
number of fields is 3.