Richards J.A., Jia X. Remote Sensing Digital Image Analysis: An Introduction
Подождите немного. Документ загружается.
John A. Richards · Xiuping Jia
Remote Sensing Digital Image Analysis
John A. Richards · Xiuping Jia
Remote Sensing Digital
Image Analysis
An Introduction
4th Edition
With 197 Figures
John A. Richards
The Australian National University
Research School of Information Sciences
and Engineering (RSISE)
Dept. of Systems Engineering
Canberra, ACT 0200
Australia
John.Richards@anu.edu.au
Xiuping Jia
The University of New South Wales
Australian Defence Force Academy
School of Information Technology and Electrical Engineering
Northcott Drive Cambell ACT 2600
Australia
x-jia@adfa.edu.au
ISBN-10 3-540-25128-6 Springer Berlin Heidelberg New York
ISBN-13 978-3-540-25128-6 Springer Berlin Heidelberg New York
This work is subject to copyright. All rights are reserved, whether the whole or part of the material
is concerned, specifi cally the rights of translation, reprinting, reuse of illustrations, recitation, broad-
casting, reproduction on microfi lm or in other ways, and storage in data banks. Duplication of this
publication or parts thereof is permitted only under the provisions of the German Copyright Law of
September 9, 1965, in its current version, and permission for use must always be obtained from Sprin-
ger. Violations are liable to prosecution under German Copyright Law.
Springer is a part of Springer Science+Business Media
springeronline.com
© Springer-Verlag Berlin Heidelberg 2006
Printed in Germany
The use of general descriptive names, registered names, trademarks, etc. in this publication does not
imply, even in the absence of a specifi c statement, that such names are exempt from the relevant pro-
tective laws and regulations and therefore free for general use.
Typesetting: PTP-Berlin Protago-T
E
X-Production GmbH, Germany
Final processing by PTP-Berlin Protago-T
E
X-Production GmbH, Germany
Cover-Design: Struve & Partner, Heidelberg
Printed on acid-free paper 62/3141/Yu – 5 4 3 2 1 0
Library of Congress Control Number: 2005926341
Front Cover
Image of the city of Canberra, the national capital of Australia, recorded by the HyMap scan-
ner manufactured by Integrated Spectronics Pty Ltd, Sydney, Australia
Preface to the Fourth Edition
This fourth edition has been developed to reflect the changes that have occurred in
techniques for the analysis of digital image data in remote sensing over the past five
years or so. Its focus is on those procedures that seem now to have become part
of the set of tools regularly used to perform thematic mapping. As with previous
revisions, the fundamental material has been preserved in its original form because
of its tutorial value; its style has been revised in places and it has been supplemented
if newer aspects have emerged in the time since the third edition appeared. The theme
of the book remains, however, on the needs of the senior student and practitioner.
The earlier editions have contained extensive material in Chapter 1 on satel-
lite programs and sensor characteristics. Although that material is important in the
context of understanding the application of image analysis procedures, the rapid de-
velopment of quasi- and fully operational programs in the past decade has meant that
expanding the material of Chap. 1, as required, would have distracted from the role
of that chapter to introduce the reader to the nature and properties of digital image
data in remote sensing. Accordingly, all of the material on satellite programs and
sensor specifications has been moved to a new Appendix A. Chapter 1 has then been
completely rewritten as a stand-alone introduction to sensors in general and the data
properties of importance in image analysis.
Many changes have been made throughout the book to meet the increasing em-
phasis on hyperspectral data and its analysis. Although much of that is contained in
Chap. 13, techniques required for hyperspectral data processing are developed in the
context of previous chapters, particularly new material on feature extraction tools
that work well on hyperspectral data sets; they are covered in Chap. 10.
Chapter 10 has been re-named. It was felt that there is too much confusion in the
term Data Fusion to retain it as the title for material that is fundamentally concerned
with thematic mapping from multiple data sources and multiple sensors.
Chapter 8, dealing with supervised classification methods, has been substantially
supplemented. Sections have been incorporated on k nearest neighbour classification,
Markov random fields and support vector classifiers.
VI Preface to the Fourth Edition
Other modifications relate to the noise adjusted principal components transfor-
mation, the definition of texture, a re-working of the contrast modification material
and the inclusion of several further illustrations and problems.
The authors continue to enjoy the very strong support and understanding of their
families, so important in the undertaking of this work, for which they express and
record their sincere gratitude.
Canberra, Australia, January 2005 John A. Richards
Xiuping Jia
Preface to the Third Edition
In the time since the second edition of this text was produced two significant trends
have been apparent. First, access to image processing technology has continued to
improve significantly with most students and practitioners now having readily avail-
able inexpensive workstations and powerful software for analysing and manipulating
image data.
The second change has been the dramatic increase in the numbers of satellite,
aircraft and sensor programs. Perhaps most significant is the widespread availability
of hyperspectral data and the special challenges presented by that data type for
information extraction.
Accordingly, this third edition has been written to reflect those trends while, at
the same time, preserving the important elements of image processing and analysis
algorithms of significance in remote sensing applications.
The major changes between the previous edition and this are an update of Chap. 1,
and the introduction of a new Chap. 13 dealing with methods for analysing hyper-
spectral data sets.
Chapter 12 has also been significantly altered to provide a focus on the inter-
pretation of data sets that are mixed and could contain, for example, different types
of imagery along with other spatial data types found in a geographic information
system. The previous knowledge-based material has been retained but material orig-
inally covered in Chap. 8 dealing with multi-source data analysis has been combined
with knowledge-based methods to create this chapter on data fusion.
The authors wish to express their appreciation to their colleagues for the assis-
tance they have received in preparing this new edition. In particular David Landgrebe
of Purdue University remains a great supporter and both authors have had the benefit
of working with Dave over the years, including periods spent at Purdue.
The authors are also enormously grateful to their families for their understanding
and support in making the completion of this third edition possible.
Canberra, Australia, August 1998 John A. Richards
Xiuping Jia
Preface to the Second Edition
Possibly the greatest change confronting the practitioner and student of remote sens-
ing in the period since the first edition of this text appeared in 1986 has been the enor-
mous improvement in accessibility to image processing technology. Falling hardware
and software costs, combined with an increase in functionality through the develop-
ment of extremely versatile user interfaces, has meant that even the user unskilled in
computing now has immediate and ready access to powerful and flexible means for
digital image analysis and enhancement. An understanding, at algorithmic level, of
the various methods for image processing has become therefore even more impor-
tant in the past few years to ensure the full capability of digital image processing is
utilised.
This period has also been a busy one in relation to digital data supply. Several
nations have become satellite data gatherers and providers, using both optical and
microwave technology. Practitioners and researchers are now faced, therefore, with
the need to be able to process imagery from several sensors, together with other forms
of spatial data. This has been driven, to an extent, by developments in Geographic
Information Systems (GIS) which, in turn, have led to the appearance of newer image
processing procedures as adjuncts to more traditional approaches.
The additional material incorporated in this edition addresses these changes. First,
Chap. 1 has been significantly revised to reflect developments in satellite and sensor
programs. Removal of information on older systems has been resisted since their
data is part of an important archive which still finds value, particularly for historical
applications.
Chapter 8, dealing with supervised classification methods, has been substantially
increased to allow context classification to be addressed, along with techniques, such
as evidential processing and neural networks, which show promise as viable image
interpretation tools. While the inclusion of these topics has caused the chapter to
become particularly large in comparison to the others, it was felt important not to
separate the material from the more traditional methods. The chapter is now pre-
sented therefore in two parts – the first covers the standard supervised classification
procedures and the second the new topics.
X Preface to the Second Edition
A departure from the application of classical digital image processing to remote
sensing over the last five years has been the adoption of knowledge-based methods,
where qualitative rather than quantitative reasoning is used to perform interpretations.
This is the subject of a new chapter which seeks to introduce reasoning based on
knowledge as a means for single image interpretation, and as an approach that can
deal successfully with the mixed spatial data types of a GIS.
Besides these changes, the opportunity has been taken to correct typographical
and related errors in the first edition and to bring other material up to date.
As with the first edition, the author wishes to record his appreciation to others
for their support and assistance. Ashwin Srinivasan, now with the Turing Institute in
Glasgow, as a very gifted graduate student, developed much of the material on which
Chap. 12 is based and, during his time as a student, helped the author understand the
mechanisms of knowledge processing. He also kindly read and offered comments
on that chapter. The author’s colleague Don Fraser similarly read and provided com-
ments on the material on neural networks. Philip Swain and David Landgrebe of
Purdue University continue to support the author in many ways: through their feed-
back on the first edition, their interaction on image processing, and Dave’s provision
of MultiSpec, the Macintosh computer version of the LARSYS software package.
Finally, the author again expresses gratitude to his family for their constant support,
without which the energy and enthusiasm needed to complete this edition might not
have been found.
Canberra, Australia, March 1993 John A. Richards
Preface to the First Edition
With the widespread availability of satellite and aircraft remote sensing image data
in digital form, and the ready access most remote sensing practitioners have to com-
puting systems for image interpretation, there is a need to draw together the range of
digital image processing procedures and methodologies commonly used in this field
into a single treatment. It is the intention of this book to provide such a function,
at a level meaningful to the non-specialist digital image analyst, but in sufficient
detail that algorithm limitations, alternative procedures and current trends can be
appreciated. Often the applications specialist in remote sensing wishing to make use
of digital processing procedures has had to depend upon either the mathematically
detailed treatments of image processing found in the electrical engineering and com-
puter science literature, or the sometimes necessarily superficial treatments given in
general texts on remote sensing. This book seeks to redress that situation.
Both image enhancement and classification techniques are covered making the
material relevant in those applications in which photointerpretation is used for infor-
mation extraction and in those wherein information is obtained by classification. It
grew out of a graduate course on digital image processing and analysis techniques
for remote sensing data given annually since 1980 at the University of New South
Wales. If used as a graduate textbook its contents with the exception of Chap. 7 can
be covered substantially in a single semester. Its function as a text is supported by
the provision of exercises at the end of each chapter. Most do not require access to
a computer for solution. Rather they are capable of hand manipulation and are in-
cluded to highlight important issues. In many cases some new material is introduced
by means of these exercises.
Each chapter concludes with a short critical bibliography that points to more
detailed treatments of specific topics and provides, where appropriate, comment on
techniques of marginal interest to the mainstream of the book’s theme.
Chapter 1 is essentially a compendium of data sources commonly encountered in
digital form in remote sensing. It is provided as supporting material for the chapters
that follow, drawing out the particular properties of each data source of importance.
The second chapter deals with radiometric and geometric errors in image data and
XII Preface to the First Edition
with means for correction. This also contains material on registration of images to
maps and images to each other. Here, as in all techniques chapters, real and modelled
image data examples are given. Chapter 3 establishes the role of computer processing
both for photointerpretation by a human analyst and for machine analysis. This may
be skipped by the remote sensing professional but is an important position chapter
if the book is to be used in teaching.
Chapters 4 and 5 respectively cover the range of radiometric and geometric en-
hancement techniques commonly adopted in practice, while Chap. 6 is addressed to
multispectral transformations of data. This includes the principal components trans-
formation and image arithmetic. Chapter 7 is given over to Fourier transformations.
This material is becoming more important in remote sensing with falling hardware
costs and the ready availability of peripheral array processors. Here the properties
of discrete Fourier analysis are given along with means by which the fast Fourier
transform algorithm can be used on image data.
Chapters 8, 9 and 10 provide a treatment of the tools used in image classification,
commencing with supervised classification methods, moving through commonly
used clustering algorithms for unsupervised classification and concluding with means
for separability analysis. These are drawn together into classification methodologies
in Chap. 11 which also provides a set of case studies.
Even though the treatment provided is intended for the non-specialist image
analyst, it is still necessary that it be cast in the context of some vector and matrix
algebra. Otherwise it would be impracticable. Consequently, an appendix is provided
on essential results on vectors and matrices, and all important points in the text are
illustrated by simple worked examples. These demonstrate how vector operations are
evaluated. Beyond this material it is assumed the reader has a passing knowledge of
basic probability and statistics including an appreciation of the multivariate normal
distribution.
Several other appendices are provided to supplement the main presentation. One
deals with developments in image processing hardware and particularly the archi-
tecture (in block form) of interactive image display sub-systems. This material high-
lights trends towards hardware implementation of image processing and illustrates
how many of the algorithms presented in the book can be executed in near real time.
Owing to common practice, some decisions have had to be taken in relation to
definitions even though they could offend the purist. For example the term “pixel"
strictly refers to a unit of digital image data and not to an area on the ground. The latter
is more properly called an effective ground resolution element. However because the
practice of referring to ground resolution elements as pixels, dimensioned in metres,
is so widespread, the current treatment seeks not to be pedantic but rather follows
common practice for simplicity. A difficulty also arises with respect to the numbering
chosen for the wavebands in the Landsat multispectral scanner. Historically these
have been referred to as bands 4 to 7 for Landsats 1 to 3. From Landsat 4 onwards
they have been renumbered as bands 1 to 4. The convention adopted herein is mixed.
When a particular satellite is evident in the discussion, the respective convention is