The Image Processing Handbook Sixth Edition 6th

Edition John C Russ download

https://ebookbell.com/product/the-image-processing-handbook-
sixth-edition-6th-edition-john-c-russ-2225300

Explore and download more ebooks at ebookbell.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

The Image Processing Handbook 7th Edition Russ John Cneal F Brent

https://ebookbell.com/product/the-image-processing-handbook-7th-
edition-russ-john-cneal-f-brent-22131752

The Image Processing Handbook Fourth Edition 4th Edition John C Russ

https://ebookbell.com/product/the-image-processing-handbook-fourth-
edition-4th-edition-john-c-russ-2325970

The Image Processing Handbook 6ed Russ Jc

https://ebookbell.com/product/the-image-processing-handbook-6ed-russ-
jc-2598084

The Image Processing Handbook 7th Edition John C Russ F Brent Neal

https://ebookbell.com/product/the-image-processing-handbook-7th-
edition-john-c-russ-f-brent-neal-5245774
The Image Processing Handbook 5th Edition John C Russ

https://ebookbell.com/product/the-image-processing-handbook-5th-
edition-john-c-russ-981882

The Handbook Of Astronomical Image Processing Richard Berry James

Burnell

https://ebookbell.com/product/the-handbook-of-astronomical-image-
processing-richard-berry-james-burnell-52165134

The Electrical Engineering Handbook Circuits Signals Speech And Image

Processing Richard C Dorf

https://ebookbell.com/product/the-electrical-engineering-handbook-
circuits-signals-speech-and-image-processing-richard-c-dorf-4719780

The Image Processing Cookbook 4th Edition John C Russ

https://ebookbell.com/product/the-image-processing-cookbook-4th-
edition-john-c-russ-5848668

The Art Of Image Processing Digital Camera Processing 1st Edition

Chris Bore

https://ebookbell.com/product/the-art-of-image-processing-digital-
camera-processing-1st-edition-chris-bore-46665556
 The
IMAGE
PROCESSING
Handbook
Sixth Edition

John C. Russ
 The

IMAGE
PROCESSING
Handbook
Sixth Edition
 The

IMAGE
PROCESSING
Handbook
Sixth Edition
John C. Russ
North Carolina State University
Materials Science and Engineering Department
Raleigh, North Carolina

Boca Raton London New York

CRC Press is an imprint of the

Taylor & Francis Group, an informa business
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2011 by Taylor and Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Printed in the United States of America on acid-free paper

10 9 8 7 6 5 4 3 2 1

International Standard Book Number-13: 978-1-4398-4063-4 (Ebook-PDF)

This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been
made to publish reliable data and information, but the author and publisher cannot assume responsibility for the valid-
ity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this
form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may
rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or uti-
lized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopy-
ing, microfilming, and recording, or in any information storage or retrieval system, without written permission from the
publishers.

For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://
www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923,
978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For
organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for
identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
Contents

Acknowledgments. . . . . . . . . . . . . . . . . . . xi
Introduction . . . . . . . . . . . . . . . . . . . . . . xiii
Author . . . . . . . . . . . . . . . . . . . . . . . . . xvii

1 Acquiring Images . . . . . . . . . . . . . . . . 1
Human reliance on images for information............ 1
Video cameras................................................... 6
CCD cameras.................................................... 7
Camera artifacts and limitations..........................13
Color cameras..................................................15
Camera resolution.............................................18
Focusing.......................................................... 20
Electronics and bandwidth limitations...................21
Pixels...............................................................24
Gray scale resolution........................................ 26
Noise............................................................. 28
High depth images........................................... 30
Color imaging...................................................34
Digital camera limitations.................................. 42
Color spaces................................................... 42
Color correction................................................52
Color displays...................................................54
Image types..................................................... 56
Range imaging................................................ 58
Multiple images............................................... 64
Stereoscopy..................................................... 69
Imaging requirements........................................ 77

v
© 2011 by Taylor & Francis Group, LLC
 2 Human Vision. . . . . . . . . . . . . . . . . . 85
What we see and why...................................... 85
Recognition..................................................... 88
Technical specs................................................ 92
Acuity..............................................................97
What the eye tells the brain..............................101
Spatial comparisons........................................103
Local to global hierarchies................................107
It’s about time................................................. 113
The third dimension.......................................... 118
How versus what............................................. 121
Seeing what isn’t there, and vice versa...............122
Image compression..........................................125
A world of light...............................................126
Size matters....................................................129
Shape (whatever that means)............................132
Context..........................................................133
Arrangements must be made.............................135
Seeing is believing.......................................... 137
So in conclusion..............................................139

3 Printing and Storage. . . . . . . . . . . . 141

Printing.......................................................... 141
Dots on paper.................................................145
Color printing.................................................150
Printing hardware............................................156
Film recorders................................................. 161
Other presentation tools...................................162
File storage.....................................................163
Storage media................................................164
Magnetic recording.........................................166
Databases for images......................................167
Browsing and thumbnails.................................. 174
Lossless coding................................................ 178
Reduced color palettes.....................................183
JPEG compression...........................................184
Wavelet compression.......................................187
Fractal compression.........................................192
Digital movies.................................................194

4 Correcting Imaging Defects . . . . . . . 199

Contrast expansion......................................... 200
Noisy images................................................ 205
Neighborhood averaging................................ 208
Neighborhood ranking.................................... 214
Other neighborhood noise reduction methods.....226

vi Contents
© 2011 by Taylor & Francis Group, LLC
 Defect removal, maximum entropy, and
maximum likelihood.........................................232
Nonuniform illumination...................................235
Fitting a background function............................238
Rank leveling..................................................244
Color images..................................................248
Non-planar views............................................250
Computer graphics..........................................252
Geometric distortion........................................254
Alignment.......................................................256
Interpolation...................................................261
Morphing.......................................................265

5 Image Enhancement in the Spatial

Domain. . . . . . . . . . . . . . . . . . . . . . 269
Contrast manipulation......................................270
Histogram equalization.................................... 274
Local equalization...........................................279
Laplacian.......................................................283
Derivatives......................................................293
Finding edges with gradients............................296
More edge detectors...................................... 306
Texture........................................................... 312
Fractal analysis............................................... 317
Implementation notes....................................... 319
Image math.................................................... 319
Subtracting images..........................................320
Multiplication and division................................323
Principal components analysis...........................325
Other image combinations...............................331

6 Processing Images in Frequency

Space. . . . . . . . . . . . . . . . . . . . . . . 337
About frequency space....................................337
The Fourier transform.......................................338
Fourier transforms of simple functions.................341
Frequencies and orientations.............................345
Preferred orientation........................................350
Texture and fractals.........................................351
Isolating periodic noise....................................356
Selective masks and filters................................361
Selection of periodic information.......................364
Convolution....................................................370
Deconvolution.................................................372
Noise and Wiener deconvolution......................378
Template matching and correlation....................385
Autocorrelation...............................................391

Contents vii
© 2011 by Taylor & Francis Group, LLC
 7 Segmentation and Thresholding. . . .395
Thresholding...................................................395
Automatic settings...........................................398
Multiband images.......................................... 403
Two-dimensional thresholds.............................. 405
Multiband thresholding................................... 408
Thresholding from texture.................................. 411
Multiple thresholding criteria............................. 414
Textural orientation.......................................... 415
Region boundaries.......................................... 419
Conditional histograms.....................................426
Boundary lines................................................427
Contours....................................................... 430
Image representation.......................................432
Other segmentation methods............................436
The general classification problem.................... 440

8 Processing Binary Images. . . . . . . . 443

Boolean operations........................................ 443
Combining Boolean operations........................ 446
Masks........................................................... 450
From pixels to features.....................................452
Boolean logic with features...............................457
Selecting features by location...........................461
Double thresholding........................................ 466
Erosion and dilation........................................ 468
Opening and closing.......................................471
Isotropy..........................................................473
Measurements using erosion and dilation...........478
Extension to gray scale images.........................481
Morphology neighborhood parameters............. 482
Examples of use..............................................484
Euclidean distance map.................................. 488
Watershed segmentation..................................491
Ultimate eroded points.....................................494
Skeletons........................................................498
Boundary lines and thickening......................... 503
Combining skeleton and EDM.......................... 506

9 Global Image Measurements. . . . . . 511

Global measurements and stereology................ 511
Surface area................................................... 516
ASTM Grain Size............................................. 521
Multiple types of surfaces.................................523
Length............................................................525
Thickness........................................................527
Sampling strategies.........................................530

viii Contents
© 2011 by Taylor & Francis Group, LLC
 Determining number........................................532
Curvature, connectivity, and the Disector............535
Anisotropy and gradients.................................538
Size distributions.............................................542
Classical stereology (unfolding).........................543

10 Feature-Specific Measurements . . . . 547

Brightness measurements..................................547
Determining location........................................556
Orientation.....................................................559
Neighbor relationships.....................................562
Alignment.......................................................567
Counting........................................................ 574
Special counting procedures.............................579
Feature size....................................................584
Circles and ellipses..........................................587
Caliper dimensions..........................................589
Perimeter........................................................592

11 Characterizing Shape . . . . . . . . . . . 597

Describing shape............................................597
Dimensionless ratios.........................................599
Fractal dimension........................................... 604
Harmonic analysis........................................... 610
Topology........................................................620
Three dimensions.............................................623

12 Feature Recognition and

Classification. . . . . . . . . . . . . . . . . . 627
Template matching and cross-correlation............628
Parametric description......................................631
Decision points...............................................635
Multidimensional classification..........................639
Learning systems..............................................646
kNN and cluster analysis.................................652
Expert systems.................................................655
Neural nets.....................................................657
Syntactical models...........................................659

13 Tomographic Imaging. . . . . . . . . . . 661

More than two dimensions................................661
Volume imaging vs. sections.............................664
Basics of reconstruction....................................670
Algebraic reconstruction methods......................676
Maximum entropy...........................................679
Defects in reconstructed images........................681

Contents ix
© 2011 by Taylor & Francis Group, LLC
 Beam hardening............................................. 686
Imaging geometries.........................................691
Three-dimensional tomography..........................695
High-resolution tomography..............................701

14 3D Visualization . . . . . . . . . . . . . . . 707
Sources of 3D data..........................................707
Serial sections.................................................709
Optical sectioning........................................... 713
Sequential removal.......................................... 715
Stereo measurement........................................ 717
3D data sets...................................................722
Slicing the data set..........................................724
Arbitrary section planes...................................727
The use of color.............................................. 731
Volumetric display...........................................732
Stereo viewing................................................736
Special display hardware.................................739
Ray tracing..................................................... 741
Reflection.......................................................746
Surfaces.........................................................750
Multiply connected surfaces..............................754
Image processing in 3D...................................759
Measurements on 3D images............................763

15 Imaging Surfaces . . . . . . . . . . . . . . 767

Producing surfaces........................................... 767
Imaging surfaces by physical contact.................770
Noncontacting measurements...........................773
Microscopy of surfaces....................................777
Surface composition imaging............................782
Processing of range images..............................783
Processing of composition maps........................787
Data presentation and visualization...................788
Rendering and visualization.............................. 791
Analysis of surface data...................................796
Profile measurements....................................... 800
The Birmingham measurement suite................... 803
Topographic analysis and fractal dimensions..... 809

References . . . . . . . . . . . . . . . . . . . . . . 817

x Contents
© 2011 by Taylor & Francis Group, LLC
Acknowledgments

All of the image processing and the creation of the resulting figures included in this
book were performed on an Apple Macintosh® and/or a Sony VAIO® computer, using
Adobe Photoshop® CS4 with the Fovea Pro plug-ins. Many of the images were acquired
directly from various microscopes and other sources that provided digital output directly
to the computer. Others were captured using a variety of digital cameras (Sony, Nikon,
Canon, and others), and some were obtained using flat-bed and slide scanners (Nikon and
Epson), often from images supplied by colleagues and researchers. These are acknowl-
edged wherever the origin of an image could be determined. A few examples, taken
from the literature, are individually referenced.
The book was delivered to the publisher in digital form (on a writable DVD), without
intermediate hard copy, negatives or prints of the images, etc. Among other things,
this means that the author must bear full responsibility for typographical errors or
problems with the figures. Every effort has been made to show enlarged image frag-
ments that will reveal pixel-level detail when it is important. The process has also
forced me to learn more than I ever hoped to know about some aspects of publish-
ing technology! However, going directly from disk file to print also shortens the time
needed in production and helps to keep costs down, while preserving the full quality
of the images. Grateful acknowledgment is made of the efforts by the excellent edi-
tors at CRC Press to educate me and to accommodate the unusually large number of
illustrations in this book (more than 2000 figures and more than a quarter of a mil-
lion words).
Special thanks are due to Chris Russ (Reindeer Graphics Inc., Asheville, NC) who has
helped to program many of these algorithms and contributed invaluable comments, and
especially to Helen Adams, who has proofread many pages, endured many discussions
about ways to present information effectively, and provided the support (and the occa-
sional glass of wine) that make writing projects such as this possible.

John C. Russ
Raleigh, NC

xi
© 2011 by Taylor & Francis Group, LLC
Introduction

Image processing is used in a wide variety of applications, for two somewhat differ-
ent purposes:

1. improving the visual appearance of images to a human observer, including their

printing and transmission, and
2. preparing images for the measurement of the features and structures which they
reveal.

The techniques that are appropriate for each of these tasks are not always the same, but
there is considerable overlap. This book covers methods that are used for both tasks.
To do the best possible job, it is important to know about the uses to which the pro-
cessed images will be put. For visual enhancement, this means having some familiarity
with the human visual process and an appreciation of what cues the viewer responds
to in images. A chapter on human vision addresses those issues. It also is useful to
know about the printing or display process, since many images are processed in the
context of reproduction or transmission. Printing technology for images has advanced
significantly with the consumer impact of digital cameras, and up-to-date information
is provided.
The measurement of images is often a principal method for acquiring scientific data and
generally requires that features or structure be well defined, either by edges or unique
brightness, color, texture, or some combination of these factors. The types of measure-
ments that can be performed on entire scenes or on individual features are important in
determining the appropriate processing steps. Several chapters deal with measurement
in detail. Measurements of size, position, and brightness deal with topics that humans
largely understand, although human vision is not quantitative and is easily fooled. Shape
is a more difficult concept, and a separate chapter added in this edition summarizes a
variety of ways that shape may be described by numbers. Measurement data may be
used for classification or recognition of objects. There are several different strategies that
can be applied, and examples are shown.
It may help to recall that image processing, like food processing or word processing,
does not reduce the amount of data present but simply rearranges it. Some arrangements

xiii
© 2011 by Taylor & Francis Group, LLC
may be more appealing to the senses, and some may convey more meaning, but these
two criteria may not be identical nor call for identical methods.
This handbook presents an extensive collection of image processing tools, so that the
user of computer-based systems can both understand those methods provided in pack-
aged software and program those additions which may be needed for particular applica-
tions. Comparisons are presented for different algorithms that may be used for similar
purposes, using a selection of representative pictures from various microscopy tech-
niques, as well as macroscopic, remote sensing, and astronomical images. It is very
important to emphasize that the scale of the image matters very little to the techniques
used to process or analyze it. Microscopes that have a resolution of nm and telescopes
that produce images covering light years produce images that require many of the same
algorithms.
The emphasis throughout the book continues to be on explaining and illustrating meth-
ods so that they can be clearly understood, rather than providing dense mathematics.
With the advances in computer speed and power, tricks and approximations in search
of efficiency are less important, so that examples based on exact implementation of
methods with full precision can generally be implemented on desktop systems. The top-
ics covered are generally presented in the same order in which the methods would be
applied in a typical workflow.
For many years, in teaching this material to students I have described achieving mastery of
these techniques as being much like becoming a skilled journeyman carpenter. The num-
ber of distinct woodworking tools — saws, planes, drills, etc. — is relatively small, and
although there are some variations — slotted vs. Phillips-head screwdrivers, for example
— knowing how to use each type of tool is closely linked to understanding what it does.
With a set of these tools, the skilled carpenter can produce a house, a boat, or a piece of
furniture. So it is with image processing tools, which are conveniently grouped into only
a few classes, such as histogram modification, neighborhood operations, Fourier-space
processing, and so on, and can be used to accomplish a broad range of purposes. Visiting
your local hardware store and purchasing the appropriate tools do not provide the skills
to use them. Understanding their use requires practice, which develops the ability to visu-
alize beforehand what each will do. The same is true of the tools for image processing.
In revising the book for this new edition, I have again tried to respond to some of the
comments and requests of readers and reviewers. New chapters on the measurement
of images and the subsequent interpretation of the data were added in the second edi-
tion, and a section on surface images in the third. The fourth edition added the stereo-
logical interpretation of measurements on sections through three-dimensional structures
and the various logical approaches to feature classification. The fifth edition brought
expanded sections on deconvolution, extended dynamic range images, and multichannel
imaging, including principal components analysis. In this sixth edition, a new chapter on
the meaning of shape has been added, as well as additional material on imaging in more
than two dimensions. The sections on the ever-advancing hardware for image capture
and printing have been expanded and information added on the newest hardware and
software technologies.
As in past editions, I have resisted suggestions to put “more of the math” into the book.
There are excellent texts on image processing, compression, mathematical morphol-
ogy, etc., that provide as much rigor and as many derivations as may be needed. Many
of them are referenced here. But the thrust of this book remains teaching by example.

xiv Introduction
© 2011 by Taylor & Francis Group, LLC
Few people learn the principles of image processing from the equations. Just as we use
images to communicate ideas and to “do science,” so most of us use images to learn
about many things, including imaging itself. The hope is that by seeing and comparing
what various operations do to representative images, you will discover how and why to
use them. Then, if you need to look up the mathematical foundations, they will be easier
to understand.
A very real concern for everyone involved in imaging, particularly in scientific and foren-
sic fields, is the question of what constitutes proper and appropriate processing and what
constitutes unethical or even fraudulent manipulation. The short answer is that anything
that alters an image so as to create a false impression on the part of the viewer is wrong.
The problem with that answer is that it does not take into account the fact that different
viewers will tend to see different things in the image anyway, and that what constitutes
a false impression for one person may not for another.
The first rule is always to store a permanent copy of the original image along with rel-
evant data on its acquisition. The second rule is to carefully document whatever steps
are taken to process the image and generally to report those steps when the processed
image is published. Most scientific publications and the editors who review submitted
papers have become more aware in recent years of the ease with which image process-
ing can be performed and the dangers of inadequate documentation. For example, see
M. Rossner and K. M. Yamada (2004; J. Cell Biology) for that journal’s policy on image
ethics and examples of improper manipulation.
For forensic purposes, there is an additional responsibility to fully record the entire step-
by-step procedures that are used and to make sure that those methods are acceptable
in court according to the U.S. Supreme Court’s Daubert ruling (Daubert v. Merrell Dow
Pharmaceuticals (92-102), 509 U.S. 579, 1993), which generally means that not only are
the methods widely accepted by professionals, but also that they have been rigorously
tested and have known performance outcomes. In a forensic setting, there will often
be a need to explain a procedure, step by step, to a non-technical jury. This frequently
requires showing that the details obtained from the image are really present in the origi-
nal but only became visually evident with the processing.
Some procedures, such as rearranging features or combining them within a single image,
or differently adjusting the contrast of several images to make them appear more alike,
are clearly misleading and generally wrong. Some, such as using copy-and-paste to dupli-
cate a portion of an image, or selectively erasing portions of an image, are out-and-out
fraudulent. Even selective cropping of an image (or choosing which field of view to
record) can create a false impression.
The general guideline to be considered is that it is never acceptable to add anything to
an image, but it may be acceptable to suppress or remove some information if it makes
the remaining details more accessible, either visually for presentation and communication
or to facilitate measurement. Of course, the procedures used must be documented and
reported. Any of the procedures shown here may be appropriate in a particular instance.
But they can also be misused and should in any case never be used without understand-
ing and careful documentation. The heart of the scientific method is replicability. If
adequate information is provided on the processing steps applied and the original image
data are preserved, then the validity of the results can be independently verified.
An important but often overlooked concern is the need to avoid using programs that
alter the image without the user being aware of it. For example, carefully correcting the

Introduction xv
© 2011 by Taylor & Francis Group, LLC
colors in an image using Photoshop® and then placing it in PowerPoint® for presentation
will cause changes even on the same computer screen (as well as discarding pixels and
reducing resolution if copy-and-paste is used for the transfer). In addition, the image
may appear different on another computer monitor or when using a projector. Pasting an
image into Microsoft® Word will reduce the resolution and color or gray scale dynamic
range. This may not affect the printed document, which has less gamut than the com-
puter screen anyway, but the image cannot be subsequently retrieved from the document
in its original form. Saving an image with a lossy compression method such as jpeg will
discard potentially important information that cannot be recovered.
The reader is encouraged to use this book in concert with a real source of images and
a computer-based system and to freely experiment with different methods to deter-
mine which are most appropriate for his or her particular needs. Selection of image
processing tools to explore images when you don’t know the contents beforehand is
a much more difficult task than using tools to make it easier for another viewer or a
measurement program to see the same things you have discovered. It places greater
demand on computing speed and the interactive nature of the interface. But it particu-
larly requires that you become a very analytical observer of images. If you can learn
to see what the computer sees and predict what various algorithms will do, you will
become a better viewer and obtain the best possible images, suitable for further pro-
cessing and analysis.
To facilitate this hands-on learning, I have collaborated with my son, Chris Russ, to
write a companion book, Introduction to Image Processing and Analysis, which teaches
how to program these algorithms and create Adobe Photoshop compatible plug-ins that
implement the methods. The downloadable solutions to the book’s worked problems
can be used to apply the routines on either Macintosh or Windows computers. There are
additional links to downloadable plug-ins and trial program packages on my Web site at
http://www.DrJohnRuss.com.

xvi Introduction
© 2011 by Taylor & Francis Group, LLC
Author

In his fifty-year career as scientist

and educator, John Russ has used
image processing and analysis as
a principal tool for understanding
and characterizing the structure and
function of materials. Images from a
wide variety of devices — including
light and electron microscopes, x-ray
and neutron tomography, and more
— require computer processing and
measurement to extract the important
data. Much of Russ’ research work
has been concerned with the micro-
structure and surface topography of
metals and ceramics. He has received
funding for his research and teaching
from government agencies and from
industry. Although retired, Dr. Russ
is currently assisting in the establish-
ment of a new laboratory and program at North Carolina State University, which will be
the first in the nation to offer advanced degrees in Forensic Science and Engineering.
Familiarity with the algorithms and instruments led to Dr. Russ’ expertise being extended
to a much broader range of images — from astronomy to biomedical research to food
science to forensics. In addition to students in NCSU’s College of Engineering, Russ has
been on graduate student commitees and collaborated with faculty in textiles, pulp
and paper products, veterinary medicine, microbiology, food science, and archaeology,
among others. Teaching the principles and methods involved to several thousand stu-
dents and consulting for many industrial clients have further broadened Dr. Russ’ experi-
ence and the scope of applications for image processing and analysis.
After retirement, Dr. Russ was Research Director for Rank Taylor Hobson, a manu-
facturer of precision instrumentation. He continues to write, to consult for a variety
of companies (and to provide expert testimony in criminal and civil cases), to teach

xvii
© 2011 by Taylor & Francis Group, LLC
workshops worldwide on image processing and analysis, and to review publications and
funding proposals. He is active in the Microscopy Society of America, the Microbeam
Analysis Society, the Society of Photo-Optical Instrumentation Engineering (SPIE), the
International Society for Stereology, is a board member of the Society for Quantitative
Morphology, and a Fellow of the Royal Microscopical Society, and has presented invited
lectures and workshops for these and other organizations. On November 16, 2006, the
New York Microscopical Society awarded John Russ the Ernst Abbe Memorial Award for
his contributions to the field of microscopy as a developer of computer-assisted micros-
copy and image analysis.

xviii Author
© 2011 by Taylor & Francis Group, LLC
 1
Acquiring Images

Human reliance on images for information

H
umans are primarily visual creatures. Not all animals depend on their eyes, as we do,
for most of the information received about their surroundings (the characteristics of
human vision are discussed in Chapter 2). This bias in everyday life extends to how
we pursue more technical goals as well. Scientific instruments commonly produce images
to communicate their results to the operator, rather than generating audible tones or emit-
ting a smell. Space missions to other planets and equally arduous explorations of the ocean
depths always include cameras as major components, and the success of those missions is
often judged by the quality of the images returned. This suggests a few of the ways in which
humans have extended the range of natural vision. Optical devices such as microscopes and
telescopes allow us to see things that are vastly smaller or larger than we could otherwise.
Beyond the visible portion of the electromagnetic spectrum (a narrow range of wavelengths
between about 400 and 700 nanometers) there are sensors capable of detecting infrared and
ultraviolet light, X-rays, and radio waves, and perhaps soon even gravity waves. Figure 1.1
shows an example, an image presenting radio telescope data in the form of an image in which
color represents the Doppler shift in the radio signal. Such devices and presentations are used
to further extend imaging capability.
Signals other than electromagnetic radiation can be used to produce images, too. Novel new
types of microscopes that use atomic-scale probes to “feel” the specimen surface present their
data as images (Figure 1.2). The data so collected may represent the surface elevation and
topography, but other signals, such as the lateral drag force on the probe, may also be used.
Acoustic waves at low frequency produce sonar images, while at gigahertz frequencies the
acoustic microscope produces images with resolution similar to that of the light microscope,
but with image contrast that is produced by local variations in the attenuation and refraction
of sound waves rather than light. Figure 1.3 shows an acoustic microscope image of a sub-
surface defect, and Figure 1.4 shows a sonogram of a baby in the womb.
Some images such as holograms or electron diffraction patterns record brightness as a func-
tion of position, but are unfamiliar to the observer. Figure 1.5 shows an image of an electron
diffraction pattern from a transmission electron microscope, in which the atomic structure of
the samples is revealed (but only by measurement and to those who know how to interpret

1
© 2011 by Taylor & Francis Group, LLC
 Figure 1.1 Radio astronomy pro-
duces images such as this view of
Messier 33 (generated with data
from telescopes of the National
Radio Astronomy Observatory,
a National Science Foundation
Facility managed by Associated
Universities, Inc.). These are
often displayed with false colors
to emphasize subtle variations
in signal strength or - as in this
example - Doppler shift.

Figure 1.2 Atomic force micro-

scope image of human chromo-
somes (courtesy S, Thalhammer,
F. Jamitzky, Helmholtz Zentrum
München, Germany).

the data). Other kinds of data, including weather maps with specialized symbols, graphs of
business profit and expenses, and charts with axes representing time, family income, choles-
terol level, or even more obscure parameters, have become part of daily life, as illustrated in
Figure 1.6. The latest developments in computer interfaces and displays make extensive use
of graphics, to take advantage of the large bandwidth of the human visual pathway. Tufte
(1990, 1997, 2001) in particular has demonstrated the power of appropriate graphics to com-
municate complex information.
There are some important differences between human vision, the kind of information it
extracts from images, and the ways in which it seems to do so, as compared to the use of
imaging devices based on computers for scientific, technical, or forensic purposes. Humans

2 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
 (a)

(b)

Figure 1.3 Acoustic microscope image of voids in solder bond beneath a GaAs die: (a) die surface; (b)
acoustic image showing strong signal reflections (white areas) from the surface of the voids (courtesy
J. E. Semmens, Sonoscan Inc).

are especially poor at judging color or brightness of objects and features within images unless
they can be exactly compared by making them adjacent. Human vision is inherently com-
parative rather than quantitative, responding to the relative size, angle, or position of several
objects but unable to supply numeric measures unless one of the reference objects is a mea-
suring scale. Overington (1976; 1992) disagrees with this widely accepted and documented
conclusion but presents no compelling counter evidence. Chapter 2 illustrates some of the
consequences of the characteristics of human vision as they affect what is perceived.
This book’s purpose is not to study the human visual pathway directly, but the overview in
Chapter 2 can help the reader to understand how humans see things so that we become bet-
ter observers. Computer-based image processing and analysis use algorithms based on human
vision methods in some cases, but also employ other methods that seem not to have direct
counterparts in human vision. In particular, some image processing methods are based on the
physics of the image formation and detection process (Sharma, 2005).
Many of the examples and much of the analysis presented in this text involve images from
various types of microscopes. The three classes of imaging applications that generally offer

Acquiring Images 3
© 2011 by Taylor & Francis Group, LLC
Figure 1.4 Surface reconstruction of sonogram imaging, showing a 26 week old fetus in the womb.

Figure 1.5 A convergent beam electron diffraction (CBED) pattern from an oxide microcrystal, which
can be indexed and measured to provide high accuracy values for the atomic unit cell dimensions.

4 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
Figure 1.6 Typical graphics used to communicate news information include one-dimensional plots
such as stock market reports and two-dimensional presentations such as weather maps.

the most straightforward types of images are microscopy, aerial (and satellite) imagery, and
industrial quality control. That is because in those situations there is the greatest knowledge
and/or control over the imaging geometry and the illumination of the scene. In more general
“real world” cases the analysis and interpretation of the image contents can be much more
difficult. Objects may lie at various distances from the camera, which complicates determin-
ing size, may have different lighting, which alters their color, and may even partially obscure
other objects. Crime scene and accident photographs are often taken under difficult condi-
tions, from less than optimum points of view, and with variable lighting, so that their analysis
can be challenging.
The basic techniques for image processing and measurement are much the same for images
regardless of their source or scale. Images ranging from microscopy to astronomy, images
formed with light photons or sound waves, magnetic resonance or scanning profilometers,
have much in common and the techniques for dealing with their imperfections, enhancing
and extracting the details, and performing measurements utilize the same algorithms and
techniques, which are set out in the following chapters. The interpretation of the measure-
ments, as presented in later chapters, does require some specialization for different viewing
geometries, but is fundamentally independent of magnification.

Acquiring Images 5
© 2011 by Taylor & Francis Group, LLC
Video cameras
When the first edition of this book was published in 1990, the most common and affordable
way of acquiring images for computer processing was with a video camera. Mounted onto a
microscope or copystand, in a satellite or space probe, or using appropriate optics to view
an experiment, the camera sent an analog signal to a separate “frame grabber” or analog-to-
digital converter (ADC) interface board in the computer, which then stored numeric values in
memory (Inoué, 1986; Inoué & Spring, 1997).
The basic form of the original type of video camera is the vidicon, illustrated in Figure 1.7. It
functions by scanning a focused beam of electrons across a phosphor coating applied to the
inside of an evacuated glass tube. The light enters the camera through the front glass surface
(and a thin metallic anode layer) and creates free electrons in the phosphor. These vary the
local conductivity of the layer, so the amount of current that flows to the anode varies as the
beam is scanned, according to the local light intensity. This analog (continuously varying)
electrical signal is amplified and, as shown in Figure 1.8, conforms to standards of voltage
and timing (the standards and timing are slightly different in Europe than the US, but the basic
principles remain the same).
Digitizing the voltage is accomplished by sampling it and generating a comparison voltage.
The child’s game of “guess a number” illustrates that it takes only eight guesses to arrive at a

Deflection and Focusing Coils Anode

Grid

Electron Beam
Cathode Glass Tube

Phosphor
Coating

Figure 1.7 Functional diagram of a vidicon tube. Light striking the phosphor coating changes its local
resistance and hence the current that flows as the electron beam scans in a raster pattern.

0.7 volt range

0.3 volt sync pulse

52 sec. picture width

63.5 sec. horizontal scan interval

Figure 1.8 Standard RS-170 video signal shows the brightness variation along one scan line (ranging
between 0 volts = black and 0.7 volts = white).

6 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
 Figure 1.9 Example of an image show-
ing pincushion distortion, as well as
loss of focus and vignetting in the
edges and corners.

value that defines the voltage to one part in 256 (the most widely used type of ADC). The first
guess is 128, or half the voltage range. If this is (e.g.) too large, the second guess subtracts
64. Each successive approximation adds or subtracts a value half as large as the previous. In
eight steps, the final (smallest) adjustment is made. The result is a number that is conveniently
stored in the 8-bit memory of most modern computers.
The tube-type camera has several advantages and quite a few drawbacks. Scanning the beam
with electromagnetic or electrostatic fields can produce a distorted scan (pincushion or barrel
distortion, or more complicated situations) and is subject to degradation by stray fields from
wiring or instrumentation. Figure 1.9 shows an example of pincushion distortion, as well
as vignetting and loss of focus. Maintaining focus in the corners of the image takes special
circuitry, and the corners may also be darkened (vignetting) by the reduction in effective lens
aperture and the additional thickness of glass through which the light must pass. The sealed
vacuum systems tend to deteriorate with time, and the “getter” used to adsorb gas molecules
may flake and fall onto the phosphor if the camera is used in a vertical orientation. The
response of the camera (voltage vs. brightness) approximates the logarithmic response of film
and the human eye, but this varies for bright and dark scenes. Recovery from bright scenes
and bright spots is slow, and blooming can occur in which bright light produces spots that
spread laterally in the coating and appear larger than the features really are, with “comet tails”
in the scan direction.
There are, however, some advantages of the tube-type camera. The spatial resolution is very
high, limited only by the grain size of the phosphor and the size of the focused beam spot.
Also, the phosphor has a spectral response that can be made similar to that of the human eye,
which sees color from red (about 0.7 µm wavelength) to blue (about 0.4 µm). Adaptations of the
basic camera design with intermediate cathode layers or special coatings for intensification are
capable of acquiring images in very dim light (e.g., night scenes, fluorescence microscopy).

CCD cameras
The tube-type camera has now been largely supplanted by the solid-state chip camera, the
oldest and simplest form of which is the CCD (charge-coupled device). The camera chip con-
tains an array of diodes that function as light buckets. Light entering the semiconductor raises
electrons from the valence to the conduction band, so the number of electrons is a direct lin-
ear measure of the light intensity. The diodes are formed by photolithography, so they have a

Acquiring Images 7
© 2011 by Taylor & Francis Group, LLC
Figure 1.10 The basic principle of CCD operation, illustrated as a set of buckets and conveyors (after
Janesick, 2001).

perfectly regular pattern with no image distortion or sensitivity to the presence of stray fields.
The devices are also inexpensive and rugged compared to tube cameras. CCDs were first
invented and patented at Bell Labs in 1969 (George Smith was awarded the 2009 Nobel Prize
in Physics for this invention), and have now largely displaced film in consumer and profes-
sional still and movie cameras.
The basic operation of a CCD is illustrated in Figure 1.10. Each bucket represents one “pixel”
in the camera (this word has a lot of different meanings in different contexts, as explained
below, so it must be used with some care). With anywhere from a few hundred thousand to
several million detectors on the chip, it is impractical to run wires directly to each one in order
to read out the signal. Instead, the electrons that accumulate in each bucket due to incident
photons are transferred, one line at a time, to a readout row. On a clock signal, each column
of pixels shifts the charge by one location. This places the contents of the buckets into the
readout row, and that row is then shifted, one pixel at a time but much more rapidly, to dump
the electrons into an amplifier, which produces an analog voltage signal that may be sent out
directly or measured to produce the numeric output from a digital camera.
The simplest way of shifting the electrons is shown in Figure 1.11. Every set of three elec-
trodes on the surface of the device constitutes one pixel. By applying a voltage to two of the
electrodes, a field is set up in the semiconductor that acts like a bucket. Electrons are trapped
in the central region by the high fields on either side. Note that this does not reduce the area
sensitive to incoming photons, because electrons generated in the high field regions quickly
migrate to the low field bucket where they are held. By changing the voltage applied to the
regions in six steps or phases, as shown in the figure, the electrons are shifted by one pixel.
First one field region is lowered and the electrons spread into the larger volume. Then the
field on the other side is raised, and the electrons have been shifted by one-third of the pixel
height. Repeating the process acts like a conveyor belt and is the reason for the name “charge-
coupled device.”

8 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
 φ1
φ2
φ3

Insulator t1 t 2 t3 t4 t5 t6 t1
Semiconductor
1 pixel period

Electron
Wells
Transferring
Charge

Figure 1.11 Varying voltages on a set of three electrodes shifts electrons from one pixel to another in a
CCD.

One significant problem with the chip camera is its spectral response. Even if the chip is
reversed and thinned so that light enters from the side opposite the electrodes, very little blue
light penetrates into the semiconductor to produce electrons. On the other hand, infrared light
penetrates easily and these cameras have red and infrared (IR) sensitivity that far exceeds that
of human vision, usually requiring the installation of a blocking filter to exclude it (because
the IR light is not focused to the same plane as the visible light and thus produces blurred
or fogged images). Figure 1.12 shows this spectral response, which can be further tailored
and extended by using materials other than silicon. The chip can reach a high total efficiency
when antireflective coatings are applied, limited primarily by the “fill factor” — the area frac-
tion of the chip that contains active devices between the narrow ditches that maintain electri-
cal separation. Also, the chip camera has an output that is linearly proportional to the incident
light intensity, convenient for some measurement purposes but very different from human
vision, the vidicon, and photographic film, which are all approximately logarithmic.
Human vision notices brightness differences of a few percent, i.e., a constant ratio of change
rather than a constant increment. Film is characterized by a response to light exposure which
(after chemical development) produces a density vs. exposure curve such as that shown in
Figure 1.13. The low end of this curve represents the fog level of the film, the density that is
present even without exposure. At the high end, the film saturates to a maximum optical den-
sity, for instance based on the maximum physical density of silver particles or dye molecules.
In between, the curve is linear with a slope that represents the contrast of the film. A steep
slope corresponds to a high-contrast film that exhibits a large change in optical density with
a small change in exposure. Conversely, a low-contrast film has a broader latitude to record a
scene with a greater range of brightnesses. The slope of the curve is usually called “gamma.”
Many chip cameras include circuitry or processing that changes their output from linear to
logarithmic so that the image contrast is more familiar to viewers. The more expensive con-
sumer cameras and most professional cameras include the possibility to read the “raw” linear
data as well as the converted image.

Acquiring Images 9
© 2011 by Taylor & Francis Group, LLC
 Figure 1.12 Spectral response:
(a) Silicon based chip.
(b) Color sensors in the human eye,
              (a)
which are commonly identified as
red, green and blue sensitive but
cover a range of long, medium and
short wavelengths.

             (b)

When film is exposed directly to electrons, as in the transmission electron micrograph, rather
than photons (visible light or X-rays), the response curve is linear rather than logarithmic. Many
light photons are needed to completely expose a single silver halide particle for development,
but only a single electron is needed. Consequently, electron image films and plates are often
very high in density (values of optical density greater than 4, which means that 9999/10000 of
incident light is absorbed), which creates difficulties for many scanners and requires more than
8 bits to record.
The trend in camera chips has been to make them smaller and to increase the number of
pixels or diodes present. Some scientific cameras, such as that used in the Hubble telescope,
occupy an entire wafer. But for consumer devices, making each chip one-third, one-quarter,
or even two-tenths of an inch in overall (diagonal) dimension places many devices on a single
wafer and allows greater economic yield. It also requires smaller, less costly lenses. Putting
more pixels into this reduced chip area (for more spatial resolution, as discussed below)
makes the individual detectors small, but the ditches between then have to remain about the
same size to prevent electrons from diffusing laterally. The result is to reduce the total effi-
ciency markedly. Some devices place small lenses over the diodes to capture light that would

10 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
 Saturation

Figure 1.13 Response of photographic film. The

central portion of the curve shows a linear

Film Density
increase in density (defined as the base-ten
logarithm of the fraction of incident light that
is transmitted) with the logarithm of exposure. Linear Range
(Slope = Gamma)
High (“hard”) contrast corresponds to a steep
curve, while low (“soft”) contrast gives a less
steep curve and films have a greater dynamic
range. Fog
Reciprocity
Level
Failure

   Log (Exposure)

otherwise fall into the ditches, but these add cost and also are not so uniform as the diodes
themselves (which are typically within 1% across the entire chip).
The other, and more important, effect of making the detectors small is to reduce their capacity
for electrons, called the well capacity. A typical 15 µm pixel in a scientific grade CCD has a
capacity of about 500,000 electrons, which with low readout noise (as can be achieved in spe-
cial situations) of a few electrons gives a dynamic range greater than photographic film. Even
larger well capacity and dynamic range can be achieved by combining (binning) more detec-
tors for each stored pixel by using more steps in the phase shifting during readout. Reducing
the area of the detector reduces the well size, and with it the dynamic range.
Increasing the noise, for instance by reading out the signal at video rates (each horizontal line
in 52 µs for US standard definition video), dramatically reduces the dynamic range so that a
typical consumer grade video camera has no more than about 64 distinguishable brightness
levels (expensive studio cameras meet the broadcast video specification of 100 levels). Since
with the chip camera these are linear with brightness, they produce even fewer viewable gray
levels, as shown in Figure 1.14. This performance is much inferior to film, which can distin-
guish thousands of brightness levels.
CMOS (Complementary Metal-Oxide Semiconductor) chips can also be used as image sensors,
and in terms of sheer numbers are now more common than the original CCD devices. They
are primarily used in relatively inexpensive consumer cameras and camera phones, although
some have found their way into digital single lens reflex cameras. The conversion of light
photons to electrons functions in the same way as in the CCD chip. The differences start with
the way the signal is read out. In the CMOS designs there are from two to four transistors
immediately adjacent to the light sensor which convert the charge to a voltage and amplify the
signal. In principle, this means that any pixel in the array can be read out directly, addressing
a pixel by row and column just as in a memory chip (Figure 1.15). This is different from the
CCD method of “sweeping” the charge out to one corner of the array, reading all of the pixels
in a fixed order.
The space taken up by these control transistors reduces the “fill factor” or active area of the
chip that is sensitive to light, but this is often compensated for by placing lenses over each
detector to collect light from the dead areas and direct it to the active sensor. The lenses, and
the use of individual amplifiers for each pixel, generally make the sensors in a CMOS detec-
tor less uniform than those in the CCD array, producing a fixed pattern that can be compen-
sated for in software (requiring recording an image with uniform illumination). In addition to
the fixed pattern noise, the CMOS detectors usually have a greater amount of random noise

Acquiring Images 11
© 2011 by Taylor & Francis Group, LLC
 Figure 1.14 Comparison of visibility of
gray level steps from linear (equal
steps) and logarithmic (equal ratios)
detectors:
(a) Plots of intensity.
(b) D
 isplay of the values from (a).

superimposed on the image signal because of the separate amplifiers, additional wiring and
its associated capacitance and thermal noise, and greater dark current. The very small active
regions (at least in the smaller chips used in pocket cameras and phones, and particularly as
the pixel counts have risen to several million) have small well capacities, resulting in limited
dynamic range for the images. The images are usually stored with 8 bits per channel, because
of the way memory is traditionally organized, but often do not have that much actual bright-
ness resolution.
Larger area CMOS chips are also made which have larger detectors and consequently a greater
well capacity and greater dynamic range. One advantage of the CMOS designs as used in more
expensive cameras arises from the fact that the circuitry to access the pixels can be arranged
along two adjacent sides of the array (addressing the rows and columns, respectively). That
makes it possible to carefully trim away the chip on the other two sides, and arrange four of
the chips together to produce a larger sensor with higher pixel counts. This approach, com-
bined with the use of much larger sensors to achieve greater sensitivity and dynamic range,

12 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
 Figure 1.15 Schematic diagram of a typical
CMOS detector. Each active light sensor
(green) has additional transistors that are
connected to addressing and output lines.

has led some manufacturers to prefer CMOS detectors as large as a traditional film negative
for digital single-lens reflex cameras.
The advantages of CMOS sensors lie primarily in three areas: they consume much less power,
and so give better battery life; the amplifier and digitization circuitry can be placed on the
same chip to reduce size and perhaps increase ruggedness; and the production methods for
the wafers are essentially the same as those used for other standard silicon devices such as
memory and processors, whereas CCD wafers require unique processing. The latter advantage
is somewhat offset by the fact that high quality CMOS sensors do require somewhat custom-
ized fabrication for greater uniformity and fewer noise-producing imperfections than can be
tolerated in other devices. While the cost to fabricate CMOS sensors is less than for CCD, the
design costs are much higher. Of course, for devices that are to be produced in large quantity,
this is a minor factor. The overall trend has been for CMOS sensors to continue to improve
in quality and performance, and while the advantages of the CCD sensor are still important
for most technical applications, it is wise to consider the trade-offs on a case-by-case basis
(Nakamura, 2006; Holst & Lomheim, 2007).

Camera artifacts and limitations

There are several problems with video cameras using chips which contribute to the specific
types of defects present in the images that must be dealt with by subsequent processing. One
is the fact that many video signals are interlaced (Figure 1.16). With high-definition video,
and with digital still cameras, the image is scanned progressively. Interlacing is a clever trick
to minimize visual flicker in broadcast television images, accomplished with tube cameras by
scanning the electron beam in the same interlace pattern as the display television set. With a
chip camera, it requires that the array be read out twice for every 30th of a second frame, once
to collect the even numbered lines and again for the odd numbered lines. In fact, many cam-
eras combine two lines to get better sensitivity, averaging lines 1 and 2, 3 and 4, 5 and 6, and
so on, in one interlace field, and then 2 and 3, 4 and 5, 6 and 7, etc. in the other. This reduces

Acquiring Images 13
© 2011 by Taylor & Francis Group, LLC
 Figure 1.16
   (a) Interlace scan covers even
Line
1
numbered lines in one six-
2 tieth-second field, and even
3
4 numbered lines in a second
5
6 field.
7
8
9
(b) W hen motion is present
10 (either in the scene or caused
11
12 by camera motion), this pro-
13
14 duces an offset in the com-
Field 1 Field 2 plete image.
               (a)

(b)

vertical resolution but for casual viewing purposes is not noticeable. Motion can cause the
even and odd fields of a full frame to be offset from each other, producing a significant deg-
radation of the image, as shown in the figure. A similar effect occurs with stationary images
if the horizontal retrace signal is imprecise or difficult for the electronics to lock onto; this is
a particular problem with signals played back from consumer video tape recorders. (Moving
images are also distorted with progressive scan cameras, due to the time required to read from
the top of the image to the bottom.)
During the transfer and readout process, unless the camera is shuttered either mechanically
or electrically, photons continue to produce electrons in the chip. This produces a large

14 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
background signal that further degrades dynamic range and may produce blurring. Electronic
shuttering is usually done line-at-a-time so that moving images are distorted. Some designs
avoid shuttering problems by doubling the number of pixels, with half of them opaque to
incoming light. A single transfer shifts the electrons from the active detectors to the hidden
ones, from which they can be read out. Of course, this reduces the active area (fill factor) of
devices on the chip, reducing sensitivity by 50%.
The high speed of horizontal line readout can produce horizontal blurring of the signal, again
reducing image resolution. This is partially due to inadequate time for the electrons to diffuse
along with the shifting fields, to the time needed to recover electrons from traps (impurities
in the silicon lattice), and partially to the inadequate frequency response of the amplifier,
which is a trade-off with amplifier noise. Even though the individual electron transfers are
very efficient, better than 99.999% in most cases, the result of being passed through many
such transfers before being collected and amplified increases the noise. This varies from one
side of the chip to the other, and from the top to the bottom, and can be visually detected in
images if there is not a lot of other detail or motion to obscure it.
Many transfers of electrons from one detector to another occur during readout of a chip, and
this accounts for some of the noise in the signal. Purely statistical variations in the produc-
tion and collection of charge is a relatively smaller effect. The conversion of the tiny charge
to a voltage and its subsequent amplification is the greatest source of noise in most systems.
Readout and amplifier noise can be reduced by slowing the transfer process so that fewer elec-
trons are lost in the shifting process and the amplifier time constant can integrate out more of
the noise, producing a cleaner signal. Cooling the chip to about –40° also reduces the noise
from these sources and from dark current, or thermal electrons. Slow readout and cooling are
used only in non-video applications, of course. Digital still cameras use the same chip technol-
ogy (but much higher numbers of detectors) as solid state video cameras, and produce higher
quality images because of the slower readout. Janesick (2001) discusses the various sources
of noise and their control in scientific CCDs of the type used in astronomical imaging (where
they have almost entirely replaced film) and in space probes.

Color cameras
Color cameras can be designed in three principal ways, as shown in Figures 1.17, 1.18,
and 1.19. For stationary images (which includes many scientific applications such as
microscopy, but excludes “real-time” applications such as video), a single detector array
can be used to acquire three sequential exposures through red, green and blue filters,
respectively (Figure 1.17), which are then combined for viewing. The advantages of this
scheme include low cost and the ability to use different exposure times for the different
color bands, which can compensate for the poorer sensitivity of the silicon chip to short
wavelength (blue) light.
Many high-end consumer and most professional grade video cameras use three sensors
(Figure 1.18). A prism array splits the incoming light into red, green, and blue components,
which are recorded by three different sensors whose outputs are combined electronically to
produce a standard video image. This approach is more costly, since three chips are needed,
but for video applications they need not be of particularly high resolution (even a high-defini-
tion video camera has many fewer pixels than a digital still camera). The optics and hardware
to keep everything in alignment add some cost, and the depth of the prism optics makes it

Acquiring Images 15
© 2011 by Taylor & Francis Group, LLC
 Figure 1.17 Schematic diagram of a color
wheel camera with red, green and blue
filters. The fourth filter position is empty,
allowing the camera to be used as a mono-
chrome detector with greater sensitivity for
dim images (e.g., fluorescence microscopy).

  

Figure 1.18 Schematic diagram of the prisms

and dichroic filters for a three chip color
camera.

  

impractical to use short focal length (wide angle) lenses. This design is rarely used in digital
still cameras.
Video images are often digitized into a 640 × 480 array of stored pixels (the dimensions of
the VGA display that was once standard for personal computers), but this is not the actual
resolution of the image. The broadcast bandwidth limits the high frequencies and eliminates
any rapid variations in brightness and color. A standard definition video image has no more
than 330 actual elements of resolution in the horizontal direction for the brightness (lumi-
nance) signal, and about half that for the color (chrominance) information. Color information
is intentionally reduced in resolution because human vision is not very sensitive to blurring
of color beyond boundary lines.
Of course, video signals can be further degraded by poor equipment. Recording video on
consumer-grade tape machines can reduce the resolution by another 50% or more, particularly
if the record head is dirty or the tape has been used many times before (an unfortunately very

16 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
 Figure 1.19
   (a) Stripe and
(b) B
 ayer filter patterns
used in single chip
cameras.

(a)               (b)

common problem with forensic examination of surveillance video is that the tapes are played
— over and over — for visual examination by local police so that by the time professionals get
them, the oxide coating — and the information — has been damaged or even removed). Video
images are not very high resolution, although HDTV (high definition television) has improved
things somewhat. Consequently, video technology is usually a poor choice for scientific imag-
ing unless there is some special need to capture “real time” images (i.e., 25–30 frames per
second) to record changes or motion. Digital still cameras have largely replaced them, as they
produce much higher resolution images with greater dynamic range.
Most digital cameras use a single pixel array, often of very high pixel (detector) count, with a
color filter that allows red, green, and blue light to reach specific detectors. Different patterns
may be used (Figure 1.19), with the Bayer pattern being very common (invented by Kodak
researcher Bryce Bayer and the basis for U.S. Patent 3,971,065 “Color Imaging Array,” issued in
1976). Notice that it assigns twice as many detectors for green as for red or blue, which mim-
ics to some extent the human eye’s greater sensitivity to green. The problem with the single-
chip camera, of course, is that the image resolution in each color channel is reduced. The red
intensity at some locations must be interpolated from nearby sensors, for example. It is also
necessary to design the filters to give the same brightness sensitivity in each channel. If this is
not done well, a herring-bone pattern (often referred to as a “zipper”) appears in images of a
uniform gray test card and color fringes appear along contrast edges in the picture, as shown
in Figure 1.20.
Interpolation techniques for Bayer pattern color filters reduce the image resolution as com-
pared to the number of individual detectors in the camera (which is generally the speci-
fication advertised by the manufacturer). Inherently, this “demosaicking” process involves
trade-offs between image sharpness, details, noise, processing time and conversion artifacts.
The quality of the result, judged by its ability to preserve sharp boundaries in brightness
while minimizing the introduction of color artifacts, varies inversely with the computational

Acquiring Images 17
© 2011 by Taylor & Francis Group, LLC
Figure 1.20 Example of “zipper” patterns resulting from poor interpolation in a single-chip digital
camera.

requirements. A comparison of several patented methods can be found in Ramanath (2000)

and Shao et al. (2005), Tamburino et al. (2010), and Guanara et al. (2010). The combination
of the specific color filter array arrangement and the camera’s interpolation firmware leaves
a signature in images that can be used in some cases to identify the model of camera used to
photograph a scene, even to identify specific fixed pattern noise from an individual camera,
and to detect alterations made to the image later (Bayram et al., 2006; Swaminathan et al.,
2007; Farid, 2008).
Pattern noise is not unique to single-chip cameras with a color filter array. Three-chip cameras
also have potential problems because all chips have some slight variations in the output from
individual transistors. In a three-chip system these produce different variations in the red,
green, and blue output that increase the color variations in the images.
Another approach to color camera design, developed by Foveon Corp. and used in a few
cameras, creates three transistors at each pixel location, stacked on top of each other, using
CMOS technology. Blue light penetrates the shortest distance in silicon and is detected in the
topmost transistor. Green light penetrates to the second transistor and red light penetrates
to the bottom one. The output signals are combined to produce the color information. This
approach does not suffer from loss of spatial resolution due to interpolation, but has potential
problems with consistent or accurate color fidelity.

Camera resolution
The signal coming from the silicon detector is analog, even if the digitization takes place within
the camera housing or even on the same chip, so the interpolation is done in the amplifier
stage. In most cases, the actual image resolution with a single chip camera and filter arrange-
ment is one-half to two-thirds the value that might be expected from the advertised number
of pixels in the camera, because of this interpolation. And some cameras record images with
many more stored pixels than the chip resolution warrants in any case. Such interpolation and
empty magnification contribute no additional information in the image.

18 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
Comparing cameras based on actual resolution rather than the stated number of recorded
pixels can be difficult. It is important to consider the multiple meanings of the word “pixel.”
In some contexts, it refers to the number of light detectors in the camera (without regard to
any color filtering, and sometimes including ones around the edges that do not contribute
to the actual image but are used to measure dark current). In some contexts it describes
the number of recorded brightness or color values stored in the computer, although these
may represent empty magnification. In other situations it is used to describe the displayed
points of color on the computer monitor, even if the image is shown in a compressed or
enlarged size. It makes much more sense to separate these various meanings and to talk
about resolution elements when considering real image resolution. This refers to the num-
ber of discrete points across the image that can be distinguished from each other and is
sometimes specified in terms of the number of line pairs that can be resolved. This is one-
third to one-half the number of resolution elements, since at least one element is needed
for the line and one for the space between lines. It depends on the amount of brightness
contrast between the lines and the spaces, and the amount of noise (random variation)
present in the image.
The situation is even more complicated with some digital still cameras that shift the detec-
tor array to capture multiple samples of the image. The most common method is to use
a piezo device to offset the array by half the pixel spacing in the horizontal and vertical
directions, capturing four images that can be combined to more or less double the resolu-
tion of the image as data are acquired from the gaps between the original pixel positions.
For an array with colored filters, additional shifts can produce color images with resolu-
tion approaching that corresponding to the pixel spacing. Some studio cameras displace
the entire sensor array to different regions of the film plane to collect tiles that are sub-
sequently assembled into an image several times as large as the detector array. Of course,
the multiple exposures required with these methods means that more time is required to
acquire the image.
Rather than a two-dimensional array of detectors, it is also possible to use a linear array (or
sometimes three, one each with red, green, and blue filters) that is swept across the image
plane to acquire the data. This method is common in desk-top scanners (which for many
applications are perfectly usable image acquisition devices). It has also been used in studio
cameras, and some light microscopes accomplish the same thing by moving the stage and
specimen under the optics so that an image of an entire 1 × 3 inch slide can be obtained
with high spatial resolution. The image file produced is huge; special software is required
to efficiently access the stored array (Bacus & Bacus, 2000, 2002) and to interactively deliver
a selected portion of the image data as the user varies position and magnification. Network
access to such stored images also presents bandwidth challenges, but facilitates collaboration
and teaching.
With either a single-chip or three-chip camera, the blue channel is typically the noisiest
due to the low chip sensitivity to blue light and the consequent need for greater amplifi-
cation. In many cases, processing software that reduces image noise using averaging or
median filters (discussed in Chapter 4) can be applied separately to each color channel,
using different parameters according to the actual noise content, to best improve image
appearance.
Digital cameras using the same chip technology as a video camera can produce much better
image quality. This is due in part to the longer exposure times, which collect more electrons
and so reduce noise due to statistics and amplification. Also, the slower readout of the data
from the chip, which may take a second or more instead of 1/60th of a second, reduces

Acquiring Images 19
© 2011 by Taylor & Francis Group, LLC
readout noise. Digital still cameras read out the data in one single pass (progressive scan), not
with an interlace. By cooling the chip and amplifier circuitry to reduce dark currents, integra-
tion (long exposures up to tens of seconds, or for some astronomical applications many min-
utes) can be used to advantage because of the high dynamic range (large well size and large
number of bits in the digitizer) of some chip designs. In addition, the ability to use a physical
rather than electronic shutter simplifies chip circuitry and increases fill factor. The number of
pixels in video cameras need not be any greater than the resolution of the video signal, which,
as noted above, is rather poor. In a digital still camera, very high pixel counts can give rise to
extremely high resolution, which rivals film in some cases.
There is also an interesting cross-over occurring between high end consumer and professional
scientific grade cameras. In addition to dedicated cameras for attachment to microscopes or
other separate optics, manufacturers are producing consumer single-lens reflex cameras with
enough resolution (15 to 20 million pixels at this writing) that it is becoming practical to use
them in technical applications, and simple optical attachments make it easy to connect them
to microscopes or other instruments (and of course the camera may also be removed and used
for other purposes). The camera may be tethered directly to a computer, but in many cases it
is more practical to record the images to memory chips that are later downloaded to the com-
puter. Professional digital cameras with large, high resolution detector arrays, interchangeable
lenses, etc., are providing capabilities that compete with traditional 35mm and larger film
cameras. Every manufacturer of cameras has recognized the shift away from film and toward
digital recording, and an incredibly wide variety of cameras is now available, with new devel-
opments appearing frequently.
The benefits of a camera with a large number of sensors (high pixel count), as well as large
individual sensors (large well size and consequent high dynamic range), seem obvious and
desirable. For some applications, high pixel counts are not so important. At high optical
magnification, the important limitation is the optical resolution. In the rather typical setup
of my bench microscope, with a 10x (low magnification) objective lens, the image projected
onto the chip by the transfer optics covers about 1600 µm width on the specimen. With a
100× (high magnification) objective lens that becomes 160 µm. For a camera with 3600 ×
2400 sensors (less than 10 megapixels) the low magnification image is recorded at about 1
pixel per micron, adequate for the resolution of the optics. The high magnification image
is recorded with 90 pixels per micron. Since the optical resolution of the microscope under
optimum conditions is about 0.5 µm with the 100× lens, this produces a vast and unneces-
sary oversampling. At low magnifications, or for viewing fine detail in large scenes (such as
aerial and satellite imagery), high pixel counts make sense. When the limitation on resolu-
tion lies with the optics, it may not.

Focusing
Regardless of what type of camera is employed to acquire images, it is important to focus the
optics correctly to capture the fine details in the image. Often the human eye is used to per-
form this task manually. In some situations, such as automated microscopy of pathology slides
or surveillance tracking of vehicles, automatic focusing is required. This brings computer
processing into the initial step of image capture. Sometimes, in the interests of speed, the
processing is performed in dedicated hardware circuits attached to the camera. But in many
cases the algorithms are the same as might be applied in the computer (described in Chapter

20 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
5), and the focusing is accomplished in software by stepping the optics through a range of
settings and choosing the one that gives the “best” picture.
Several different approaches to automatic focus are used. Cameras used for macroscopic
scenes may employ methods that use some distance measuring technology, e.g., using high
frequency sound or infrared light, to determine the distance to the subject so that the lens
position can be adjusted. In microscopy applications this is impractical, and the variation
with focus adjustment captured in the image itself must be used. Various algorithms are used
to detect the quality of image sharpness, and all are successful for the majority of images in
which there is good contrast and fine detail present. Each approach selects some implementa-
tion of a high-pass filter output which can be realized in various ways, using either hardware
or software, but must take into account the effect of high frequency noise in the image and
the optical transfer function of the optics (Green et al., 1985; Firestone et al., 1991; Boddeke
et al., 1994; Sun et al., 2004; Buena-Ibarra, 2005; Bueno et al., 2005; Brazdilova & Kozubek,
2009; Shim et al., 2010.)

Electronics and bandwidth limitations

Video cameras of either the solid-state chip or tube type produce analog voltage signals cor-
responding to the brightness at different points in the image. In the standard definition RS-170
signal convention, the voltage varies over a 0.7-volt range from minimum to maximum bright-
ness, as shown above in Figure 1.8. The scan is nominally 525 lines per full frame, with two
interlaced 1/60th-second fields combining to make an entire image. Only about 480 of the
scan lines are usable, with the remainder lost during vertical retrace. In a typical broadcast
television picture, more of these lines are lost due to overscanning, leaving about 400 lines in
the actual viewed area. The time duration of each scan line is 62.5 µs, part of which is used
for horizontal retrace. This leaves 52 µs for the image data, which must be subdivided into
the horizontal spacing of discernible pixels. For PAL (European) television, these values are
slightly different based on a 1/25th-second frame time and more scan lines, and the resulting
resolution is slightly higher.
Until recently in the United States, broadcast television stations were given only a 4-MHz
bandwidth for their signals, which must carry color and sound information as well as the
brightness signal. This narrow bandwidth limits the number of separate voltage values that
can be distinguished along each scan line to a maximum of 330, as mentioned above, and this
value is reduced if the signal is degraded by the electronics or by recording using standard
videotape recorders. Consumer-quality videotape recorders reduce the effective resolution
substantially; in “freeze frame” playback, they display only one of the two interlaced fields,
so that only about 200 lines are resolved vertically. Using such equipment as part of an image
analysis system makes choices of cameras or digitizer cards on the basis of resolution (actually
the number of sampled pixels) irrelevant.
There is a major difference between the interlace scan used in standard definition television
and a non-interlaced or “progressive” scan. The latter gives better quality because there
are no line-to-line alignment or shift problems. Most high definition television (HDTV)
modes use progressive scan. The format requires a higher rate of repetition of frames to
fool the human eye into seeing continuous motion without flicker, but it has many other
advantages. These include simpler logic to read data from the camera (which may be incor-
porated directly on the chip), more opportunity for data compression because of redun-
dancies between successive lines, and simpler display and storage devices. Practically all

Acquiring Images 21
© 2011 by Taylor & Francis Group, LLC
scientific imaging systems such as digital cameras, direct-scan microscopes (the scanning
electron microscope or SEM, scanning tunneling microscope or STM, the atomic force
microscope or AFM, etc.), flat-bed scanners, film or slide digitizers, and similar devices use
progressive scan.
HDTV modes include many more differences from conventional television than the use
of progressive scan. The pixel density is much higher, with a wider aspect ratio of 16:9
(instead of the 4:3 used in NTSC television) and the pixels are square. A typical HDTV
mode presents 1920 × 1080 pixel images at the rate of 60 full scans per second, for a total
data rate exceeding 2 gigabits per second, several hundred times as much data as analog
broadcast television. One consequence of this high data rate is the use of data compression
techniques, which are discussed in Chapter 3, and the use of digital transmission tech-
niques using cable or optical fiber instead of broadcast channels. Over-the-air, satellite,
and cable transmission of HDTV signals all involve compression, often with a significant
loss of image quality.
Regardless of the effects on consumer television, the development of HDTV hardware is likely
to produce spin-off effects for computer imaging, such as high pixel density cameras with pro-
gressive scan output, high bandwidth recording devices, and superior CRT or LCD displays.
For example, color cameras being designed for HDTV applications output digital rather than
analog information by performing the analog-to-digital conversion within the camera, with at
least 10 bits each for red, green, and blue.
Even the best system can be degraded in performance by such simple things as cables, con-
nectors, or incorrect termination impedance. Another practical caution in the use of standard
cameras is to avoid automatic gain or brightness compensation circuits. These can change the
image contrast or linearity in response to bright or dark regions that do not even lie within
the digitized portion of the image, make comparison between images difficult, and increase
the gain and noise for a dim signal.
Figure 1.21 shows a micrograph with its brightness histogram. This is an important tool for
image analysis, which plots the number of pixels as a function of their brightness values. It
is used extensively in subsequent chapters. The histogram shown is well spread out over the
available 256 brightness levels, with peaks corresponding to each of the structures in the
metal sample. If a bright light falls on a portion of the detector in the solid-state camera that is

Figure 1.21 A gray scale image digitized from a metallographic microscope and its brightness histo-
gram, which plots the number of pixels with each possible brightness value.

22 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
not part of the image area of interest (e.g., due to internal reflections in the optics), automatic
gain circuits in the camera may alter the brightness-voltage relationship so that the image
changes. This same effect occurs when a white or dark mask is used to surround images
placed under a camera on a copy stand. The relationship between structure and brightness is
changed, making subsequent analysis more difficult.
Issues involving color correction and calibration are dealt with below, but obtaining absolute
color information from video cameras is not possible because of the broad range of wave-
lengths passed through each filter, the variation in illumination color (even with slight voltage
changes on an incandescent bulb), and the way the color information is encoded. Matching
colors so that the human impression of color is consistent requires calibration, which is dis-
cussed in Chapter 4.
The color temperature of the illumination used is critical to matching colors in images.
Figure 1.22 shows an image recorded using filtered sunlight, with an effective color tem-
perature (described more fully in Chapter 3) of approximately 5000K, using a white card
and prior exposure to allow the camera to perform a color balance adjustment. Opening the
raw image file with different assumed color temperatures produces substantial changes in the
visual perception of the colors.
Digitization of the analog voltage signal from the detector may be done either in the camera
or in a separate external circuit (such as a “frame grabber” board placed inside the computer).
The analog signal is usually digitized with a “flash” ADC (analog-to-digital converter). This
is a chip using successive approximation techniques (described above) to rapidly sample and
measure the voltage. For video-rate imaging this must be done in less than 100 ns, producing
a number value from 0 to 255 that represents the brightness. Slower readout allows for more
than 8 bit conversion, and many digital still cameras have 12 or even 14 bit ADCs, although the
dynamic range and noise level in the detector may not be that good. The brightness number is
stored in memory and another reading made, so that a series of brightness values is obtained
along each scan line. Figure 1.23 illustrates the digitization of a signal into equal steps in
both time and value. Additional circuitry is needed to trigger each series of readings so that
positions along successive lines are consistent. Digitizing several hundred or thousand points
along each scan line, repeating the process for each line, and transmitting or storing the val-
ues into memory produces a digitized image for further processing or analysis.

Figure 1.22 An image taken with filtered sunlight and an effective color temperature of 5000K, but
stored as a raw file and opened using different assumed color temperatures. From left to right, 3500,
4500, 5500, 6500K.

Acquiring Images 23
© 2011 by Taylor & Francis Group, LLC
 Figure 1.23 Digitization of an analog
voltage signal along one line in an
image (blue) produces a series of
values that correspond to a series
of steps (red) equal in time and
rounded to integral multiples of the
smallest measurable increment.

Pixels
It is most desirable to have the spacing of the pixel values be the same in the horizontal and
vertical directions (i.e., square pixels), as this simplifies many processing and measurement
operations. There are some theoretical advantages to having pixels arranged as a hexagonal
grid, but because of the way that all acquisition hardware actually functions, and to simplify
the addressing of pixels in computer memory, this is almost never done.
Accomplishing the goal of square pixels with an analog video camera requires a well-adjusted
clock to control the acquisition. Since the standard-definition video image is not square, but
has a width-to-height ratio of 4:3, the digitized image may represent only a portion of the
entire field of view. Digitizing boards (frame grabbers) were first designed to record 512 × 512
arrays of values, since the power-of-two dimension simplified design and memory addressing.
Later generations acquired a 640 wide by 480 high array, which matched the image propor-
tions and the size of standard VGA display monitors while keeping the pixels square. Because
of the variation in clocks between cameras and digitizers, it was common to find distortions of
several percent in pixel squareness. This can be measured and compensated for after acquisi-
tion by resampling the pixels in the image, as Chapter 4 describes. Most digital still cameras
acquire images that have a width-to-height ratio of 4:3 (the aspect ratio of conventional video)
or 3:2 (the aspect ratio of 35mm film) and have square pixels.
Since pixels have a finite area, those which straddle a boundary in the scene effectively aver-
age the brightness levels of two regions and have an intermediate brightness that depends on
how the pixels lie with respect to the boundary. This means that a high lateral pixel resolution
and a large number of distinguishable gray levels are needed to accurately locate boundaries.
Figure 1.24 shows several examples of an image with varying numbers of pixels across its
width, and Figure 1.25 shows the same image with varying numbers of gray levels.
For the most common types of image acquisition devices, such as cameras, the pixels rep-
resent an averaging of the signal across a finite area of the scene or specimen. However,
there are other situations in which this is not so. At low magnification, for example, the
scanning electron microscope beam samples a volume of the specimen much smaller
than the dimension of a pixel in the image. So does the probe tip in a scanned probe

24 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
Figure 1.24 Four representations of the same image, showing a variation in the number of pixels used.
From the upper left: 256 × 256; 128 × 128; 64 × 64; 32 × 32. In all cases, a full 256 gray values are
retained. Each step in coarsening of the image is accomplished by averaging the brightness of the
region covered by the larger pixels.

microscope. Range imaging of the moon from the Clementine orbiter determined the
elevation of points about 10 cm in diameter using a laser rangefinder, but at points spaced
apart by 100 meters or more.
In these cases, the interpretation of the relationship between adjacent pixels is slightly dif-
ferent. Instead of averaging across boundaries, the pixels sample points that are discrete and
well separated. Cases of intermediate or gradually varying values from pixel to pixel are rare,
and the problem instead becomes how to locate a boundary between two sampled points on
either side. If there are many points along both sides of the boundary, and the boundary can
be assumed to have some geometric shape (such as a locally straight line), fitting methods can

Acquiring Images 25
© 2011 by Taylor & Francis Group, LLC
Figure 1.25 Four representations of the same image, with variation in the number of gray levels used.
From the upper left: 32; 16; 8; 4. In all cases, a full 256 × 256 array of pixels is retained. Each step in
the coarsening of the image is accomplished by rounding the brightness of the original pixel value.

be used to locate it to a fraction of the pixel spacing. These methods are discussed further in
Chapter 10 on image measurements.

Gray scale resolution

In addition to defining the number of sampled points along each scan line, and hence the
resolution of the image, the design of the ADC also controls the precision of each measure-
ment. High speed flash analog-to-digital converters usually measure each voltage reading to
produce an 8-bit number from 0 to 255. This full range may not be used for an actual image,
which may not vary from full black to white. Also, the quality of most analog video cameras

26 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
and other associated electronics rarely produces voltages that are free enough from electronic
noise to justify full 8-bit digitization anyway. A typical “good” camera specification of 49 dB
signal-to-noise ratio implies that only 7 bits of real information are available, and the eighth
bit is random noise. But 8 bits corresponds nicely to the most common organization of com-
puter memory into bytes, so that 1 byte of storage can hold the brightness value from 1 pixel
in the image.
High end digital still cameras and most scanners produce more than 256 distinguishable
brightness values, and for these it is common to store the data in 2 bytes or 16 bits, giving a
possible range of 65536:1, which exceeds the capability of any current imaging device (but not
some other sources of data that may be displayed as images, such as surface elevation mea-
sured with a scanned probe, a topic in Chapter 15). For a camera with a 10 or 12 bit output,
the values are shifted over to the most significant bits and the low order bits are either zero or
random values. For display and printing purposes 8 bits is enough, but the additional depth
can be very important for processing and measurement, as discussed in subsequent chapters.
In many systems the histogram of values is still expressed as 0.255 for compatibility with the
more common 8-bit range, but instead of being restricted to integers the brightness consists
of floating point values. That is the convention used in this book.
When the stored image is displayed from computer memory, the numbers are used in a dig-
ital-to-analog converter to produce voltages that control the brightness of a display monitor,
often a cathode ray tube (CRT) or liquid crystal display (LCD). This process is comparatively
noise-free and high resolution, since computer display technology has been developed to a
high level for other purposes. These displays typically have 256 steps of brightness for the red,
green, and blue signals, and when equal values are supplied to all three the result is perceived
as a neutral gray value.
The human eye cannot distinguish all 256 different levels of brightness in this type of display,
nor can they be successfully recorded or printed using ink-jet or laser printers (discussed in
Chapter 3). About 20–40 brightness levels can be visually distinguished on a CRT, LCD, or
photographic print, suggesting that the performance of the digitizers in this regard is more
than adequate, at least for those applications where the performance of the eye is enough to
begin with, or the purpose of the imaging is to produce prints.
A somewhat different situation that results in another limitation arises with images that cover
a very large dynamic range. Real-world scenes often include brightly lit areas and deep shade.
Scientific images such as SEM pictures have very bright regions corresponding to edges and
protrusions and very dark ones such as the interiors of depressions. Astronomical pictures
range from the very bright light of stars to the very dark levels of dust clouds or interstellar
space. If only 256 brightness levels are stretched to cover this entire range, there is not enough
sensitivity to small variations to reveal detail in either bright or dark areas. Capturing images
with higher bit depth, for instance 12 bits (4096 brightness levels, which is approximately the
capability of a film camera), can record the data, but it cannot be viewed successfully on a
display screen or in a print. Processing methods that can deal with such high dynamic range
images to facilitate visual interpretation are shown in Chapter 5.
Images acquired in very dim light, or some other imaging modalities such as X-ray mapping in
the scanning electron microscope (SEM), impose another limitation of the gray scale depth of
the image. When the number of photons (or other particles) collected for each image pixel is
low, statistical fluctuations and random noise become important. Figure 1.26 shows the effect
of high ASA settings (high amplifier gain) on random pixel variations in an image. The two

Acquiring Images 27
© 2011 by Taylor & Francis Group, LLC
Figure 1.26 The same image recorded with an ASA setting of 1600 (top) and 100 (bottom), showing the
increase in random pixel noise produced by higher gain in the camera.

images were recorded with the same camera and identical illumination and aperture settings;
changing the ASA setting on the camera resulted in different exposure times.
Figure 1.27 shows a fluorescence microscope image in which a single video frame illustrates
substantial statistical noise, which prevents distinguishing or measuring the structures pres-
ent. Averaging together multiple frames collects more signal and results in an improvement in
the signal- to noise-ratio, and hence in the visibility of detail.

Noise
Images in which the pixel values vary within regions that are ideally uniform in the original scene
can arise either because of limited counting statistics for the photons or other signals, losses intro-
duced in the shifting of electrons within the chip, or due to electronic noise in the amplifiers or
cabling. In any case, the variation is generally referred to as noise, and the ratio of the contrast
which is due to differences present in the scene represented by the image to the noise level is
the signal-to-noise ratio. When this is low, the features present may be invisible to the observer.
Figure 1.28 shows an example in which several features of different size and shape are super-
imposed on a noisy background with different signal-to-noise ratios. The ability to discern the
presence of the features is generally proportional to their area and independent of shape.
In the figure, a smoothing operation is performed on the image with the poorest signal-to-
noise ratio, which somewhat improves the visibility of the features. The methods available
for improving noisy images by image processing are discussed in the chapters on spatial and
frequency domain methods. However, the best approach to noisy images, when it is available,
is to collect more signal and improve the statistics.
Increasing the exposure (either by increasing the exposure time, the lens aperture, or the
illumination) reduces noise due to statistical effects, as shown in Figures 1.26 and 1.27. The
improvement in quality is proportional to the square root of the amount of light (or other

28 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
Figure 1.27 Averaging of a noisy (low photon intensity) image (light microscope image of bone mar-
row). From the upper left: one frame; averaging of 4; 16; 256 frames.

signal) collected. It is necessary to use a detector with a sufficient well size to hold the elec-
trons and to use a sufficiently high bit depth in the image to preserve the contrast details.
Cooling the detector and associated electronics chips can reduce electronic noise during long
acquisitions of many minutes. Most uncooled camera chips begin to show unacceptable pixel
noise due to dark current with integration times of more than a few seconds.
Acquiring images at video rates of 25–30 frames per second is sometimes referred to as “real
time” imaging, but of course this term should properly be reserved for any imaging rate that
is adequate to reveal temporal changes in a particular application. For some situations, time-
lapse photography may only require one frame to be taken at periods of many minutes, hours,
or even days. For others, very short exposures and high rates are needed. Special cameras
that do not use video frame rates or bandwidths can achieve rates up to ten times that of a
standard video camera for full frames and even higher for small image dimensions. These
cameras typically use a single line of detectors and optical deflection (e.g., a rotating mirror
or prism) to cover the image area.
For many applications, the repetition rate does not need to be that high. Either stroboscopic
imaging or a fast shutter speed may be enough to stop the important motion to provide a
sharp image. Electronic shutters can be used to control solid state imaging devices, instead

Acquiring Images 29
© 2011 by Taylor & Francis Group, LLC
 Figure 1.28 Features on a
noisy background:
   (a) signal-to-noise ratio 1:1;
(a)                (b)
(b) signal-to-noise ratio 1:3;
(c) s ignal-to-noise ratio 1:7;
(d) i mage (c) after spatial
smoothing.

(c)                (d)

of a mechanical shutter. Exposure times under 1/10,000th of a second can easily be achieved,
but of course this short exposure requires high illumination intensity.

High depth images

Other devices that produce data sets that are often treated as images for viewing and measure-
ment produce data with a much greater range than a camera. For instance, a scanned stylus
instrument that measures the elevation of points on a surface may have a vertical resolution of
a few nanometers with a maximum vertical travel of hundreds of micrometers, for a range-to-
resolution value of 105. This requires storing data in a format that preserves the full resolution
values, and such instruments typically use 4 bytes per pixel. An elevation map of the earth’s
surface encoded with 8 bits (256 values) spread over the range from sea level to the top of Mt.
Everest corresponds to about 100 feet per bit and would not not show most of Florida as dis-
tinct from sea level. With 2 bytes per pixel (65,536 values) each bit represents about 6 inches
and the map can distinguish the curbs along most streets. With 4 bytes per pixel (4 billion
values) each bit corresponds to less than 200 µm, and the roughness of sand on a beach can
be recorded.
In some cases with cameras having a large brightness range, the entire 12- or 14-bit depth
of each pixel is stored. However, since this depth exceeds the capabilities of most CRTs to
display, or of the user to see or print, reduction may be appropriate. If the actual bright-
ness range of the image does not cover the entire possible range, scaling (either manual or

30 The Image Processing Handbook, Sixth Edition

© 2011 by Taylor & Francis Group, LLC
automatic) to select just the range actually used can significantly reduce storage require-
ments. Many computer programs (e.g., Adobe Photoshop®) offer routines to read the raw
linear format from the camera and convert it with adjustments for lighting, vignetting, and
contrast.
In other cases, especially when performing densitometry, a conversion table is used. For den-
sitometry, the desired density value varies as the logarithm of the brightness; this is discussed
in detail in Chapter 10. A range of 256 brightness steps is not adequate to cover a typical
range from 0 to greater than 3.0 in optical density (i.e., one part in 103 of the incident illumina-
tion is transmitted) with useful precision, because at the dark end of the range, 1 step in 256
represents a very large step in optical density. Using a digitization with 12 bits (1 step in 4096)
solves this problem, but it may be efficient to convert the resulting value with a logarithmic
lookup table to store an 8-bit value (occupying a single computer byte) that represents the
optical density.
Lookup tables (LUTs) can be implemented either in hardware or software. They use the origi-
nal value as an index into a stored or precalculated table, which then provides the derived
value. This process is fast enough that acquisition is not affected. Many digital still cameras
use LUTs to convert the linear output from the detector and ADC to a value that mimics the
behavior of film. The LUTs discussed here are used for image acquisition, converting a 10-,
12-, or 14-bit digitized value with a nonlinear table to an 8-bit value that can be stored. LUTs
are also used for displaying stored images, particularly to substitute colors for gray scale
values to create pseudo-color displays, but also to apply correction curves to output devices
(displays and printers) in order to match colors. This topic is discussed later in this chapter
and in Chapter 3.
Many images do not have a brightness range that covers the full dynamic range of the digi-
tizer. The result is an image whose histogram covers only a portion of the available values
for storage or for display. Figure 1.29 shows a histogram of such an image. The flat (empty)
regions of the plot indicate brightness values at both the light and dark ends that are not
used by any of the pixels in the image. Expanding the brightness scale by spreading the his-
togram out to the full available range, as shown in the figure, may improve the visibility of
features and the perceived contrast in local structures. The same number of brightness values
are missing from the image, as shown by the gaps in the histogram, but now they are spread
uniformly throughout the range. Other ways to stretch the histogram nonlinearly are shown
in Chapter 5.

Figure 1.29 Linear expansion of a

histogram to cover the full range
of storage or display.

Acquiring Images 31
© 2011 by Taylor & Francis Group, LLC
Exploring the Variety of Random
Documents with Different Content
 "Men who, 'mid noise and dirt, and play and prate,
Could calmly mend the pen, and wash the slate."

Punishments were rare; indeed, flogging was absolutely prohibited;

and the setting an imposition would have been equally against the
genius loci, had lesson-books existed out of which to hear it
afterwards. A short imprisonment in an unfurnished room—a not
very formidable black-hole—with the loss of a goutte, now and then,
and at very long intervals, formed the mild summary of the penal
"code Pestalozzi."
It was Saturday, and a half holiday, when we arrived at Yverdun, and
oh the confusion of tongues which there prevailed! All Bedlam and
Parnassus let loose to rave together, could not have come up to that
diapason of discords with which the high corridors were ringing, as,
passing through the throng, we were conducted to the venerable
head of the establishment in his private apartments beyond. In this
gallery of mixed portraits might be seen long-haired, highborn, and
high-cheek-boned Germans; a scantling of French gamins much
better dressed; some dark-eyed Italians; Greeks in most foreign
attire; here and there a fair ingenuous Russian face; several swart
sinister-looking Spaniards, models only for their own Carravagio;
some dirty specimens of the universal Pole; one or two
unmistakeable English, ready to shake hands with a compatriot; and
Swiss from every canton of the Helvetic confederacy. To this
promiscuous multitude we were shortly introduced, the kind old man
himself taking us by the hand, and acting as master of the
ceremonies. When the whole school had crowded round to stare at
the new importation, "Here," said he, "are four English boys come
from their distant home, to be naturalised in this establishment, and
made members of our family. Boys, receive them kindly, and
remember they are henceforth your brothers." A shout from the
crowd proclaiming its ready assent and cordial participation in the
adoption, nothing remained but to shake hands à l'Anglaise, and to
fraternise without loss of time. The next day being Sunday, our
skulls were craniologically studied by Herr Schmidt, the head usher;
and whatever various bumps or depressions phrenology might have
discovered thereon were all duly registered in a large book. After
this examination was concluded, a week's furlough was allowed, in
order that Herr Schmidt might have an opportunity afforded him of
seeing how far our real character squared with phrenological
observation and measurement, entering this also into the same
ledger as a note. What a contrast were we unavoidably drawing all
this time between Yverdun and Westminster, and how enjoyable was
the change to us! The reader will please to imagine as well as he
can, the sensations of a lately pent up chrysalis, on first finding
himself a butterfly, or the not less agreeable surprise of some newly
metamorphosed tadpole, when, leaving his associates in the mud
and green slime, he floats at liberty on the surface of the pool,
endowed with lungs and a voice,—if he would at all enter into the
exultation of our feelings on changing the penitential air of Millbank
for the fresh mountain breezes of the Pays de Vaud. It seemed as if
we had—nay, we had actually entered upon a new existence, so
thoroughly had all the elements of the old been altered and
improved. If we looked back, and compared past and present
experiences, there, at the wrong end of the mental telescope, stood
that small dingy house, in that little mis-yclept Great Smith Street,
with its tiny cocoon of a bedroom, whilom our close and airless
prison; here, at the other end, and in immediate contact with the
eye, a noble chateau, full of roomy rooms, enough and to spare.
Another retrospective peep, and there was Tothill Fields, and its
seedy cricket ground; and here, again, a level equally perfect, but
carpeted with fine turf, and extending to the margin of a broad living
lake, instead of terminating in a nauseous duck-pond; while the cold
clammy cloisters adjoining Dean's Yard were not less favourably
replaced by a large open airy play-ground, intersected by two clear
trout-streams—and a sky as unlike that above Bird-Cage Walk as the
interposed atmosphere was different; whilst, in place of the startling,
discordant Keleusmata of bargees, joined to the creaking, stunning
noise of commerce in a great city, few out-of-door sounds to meet
our ear, and these few, with the exception of our own, all quiet,
pastoral, and soothing, such as, later in life, make
 "Silence in the heart
For thought to do her part,"

and which are not without their charm even to him "who whistles as
he goes for want of thought." No wonder, then, if Yverdun seemed
Paradisaical in its landscapes. Nor was this all. If the views outside
were charming, our domestic and social relations within doors were
not less pleasing. At first, the unwelcome vision of the late head-
master would sometimes haunt us, clad in his flowing black D.D.
robes—"tristis severitas in vultu, atque in verbis fides," looking as if
he intended to flog, and his words never belying his looks. That
terrible Olympian arm, raised and ready to strike, was again
shadowed forth to view; while we could almost fancy ourselves once
more at that judicial table, one of twenty boys who were to draw
lots for a "hander." How soothingly, then, came the pleasing
consciousness, breaking our reverie, that a very different person was
now our head-master—a most indulgent old man whom we should
meet ere long, with hands uplifted, indeed, but only for the purpose
of clutching us tight while he inflicted a salute on both cheeks, and
pronounced his affectionate guten morgen, liebes kind, as he
hastened on to bestow the like fatherly greeting upon every pupil in
turn.

THE DORMITORY.
The sleeping apartments at the chateau occupied three of the four
sides of its inner quadrangle, and consisted of as many long rooms,
each with a double row of windows; whereof one looked into the
aforesaid quadrangle, while the opposite rows commanded,
severally, views of the garden, the open country, and the Grande
Place of the town. They were accommodated with sixty uncurtained
stump bedsteads, fifty-nine of which afforded gîte to a like number
of boys; and one, in no respect superior to the rest, was destined to
receive the athletic form of Herr Gottlieb, son-in-law to Vater
Pestalozzi, to whose particular charge we were consigned during the
hours of the night. These bedrooms, being as lofty as they were
long, broad, and over-furnished with windows, were always
ventilated; but the in-draught of air, which was sufficient to keep
them cool during the hottest day in summer, rendered them cold,
and sometimes very cold, in the winter. In that season, accordingly,
especially when the bise blew, and hail and sleet were pattering
against the casements, the compulsory rising to class by candlelight
was an ungenial and unwelcome process; for which, however, there
being no remedy, the next best thing was to take it as coolly, we
were going to say—that of course—but, as patiently as might be.
The disagreeable anticipation of the réveil was frequently enough to
scare away sleep from our eyes a full hour before the command to
jump out of bed was actually issued. On such occasions we would lie
awake, and, as the time approached, begin to draw in our own
breath, furtively listening, not without trepidation, to the loud nose
of a distant comrade, lest its fitful stertor should startle another pair
of nostrils, on whose repose that of the whole dormitory depended.
Let Æolus and his crew make what tumult they liked inside or
outside the castle—they disturbed nobody's dreams—they never
murdered sleep. Let them pipe and whistle through every keyhole
and crevice of the vast enceinte of the building—sigh and moan as
they would in their various imprisonments of attic or corridor; howl
wildly round the great tower, or even threaten a forcible entry at the
windows, nobody's ears were scared into unwelcome consciousness
by sounds so familiar to them all. It was the expectation of a blast
louder even than theirs that would keep our eyes open—a blast
about to issue from the bed of Herr Gottlieb, and thundering
enough, when it issued, to startle the very god of winds himself!
Often, as the dreaded six A.M. drew nigh, when the third quarter
past five had, ten minutes since, come with a sough and a rattle
against the casements, and still Gottlieb slept on, we would take
courage, and begin to dream with our eyes open, that his slumbers
might be prolonged a little; his face, turned upwards, looked so
calm, the eyes so resolutely closed—every feature so perfectly at
rest. It could not be more than five minutes to six—might not he
who had slept so long, for once oversleep himself? Never! However
placid those slumbers might be, they invariably forsook our
"unwearied one" just as the clock was on the point of striking six. To
judge by the rapid twitchings—they almost seemed galvanic—first of
the muscles round the mouth, then of the nose and eyes, it
appeared as though some ill-omened dream, at that very nick of
time, was sent periodically, on purpose to awaken him; and, if so, it
certainly never returned απρακτος. Gottlieb would instantly set to
rubbing his eyes, and as the hour struck, spring up wide awake in
his shirt sleeves—thus destroying every lingering, and, as it always
turned out, ill-founded hope of a longer snooze. Presently we beheld
him jump into his small-clothes, and, when sufficiently attired to be
seen, unlimber his tongue, and pour forth a rattling broadside—Auf,
kinder! schwind!—with such precision of delivery, too, that few
sleepers could turn a deaf ear to it. But, lest any one should still lurk
under his warm coverlet out of earshot, at the further end of the
room, another and a shriller summons to the same effect once more
shakes the walls and windows of the dormitory. Then every boy
knew right well that the last moment for repose was past, and that
he must at once turn out shivering from his bed, and dress as fast as
possible; and it was really surprising to witness how rapidly all could
huddle on their clothes under certain conditions of the atmosphere!
In less than five minutes the whole school was dressed, and
Gottlieb, in his sounding shoes, having urged the dilatory with
another admonitory schwind, schwind! has departed, key and candle
in hand, to arouse the remaining sleepers, by ringing the "Great
Tom" of the chateau. So cold and cheerless was this matutinal
summons, that occasional attempts were made to evade it by
simulated headach, or, without being quite so specific, on the plea of
general indisposition, though it was well known beforehand what the
result would be. Herr Gottlieb, in such a case, would presently
appear at the bedside of the delinquent patient, with very little
compassion in his countenance, and, in a business tone, proceed to
inquire from him, Why not up?—and on receiving for reply, in a
melancholy voice, that the would-be invalid was sehr krank, would
instantly pass the word for the doctor to be summoned. That doctor
—we knew him well, and every truant knew—was a quondam French
army surgeon—a sworn disciple of the Broussais school, whose
heroic remedies at the chateau resolved themselves into one of two
—i. e., a starve or a vomit, alternately administered, according as
the idiosyncracy of the patient, or as this or that symptom turned
the scale, now in favour of storming the stomach, now of starving it
into capitulation. Just as the welcome hot mess of bread and milk
was about to be served to the rest, this dapper little Sangrado would
make his appearance, feel the pulse, inspect the tongue, ask a few
questions, and finding, generally, indications of what he would term
une légère gastrite, recommend diète absolue; then prescribing a
mawkish tisane, composed of any garden herbs at hand, and
pocketing lancets and stethoscope, would leave the patient to
recover sans calomel—a mode of treatment to which, he would tell
us, we should certainly have been subjected in our own country.
Meanwhile, the superiority of his plan of treatment was
unquestionable. On the very next morning, when he called to visit
his cher petit malade, an empty bed said quite plainly, "Very well, I
thank you, sir, and in class." But these feignings were comparatively
of rare occurrence; in general, all rose, dressed, and descended
together, just as the alarum-bell had ceased to sound; and in less
than two minutes more all were assembled in their respective class-
rooms. The rats and mice, which had had the run of these during
the night, would be still in occupation when we entered; and such
was the audacity of these vermin that none cared alone to be the
first to plant a candle on his desk. But, by entering en masse, we
easily routed the Rodentia, whose forces were driven to seek shelter
behind the wainscot, where they would scuffle, and gnaw, and
scratch, before they finally withdrew, and left us with blue fingers
and chattering teeth to study to make the best of it. Uncomfortable
enough was the effort for the first ten minutes of the session; but by
degrees the hopes of a possible warming of hands upon the surface
of the Dutch stoves after class, if they should have been lighted in
time, and at any rate the certainty of a hot breakfast, were
entertained, and brought their consolation; besides which, the being
up in time to welcome in the dawn of the dullest day, while health
and liberty are ours, is a pleasure in itself. There was no exception to
it here; for when the darkness, becoming every moment less and
less dark, had at length given way, and melted into a gray gloaming,
we would rejoice, even before it appeared, at the approach of a new
day. That approach was soon further heralded by the fitful notes of
small day-birds chirping under the leaves, and anon by their sudden
dashings against the windows, in the direction of the lights not yet
extinguished in the class-rooms. Presently the pigs were heard
rejoicing and contending over their fresh wash; then the old horse
and the shaggy little donkey in the stable adjoining the styes,
knowing by this stir that their feed was coming, snorted and brayed
at the pleasant prospect. The cocks had by this time roused their
sleepy sultanas, who came creeping from under the barn-door to
meet their lords on the dunghill. Our peacock, to satisfy himself that
he had not taken cold during the night, would scream to the utmost
pitch of a most discordant voice; then the prescient goats would
bleat from the cabins, and plaintively remind us that, till their door is
unpadlocked, they can get no prog; then the punctual magpie, and
his friend the jay, having hopped all down the corridor, would be
heard screaming for broken victuals at the school-room door, till our
dismissal bell, finding so many other tongues loosened, at length
wags its own, and then for the next hour and a half all are free to
follow their own devices. Breakfast shortly follows; but, alas! another
cold ceremony must be undergone first. A preliminary visit to pump
court, and a thorough ablution of face and hands, is indispensable to
those who would become successful candidates for that long-
anticipated meal. This bleaching process, at an icy temperature, was
never agreeable; but when the pipes happened to be frozen—a
contingency by no means unfrequent—and the snow in the yard
must be substituted for the water which was not in the pump, it
proved a difficult and sometimes a painful business; especially as
there was always some uncertainty afterwards, whether the
chilblained paws would pass muster before the inspector-general
commissioned to examine them—who, utterly reckless as to how the
boys might "be off for soap," and incredulous of what they would
fain attribute to the adust complexion of their skin, would require to
have that assertion tested by a further experiment at the "pump
head."

THE REFECTORY.
"Forbear to scoff at woes you cannot feel,
Nor mock the misery of a stinted meal."—Crabbe.

The dietary tables at the chateau, conspicuous alike for the paucity
and simplicity of the articles registered therein, are easily recalled to
mind. The fare they exhibited was certainly coarse—though, by a
euphemism, it might have been termed merely plain—and spare
withal. The breakfast would consist of milk and water—the first
aqueous enough without dilution, being the produce of certain ill-
favoured, and, as we afterwards tasted their flesh, we may add ill-
flavoured kine, whose impoverished lacteals could furnish out of
their sorry fodder no better supplies. It was London sky-blue, in
short, but not of the Alderney dairy, which was made to serve our
turn at Yverdun. This milk, at seven in summer, and at half-past
seven in winter, was transferred boiling, and as yet unadulterated,
into earthenware mixers, which had been previously half-filled with
hot water from a neighbouring kettle. In this half-and-half state it
was baled out for the assembled school into a series of pewter
platters, ranged along the sides of three bare deal boards, some
thirty feet long by two wide, and mounted on tressels, which served
us for tables. The ministering damsels were two great German
Fraus, rejoicing severally in the pleasing names of Gretchen and
Bessie. When Frau Gretchen, standing behind each boy, had dropt
her allowance of milk over his right shoulder—during which process
there was generally a mighty clatter for full measure and fair play—
the other Frau was slicing off her slices of bread from a brown loaf a
yard long, which she carried under her arm, and slashed clean
through with wonderful precision and address. It was now for all
those who had saved pocket-money for menus-plaisirs to produce
their cornets of cinnamon or sugar, sprinkle a little into the milk, and
then fall to sipping and munching with increased zest and
satisfaction. So dry and chaffy was our pain de ménage that none
ventured to soak it entire, or at once, but would cut it into
frustrums, and retain liquid enough to wash down the boluses
separately. In a few minutes every plate was completely cleaned out
and polished; and the cats, that generally entered the room as we
left it, seldom found a drop with which they might moisten their
tongues, or remove from cheeks and whiskers the red stains of
murdered mice on which they had been breaking their fast in the
great tower. So much for the earliest meal of the day, which was to
carry us through five hours, if not of laborious mental study, at least
of the incarceration of our bodies in class, which was equally
irksome to them as if our minds had been hard at work. These five
hours terminated, slates were once more insalivated and put by
clean, and the hungry garrison began to look forward to the
pleasures of the noon-day repast. The same bell that had been
calling so often to class would now give premonitory notice of
dinner, but in a greatly changed tone. In place of the shrill snappish
key in which it had all the morning jerked out each short unwelcome
summons from lesson to lesson, as if fearful of ringing one note
beyond the prescribed minute, it now would take time, vibrate far
and wide in its cage, give full scope to its tongue, and appear, from
the loud increasing swell of its prolonged oyez, to announce the
message of good cheer like a herald conscious and proud of his
commission. Ding-dong!—come along! Dinner's dishing!—ding-dong!
Da capo and encore! Then, starting up from every school-room form
throughout the chateau, the noisy boys rushed pell-mell, opened all
the doors, and, like emergent bees in quest of honey, began
coursing up and down right busily between the salle-à-manger and
the kitchen—snuffing the various aromas as they escaped from the
latter into the passage, and inferring from the amount of exhaled
fragrance the actual progress of the preparations for eating.
Occasionally some "sly Tom" would peep into the kitchen, while the
Fraus were too busy to notice him, and watch the great cauldron
that had been milked dry of its stores in the morning, now
discharging its aqueous contents of a much-attenuated bouillon—the
surface covered with lumps of swimming bread, thickened
throughout with a hydrate of potatoes, and coloured with coarse
insipid carrots, which certainly gave it a savoury appearance. It was
not good broth—far from it, for it was both sub-greasy and super-
salted; but then it was hot, it was thick, and there was an abundant
supply. It used to gush, as we have said, from the great stop-cock of
the cauldron, steaming and sputtering, into eight enormous tureens.
The shreds of beef, together with whatever other solids remained
behind after the fluid had been drawn off, were next fished up from
the abyss with long ladles, and plumped into the decanted liquor.
The young gastronome who might have beheld these proceedings
would wait till the lid was taken off the sauerkraut; and then, the
odour becoming overpoweringly appetising, he would run, as by
irresistible instinct, into the dining-room, where most of the boys
were already assembled, each with a ration of brown bread in his
hand, and ready for the Fraus, who were speedily about to enter.
The dinner was noisy and ungenteel in the extreme—how could it be
otherwise? ventre affamé n'a point d'oreilles. Hardly was the German
grace concluded, and the covers removed, when that bone of
contention, the marrow bone, was caught up by some big boy near
the top of the table, and became the signal for a general row. All in
his neighbourhood would call out second, third, fourth, fifth, &c., for
said bone; and thus it would travel from plate to plate, yielding its
contents freely to the two or three first applicants, but wholly
inadequate—unless it could have resolved itself altogether into
marrow—to meet all the demands made upon its stores. Then arose
angry words of contention, which waxed hot as the marrow waxed
cold, every candidate being equally vociferous in maintaining the
priority of his particular claim. Earnest appeals in German, French,
Spanish, English, &c., were bandied from one to the other in
consequence, as to who had really said après toi first! At last the
"dry bone" was found undeserving of further contention; and,
ceasing to drop any more fatness upon any boy's bread, the
competition for it was dropt too. When now we had half-filled our
stomachs with a soup which few physicians would have withheld
from their fever patients on the score of its strength, we threw in a
sufficiency of bread and sauerkraut to absorb it; and, after the post-
prandial German grace had been pronounced, the boys left the
table, generally with a saved crust in their pockets, to repair to the
garden and filch—if it was filching—an alliaceous dessert from the
beds, which they washed in the clear stream, and added, without
fear of indigestion, to the meal just concluded within the chateau.
Most of us throve upon this Spartan diet; but some delicate boys,
unendowed with the ostrich power of assimilation usual at that
period—for boys, like ostriches, can digest almost anything—became
deranged in their chylopoietics, and continued to feel its ill effects in
mesenteric and other chronic ailments for years afterwards. An hour
was given for stomachs to do their work, before we reassembled to
ours in the class-room. At half-past four precisely, a gouté, was
served out, which consisted of a whacking slice of bread, and either
a repetition of the morning's milk and water, or café au lait, (without
sugar "bien entendu,") or twenty-five walnuts, or a couple of ounces
of strong-tasted gruyère, or a plateful of schnitz (cuttings of dried
apples, pears, and plums). We might choose any one of these
several dainties we liked, but not more. Some dangerous characters
—not to be imitated—would occasionally, while young Frau Schmidt
stood doling out the supplies from her cupboard among the
assembled throng, make the disingenuous attempt to obtain cheese
with one hand and schnitz with the other. But the artifice, we are
happy to say, seldom succeeded; for that vigilant lady, quick-eyed
and active, and who, of all things, hated to be imposed upon, would
turn round upon the false claimant, and bid him hold up both his
hands at once—which he, ambidexter as he was, durst not do, and
thus he was exposed to the laughter and jeers of the rest. At nine,
the bell sounded a feeble call to a soi-disant supper; but few of us
cared for a basin of tisane under the name of lentil soup—or a pappy
potato, salted in the boiling—and soon after we all repaired to our
bedrooms—made a noise for a short time, then undressed, and were
speedily asleep under our duvets, and as sound, if not as musical, as
tops.
Our common fare, as the reader has now seen, was sorry enough;
but we had our Carnival and gala days as well as our Lent. Vater
Pestalozzi's birthday, in summer, and the first day of the new year,
were the most conspicuous. On each of these occasions we enjoyed
a whole week's holiday; and as these were also the periods for
slaughtering the pigs, we fed (twice a-year for a whole week!) upon
black puddings and pork à discretion, qualified with a sauce of
beetroot and vinegar, and washed down with a fluid really like small-
beer.

CLASSES.
The school-rooms, which lay immediately under the dormitories on
the ground-floor, consisted of a number of detached chambers, each
of which issued upon a corridor. They were airy—there was plenty of
air at Yverdun—and lofty as became so venerable a building; but
they were unswept, unscrubbed, peeled of their paint, and, owing to
the little light that could find its way through two very small windows
punched out of the fortress walls, presented, save at mid-day, or as
the declining sun illumined momentarily the dark recess, as
comfortless a set of interiors as you could well see. It required,
indeed, all the elasticity of youth to bear many hours' daily
incarceration in such black-holes, without participating in the
pervading gloom. Such dismal domiciles were only fit resorts for the
myoptic bat, who would occasionally visit them from the old tower;
for the twilight horde of cockroaches, which swarmed along the
floor, or the eight-eyed spiders who colonised the ceiling. The tender
sight, too, of a patient just recovering from ophthalmia would here
have required no factitious or deeper shade—but merits like these
only rendered them as ungenial as possible to the physiology and
feelings of their youthful occupants. If these apartments looked
gloomy in their dilapidations and want of sun, the sombre effect was
much heightened by the absence of the ordinary tables and chairs,
and whatever else is necessary to give a room a habitable
appearance. Had an appraiser been commissioned to make out a
complete list of the furniture and the fixtures together, a mere
glance had sufficed for the inventory. In vain would his practised eye
have wandered in quest of themes for golden sentences, printed in
such uncial characters that all who run may read; in vain for the
high-hung well-backed chart, or for any pleasing pictorial souvenirs
of Æsop or the Ark—neither these nor the long "coloured Stream of
Time," nor formal but useful views in perspective, adorned our sorry
walls. No old mahogany case clicked in a corner, beating time for the
class, and the hour up-striking loud that it should not be defrauded
of its dues. No glazed globe, gliding round on easy axis, spun under
its brassy equator to the antipodes on its sides being touched. No
bright zodiac was there to exhibit its cabalistic figures in pleasing
arabesques. In place of these and other well-known objects, here
stood a line of dirty, much-inked desks, with an equally dirty row of
attendant forms subjacent alongside. There was a scantling—it
seldom exceeded a leash—of rickety rush-bottom chairs distributed
at long intervals along the walls; a coal-black slate, pegged high on
its wooden horse; a keyless cupboard, containing the various
implements of learning, a dirty duster, a pewter plate with
cretaceous deposits, a slop-basin and a ragged sponge;—and then,
unless he had included the cobwebs of the ceiling, (not usually
reckoned up in the furniture of a room,) no other movables
remained. One conspicuous fixture, however, there was, a gigantic
Dutch stove. This lumbering parallelogram, faggot-fed from the
corridor behind, projected several feet into the room, and shone
bright in the glaze of earthenware emblazonments. Around it we
would sometimes congregate in the intervals of class: in winter to
toast our hands and hind quarters, as we pressed against the heated
tiles, with more or less vigour according to the fervency of the
central fire; and in summer either to tell stories, or to con over the
pictorial History of the Bible, which adorned its frontispiece and
sides. We cannot say that every square exactly squared with even
our schoolboy notions of propriety in its mode of teaching religious
subjects; there was a Dutch quaintness in the illustrations, which
would sometimes force a smile from its simplicity, at others shock,
from its apparent want of decorum and reverence. Preeminent of
course among the gems from Genesis, Adam and Eve, safe in
innocency and "naked truth," here walked unscathed amidst a
menagerie of wild beasts—there, dressed in the costume of their fall,
they quitted Eden, and left it in possession of tigers, bears, and
crocodiles. Hard by on a smaller tile, that brawny "knave of clubs,"
Cain, battered down his brother at the altar; then followed a long
picture-gallery of the acts of the patriarchs, and another equally long
of the acts of the apostles. But, queer as many of these
misconceptions might seem, they were nothing to the strange
attempts made at dramatising the parables of the New Testament—
e. g. a stout man, staggering under the weight of an enormous
beam which grows out of one eye, employs his fingers, assisted by
the other, to pick out a black speck from the cornea of his neighbour.
Here, an unclean spirit, as black as any sweep, issues from the
mouth of his victim, with wings and a tail! Here again, the good
Samaritan, turbaned like a Turk, is bent over the waylaid traveller,
and pours wine and oil into his wounds from the mouths of two
Florence flasks; there, the grain of mustard-seed, become a tree,
sheltering already a large aviary in its boughs; the woman, dancing
a hornpipe with the Dutch broom, has swept her house, and lo! the
piece of silver that was lost in her hand; a servant, who is digging a
hole in order to hide his lord's talent under a tree, is overlooked by a
magpie and two crows, who are attentive witnesses of the deposit:—
and many others too numerous to mention. So much for the empty
school-room, but what's a hive without bees, or a school-room
without boys? The reader who has peeped into it untenanted, shall
now, if he pleases, be introduced, dum fervet opus full and alive.
Should he not be able to trace out very clearly the system at work,
he will at least be no worse off than the bee-fancier, who hears
indeed the buzzing, and sees a flux and reflux current of his winged
confectioners entering in and passing out, but cannot investigate the
detail of their labours any farther. In the Yverdun, as in the
hymenopterus apiary, we swarmed, we buzzed, dispersed,
reassembled at the sound of the bell, flocked in and flocked out, all
the day long; exhibited much restlessness and activity, evincing that
something was going on, but what, it would have been hard to
determine. Here the comparison must drop. Bees buzz to some
purpose; they know what they are about; they help one another;
they work orderly and to one end,—

"How skilfully they build the cell,

How neat they spread the wax,
And labour hard to store it well
With the sweet food," &c. &c.

In none of these particulars did we resemble the "busy bee." This

being admitted, our object in offering a few words upon the course
of study pursued at the chateau is not with any idea of enlightening
the reader as to anything really acquired during the long ten hours'
session of each day; but rather to show how ten hours'
imprisonment may be inflicted upon the body for the supposed
advantage of the mind, and yet be consumed in "profitless labour,
and diligence which maketh not rich;" to prove, by an exhibition of
their opposites, that method and discipline are indispensable in
tuition, and (if he will accept our "pathemata" for his "mathemata"
and guides in the bringing up of his sons) to convince him that
education, like scripture, admits not of private interpretation. Those
who refuse to adopt the Catholic views of the age, and the general
sense of the society in which they live, must blame themselves if
they find the experiment of foreign schools a failure, and that they
have sent their children "farther to fare worse."
And now to proceed to the geography class, which was the first after
breakfast, and began at half-past eight. As the summons-bell
sounded, the boys came rushing and tumbling in, and ere a minute
had elapsed were swarming over, and settling upon, the high
reading-desks: the master, already at his work, was chalking out the
business of the hour; and as this took some little time to accomplish,
the youngsters, not to sit unemployed, would be assiduously
engaged in impressing sundry animal forms—among which the
donkey was a favourite—cut out in cloth, and well powdered, upon
one another's backs. When Herr G—— had finished his chalkings,
and was gone to the corner of the room for his show-perch, a
skeleton map of Europe might be seen, by those who chose to look
that way, covering the slate: this, however, was what the majority of
the assembly never dreamt of, or only dreamt they were doing. The
class generally—though ready when called upon to give the efficient
support of their tongues—kept their eyes to gape elsewhere, and,
like Solomon's fool, had them where they had no business to be.
The map, too often repeated to attract from its novelty, had no claim
to respect on other grounds. It was one of a class accurately
designated by that careful geographer, old Homer, as "μαπς ου Κατα
Κοσμον." Coarse and clumsy, however, as it necessarily would be, it
might still have proved of service had the boys been the
draughtsmen. As it was, the following mechanically Herr G——'s
wand to join in the general chorus of the last census of a city, the
perpendicular altitude of a mountain, or the length and breadth of a
lake, could obviously convey no useful instruction to any one. But,
useful or otherwise, such was our regime,—to set one of from fifty
to sixty lads, day after day, week after week, repeating facts and
figures notorious to every little reader of penny guides to science, till
all had the last statistical returns at their tongue's tip; and knew,
when all was done, as much of what geography really meant as on
the day of their first matriculation. Small wonder, then, if some
should later have foresworn this study, and been revolted at the
bare sight of a map! All our recollections of map, unlike those of
personal travel, are sufficiently distasteful. Often have we yawned
wearily over them at Yverdun, when our eyes were demanded to
follow the titubations of Herr G——'s magic wand, which, in its
uncertain route, would skip from Europe to Africa and back again—
qui modo Thebas modo me ponit Athenis; and our dislike to them
since has increased amazingly. Does the reader care to be told the
reason of this? Let him—in order to obtain the pragmatic sanction of
some stiff-necked examiner—have to "get up" all the anastomosing
routes of St Paul's several journeyings; have to follow those
rebellious Israelites in all their wanderings through the desert; to
draw the line round them when in Palestine; going from Dan to
Beersheba, and "meting out the valley of Succoth;" or, finally, have
to cover a large sheet of foolscap with a progressive survey of the
spread of Christianity during the three first centuries—and he will
easily enter into our feelings. To return to the class-room: The
geographical lesson, though of daily infliction, was accurately
circumscribed in its duration. Old Time kept a sharp look-out over his
blooming daughters, and never suffered one hour to tread upon the
heels or trench upon the province of a sister hour. Sixty minutes to
all, and not an extra minute to any, was the old gentleman's
impartial rule; and he took care to see it was strictly adhered to. As
the clock struck ten, geography was shoved aside by the muse of
mathematics. A sea of dirty water had washed out in a twinkling all
traces of the continent of Europe, and the palimpsest slate
presented a clean face for whatever figures might next be traced
upon it.
The hour for Euclidising was arrived, and anon the black
parallelogram was intersected with numerous triangles of the
Isosceles and Scalene pattern; but, notwithstanding this promising
début, we did not make much quicker progress here than in the
previous lesson. How should we, who had not only the difficulties
inseparable from the subject to cope with, but a much more
formidable difficulty—viz. the obstruction which we opposed to each
other's advance, by the plan, so unwisely adopted, of making all the
class do the same thing, that they might keep pace together. It is a
polite piece of folly enough for a whole party to be kept waiting
dinner by a lounging guest, who chooses to ride in the park when he
ought to be at his toilet; but we were the victims of a much greater
absurdity, who lost what might have proved an hour of profitable
work, out of tenderness to some incorrigibly idle or Bœotian boy,
who could not get over the Pons Asinorum, (every proposition was a
pons to some asinus or other,) and so made those who were over
stand still, or come back to help him across. Neither was this,
though a very considerable drawback, our only hindrance—the
guides were not always safe. Sometimes he who acted in that
capacity would shout "Eureka" too soon; and having undertaken to
lead the van, lead it astray till just about, as he supposed, to come
down upon the proof itself, and to come down with a Q. E. D.: the
master would stop him short, and bid him—as Coleridge told the
ingenious author of Guesses at Truth—"to guess again." But suppose
the "guess" fortunate, or that a boy had even succeeded, by his own
industry or reflection, in mastering a proposition, did it follow that he
would be a clear expositor of what he knew? It was far otherwise.
Our young Archimedes—unacquainted with the terms of the science,
and being also (as we have hinted) lamentably defective in his
knowledge of the power of words—would mix up such a "farrago" of
irrelevancies and repetitions with the proof, as, in fact, to render it to
the majority no proof at all. Euclid should be taught in his own
words,—just enough and none to spare: the employment of less
must engender obscurity; and of more, a want of neatness and
perspicacity. The best geometrician amongst us would have cut but
a bad figure by the side of a lad of very average ability brought up
to know Euclid by book.
Another twitch of the bell announced that the hour for playing at
triangles had expired. In five minutes the slate was covered with
bars of minims and crotchets, and the music lesson begun. This, in
the general tone of its delivery, bore a striking resemblance to the
geographical one of two hours before; the only difference being that
"ut, re, me" had succeeded to names of certain cities, and "fa, so,
la" to the number of their inhabitants. It would be as vain an
attempt to describe all the noise we made as to show its rationale or
motive. It was loud enough to have cowed a lion, stopped a donkey
in mid-bray—to have excited the envy of the vocal Lablache, or to
have sent any prima donna into hysterics. When this third hour had
been bellowed away, and the bell had rung unheard the advent of a
fourth—presto—in came Mons. D——, to relieve the meek man who
had acted as coryphæus to the music class; and after a little
tugging, had soon produced from his pocket that without which you
never catch a Frenchman—a thème. The theme being announced,
we proceeded (not quite tant bien que mal) to scribble it down at his
dictation, and to amend its orthography afterwards from a corrected
copy on the slate. Once more the indefatigable bell obtruded its
tinkle, to proclaim that Herr Roth was coming with a Fable of Gellert,
or a chapter from Vater Pestalozzi's serious novel, Gumal und Lina,
to read, and expound, and catechise upon. This last lesson before
dinner was always accompanied by frequent yawns and other
unrepressed symptoms of fatigue; and at its conclusion we all rose
with a shout, and rushed into the corridors.
On resuming work in the afternoon, there was even less attention
and method observed than before. The classes were then broken up,
and private lessons were given in accomplishments, or in some of
the useful arts. Drawing dogs and cows, with a master to look after
the trees and the hedges; whistling and spitting through a flute;
playing on the patience of a violin; turning at a lathe; or fencing with
a powerful maître d'armes;—such were the general occupations. It
was then, however, that we English withdrew to our Greek and
Latin; and, under a kind master, Dr M——, acquired (with the
exception of a love for natural history, and a very unambitious turn
of mind) all that really could deserve the name of education.
We have now described the sedentary life at the chateau. In the
next paper the reader shall be carried to the gymnasium; the drill
ground behind the lake; to our small menageries of kids, guinea
pigs, and rabbits; be present at our annual ball and skating bouts in
winter, and at our bathings, fishings, frog-spearings, and rambles
over the Jura in summer.

FOOTNOTES:
[14] Cicero, De Fin., ii. 1.
 THE CROWNING OF THE COLUMN,
AND CRUSHING OF THE PEDESTAL.
It was said in the debate on the Navigation Laws, in the best speech
made on the Liberal side, by one of the ablest of the Liberal party,
that the repeal of the Navigation Laws was the crowning of the
column of free trade. There is no doubt it was so; but it was
something more. It was not only the carrying out of a principle, but
the overthrow of a system; it was not merely the crowning of the
column, but the crushing of the pedestal.
And what was the system which was thus completely overthrown,
for the time at least, by this great triumph of Liberal doctrines? It
was the system under which England had become free, and great,
and powerful; under which, in her alone of all modern states, liberty
had been found to coexist with law, and progress with order; under
which wealth had increased without producing divisions, and power
grown up without inducing corruption; the system which had
withstood the shocks of two centuries, and created an empire
unsurpassed since the beginning of the world in extent and
magnificence. It was a system which had been followed out with
persevering energy by the greatest men, and the most commanding
intellects, which modern Europe had ever produced; which was
begun by the republican patriotism of Cromwell, and consummated
by the conservative wisdom of Pitt; which had been embraced alike
by Somers and Bolingbroke, by Walpole and Chatham, by Fox and
Castlereagh; which, during two centuries, had produced an
unbroken growth of national strength, a ceaseless extension of
national power, and at length reared up a dominion which embraced
the earth in its grasp, and exceeded anything ever achieved by the
legions of Cæsar, or the phalanx of Alexander. No vicissitudes of
time, no shock of adverse fortune, had been able permanently to
arrest its progress. It had risen superior alike to the ambition of
Louis XIV. and the genius of Napoleon; the rude severance of the
North American colonies had thrown only a passing shade over its
fortunes; the power of Hindostan had been subdued by its force, the
sceptre of the ocean won by its prowess. It had planted its colonies
in every quarter of the globe, and at once peopled with its
descendants a new hemisphere, and, for the first time since the
creation, rolled back to the old the tide of civilisation. Perish when it
may, the old English system has achieved mighty things; it has
indelibly affixed its impress on the tablets of history. The children of
its creation, the Anglo-Saxon race, will fill alike the solitudes of the
Far West, and the isles of the East; they will be found equally on the
shores of the Missouri, and on the savannahs of Australia; and the
period can already be anticipated, even by the least imaginative,
when their descendants will people half the globe.
It was not only the column of free trade which has been crowned in
this memorable year. Another column, more firm in its structure,
more lasting in its duration, more conspicuous amidst the wonders
of creation, has, in the same season, been crowned by British hands.
While the sacrilegious efforts of those whom it had sheltered were
tearing down the temple of protection in the West, the last stone
was put to the august structure which it had reared in the East. The
victory of Goojerat on the Indus was contemporary with the repeal
of the Navigation Laws on the Thames. The completion of the
conquest of India occurred exactly at the moment when the system
which had created that empire was repudiated. Protection placed the
sceptre of India in our hands, when free trade was surrendering the
trident of the ocean in the heart of our power. With truth did Lord
Gough say, in his noble proclamation to the army of the Punjaub, on
the termination of hostilities, that "what Alexander had attempted
they had done." Supported by the energy of England, guided by the
principles of protection, restrained by the dictates of justice, backed
by the navy which the Navigation Laws had created, the British arms
had achieved the most wonderful triumph recorded in the annals of
mankind. They had subjugated a hundred and forty millions of men
in the Continent of Hindostan, at the distance of ten thousand miles
from the parent state; they had made themselves felt alike, and at
the same moment, at Nankin, the ancient capital of the Celestial
Empire, and at Cabool, the cradle of Mahommedan power.
Conquering all who resisted, blessing all who submitted, securing
the allegiance of the subjects by the justice and experienced
advantages of their government, they had realised the boasted
maxim of Roman administration—

"Parcere subjectis et debellare superbos,"

and steadily advanced through a hundred years of effort and glory,

not unmixed with disaster, from the banks of the Hoogley to the
shores of the Indus—from the black hole of Calcutta to the throne of
Aurengzebe.
"Nulla magna civitas," said Hannibal, "diu quiescere potest—si foris
hostem non habet, domi invenit: ut praevalida corpora ab externis
causis tuta videntur, suis ipsis viribus conficiuntur."[15] When the
Carthaginian hero made this mournful reflection on the infatuated
spirit which had seized his own countrymen, and threatened to
destroy their once powerful dominion, he little thought what a
marvellous confirmation of it a future empire of far greater extent
and celebrity was to afford. That the system of free trade—that is,
the universal preference of foreigners, for the sake of the smallest
reduction of price, to your own subjects—must, if persisted in, lead
to the dismemberment and overthrow of the British empire, cannot
admit of a moment's doubt, and will be amply proved to every
unbiassed reader in the sequel of this paper. Yet the moment chosen
for carrying this principle into effect was precisely that, when the
good effects of the opposite system had been most decisively
demonstrated, and an empire unprecedented in magnitude and
magnificence had reached its acme under its shadow. It would be
impossible to explain so strange an anomaly, if we did not recollect
how wayward and irreconcilable are the changes of the human
mind: that action and reaction is the law not less of the moral than
of the material world; that nations become tired of hearing a policy
called wise, not less than an individual called the just; and that if a
magnanimous and truly national course of government has been
pursued by one party long in possession of power, this is quite
sufficient to make its opponents embrace the opposite set of tenets,
and exert all their influence to carry them into effect when they
succeed to the direction of affairs, without the slightest regard to the
ruin they may bring on the national fortunes.
The secret of the long duration and unexampled success of the
British national policy is to be found in the protection which it
afforded to all the national interests. But for this, it must long since
have been overthrown, and with it the empire which was growing up
under its shadow. No institutions or frames of government can long
exist which are not held together by that firmest of bonds,
experienced benefits. What made the Roman power steadily
advance during seven centuries, and endure in all a thousand years?
The protection which the arms of the legions afforded to the
industry of mankind, the international wars which they prevented,
the general peace they secured, the magnanimous policy which
admitted the conquered states to the privileges of Roman citizens,
and caused the Imperial government to be felt through the wide
circuit of its power, only by the vast market it opened to the industry
of its multifarious subjects, and the munificence with which local
undertakings were everywhere aided by the Imperial treasury. Free
trade in grain at length ruined it: the harvests of Libya and Egypt
came to supersede those of Greece and Italy,—and thence its fall. To
the same cause which occasioned the rise of Rome, is to be ascribed
the similar unbroken progress of the Russian territorial dominion,
and that of the British colonial empire in modern times. What, on the
other hand, caused the conquests of Timour and Charlemagne,
Alexander the Great and Napoleon, to be so speedily obliterated,
and their vast empires to fall to pieces the moment the powerful
hand which had created them was laid in the dust? The want of
protection to general interests, the absence of the strong bond of
experienced benefits; the oppressive nature of the conquering
Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.

More than just a book-buying platform, we strive to be a bridge

connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.

Join us on a journey of knowledge exploration, passion nurturing, and

personal growth every day!

ebookbell.com