Full download ebook at ebookgate.

com

The Image processing handbook Seventh Edition

Russ

https://ebookgate.com/product/the-image-
processing-handbook-seventh-edition-russ/

Download more ebook from https://ebookgate.com

More products digital (pdf, epub, mobi) instant
download maybe you interests ...

The Image Processing Handbook 5th Edition John C. Russ

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

The Image Processing Handbook 7th Edition John C. Russ

https://ebookgate.com/product/the-image-processing-handbook-7th-
edition-john-c-russ/

Information Fusion in Signal and Image Processing

Digital Signal and Image Processing 1st Edition
Isabelle Bloch

https://ebookgate.com/product/information-fusion-in-signal-and-
image-processing-digital-signal-and-image-processing-1st-edition-
isabelle-bloch/

Image Analysis of Food Microstructure 1st Edition John

C. Russ (Author)

https://ebookgate.com/product/image-analysis-of-food-
microstructure-1st-edition-john-c-russ-author/
Logic based Nonlinear Image Processing Stephen Marshall

https://ebookgate.com/product/logic-based-nonlinear-image-
processing-stephen-marshall/

Image Processing with ImageJ 2nd Edition Jurjen Broeke

https://ebookgate.com/product/image-processing-with-imagej-2nd-
edition-jurjen-broeke/

Image and Video Processing in the Compressed Domain

Jayanta Mukhopadhyay

https://ebookgate.com/product/image-and-video-processing-in-the-
compressed-domain-jayanta-mukhopadhyay/

Image and Video Processing in the Compressed Domain

Jayanta Mukhopadhyay

https://ebookgate.com/product/image-and-video-processing-in-the-
compressed-domain-jayanta-mukhopadhyay-2/

Image Processing Principles and Applications 1st

Edition Tinku Acharya

https://ebookgate.com/product/image-processing-principles-and-
applications-1st-edition-tinku-acharya/
SE VE N T H E D I T I O N

© 2016 by Taylor & Francis Group, LLC

© 2016 by Taylor & Francis Group, LLC
SE VE N T H E D I T I O N

JOHN C. RUSS • F. BRENT NEAL

Boca Raton London New York

CRC Press is an imprint of the

Taylor & Francis Group, an informa business

© 2016 by Taylor & Francis Group, LLC

CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2016 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Version Date: 20150804

International Standard Book Number-13: 978-1-4987-4028-9 (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

© 2016 by Taylor & Francis Group, LLC

Contents

Introduction . . . . . . . . . . . . . . . . . . . . . . xiii
About this text.................................................. xiii
A word of caution............................................. xv
A personal note............................................... xvi

1 Acquiring Images . . . . . . . . . . . . . . . . 1
Human reliance on images.................................. 1
Extracting information......................................... 4
Video cameras................................................... 6
CCD cameras.................................................... 8
CMOS detectors...............................................12
Camera artifacts and limitations..........................13
Color cameras..................................................15
Camera resolution.............................................18
Electronics and bandwidth limitations.................. 20
Handling color data..........................................21
Color encoding................................................ 22
Other image sources..........................................24
Pixels.............................................................. 25
Tonal resolution................................................ 29
The image contents........................................... 30
Camera limitations.............................................31
Noise............................................................. 33
High-­depth images............................................34
Focusing.......................................................... 35
Color displays.................................................. 36
Image types......................................................37

v
© 2016 by Taylor & Francis Group, LLC
 Multiple images............................................... 40
Imaging requirements........................................ 44

2 Printing and Storage. . . . . . . . . . . . . 47

Hard copies......................................................47
Halftoning....................................................... 49
Dots on paper...................................................52
Color printing.................................................. 55
Adding black—CMYK....................................... 58
Printing hardware............................................. 62
Film recorders.................................................. 66
Presentation tools...............................................67
File storage.......................................................67
Storage media................................................. 69
Magnetic recording...........................................70
Databases for images........................................72
Searching by content.........................................76
Browsing and thumbnails....................................79
File formats.......................................................81
Lossless coding................................................. 82
Reduced color palettes...................................... 86
JPEG compression.............................................87
Wavelet compression........................................ 93
Fractal compression.......................................... 96
Digital movies.................................................. 98

3 Human Vision. . . . . . . . . . . . . . . . . 101

What we see and why.....................................101
Recognition....................................................104
Technical specs...............................................107
Seeing color...................................................108
Acuity............................................................ 112
What the eye tells the brain.............................. 116
Spatial comparisons........................................ 118
Local to global hierarchies................................124
Grouping.......................................................127
It’s about time.................................................130
The third dimension..........................................134
How versus what.............................................138
Seeing what isn’t there, and vice versa...............140
Image compression..........................................142
A world of light...............................................144
Size matters....................................................148
Shape (whatever that means)............................152
Context..........................................................154
Arrangements must be made.............................156
Seeing is believing..........................................158
Learning more.................................................162

vi Contents
© 2016 by Taylor & Francis Group, LLC
4 Correcting Imaging Defects . . . . . . . 163
Color adjustments............................................164
Hue, saturation, intensity..................................166
Other spaces..................................................168
Color correction.............................................. 171
Noisy images................................................. 174
Neighborhood averaging................................. 178
Gaussian smoothing........................................ 181
Neighborhood ranking....................................185
The color median............................................190
More median filters..........................................193
Weighted, conditional, and adaptive
neighborhoods..............................................196
Other neighborhood noise reduction methods.... 204
Defect removal, maximum entropy, and
maximum likelihood...................................... 208
Nonuniform illumination................................... 214
Fitting a background function............................ 217
Rank leveling..................................................222
Color images..................................................225
Nonplanar views.............................................227
Computer graphics..........................................228
Geometric distortion........................................230
Alignment.......................................................234
Interpolation...................................................236
Morphing.......................................................241

5 Image Enhancement in the Spatial

Domain. . . . . . . . . . . . . . . . . . . . . . 243
Purposes for enhancement................................243
Contrast expansion..........................................244
False color lookup tables (LUTs).........................248
Contrast manipulation......................................250
Histogram equalization....................................253
Contrast in color images...................................255
Local equalization...........................................258
Laplacian sharpening.......................................261
The unsharp mask............................................269
Derivatives......................................................272
Edges and gradients........................................276
Edge orientation..............................................279
More edge detectors.......................................284
Rank-­based methods....................................... 290
Texture...........................................................294
Implementation notes.......................................298
Image math....................................................299
Subtracting images......................................... 300
Multiplication and division............................... 303

Contents vii
© 2016 by Taylor & Francis Group, LLC
 Principal component analysis........................... 305
Principal component analysis for contrast
enhancement................................................ 310
Other image combinations............................... 313
Cross- ­correlation............................................. 317

6 Processing Images in Frequency

Space . . . . . . . . . . . . . . . . . . . . . . . 321
About frequency space....................................321
The Fourier transform.......................................322
Fourier transforms of simple functions.................325
Moving to two dimensions................................327
Frequencies and spacings................................331
Preferred orientation........................................333
Texture and fractals.........................................336
Removing selected frequencies..........................338
Periodic noise removal.....................................345
Selection of periodic information.......................349
Convolution....................................................352
Deconvolution.................................................354
Noise and Wiener deconvolution......................358
Other deconvolution methods...........................362
Additional notes on deconvolution.....................365
Template matching and correlation....................368
Autocorrelation...............................................373
Wavelets........................................................ 374

7 Segmentation and Thresholding. . . . 381

Brightness thresholding.....................................381
Automatic settings...........................................384
Multiband images...........................................391
Color thresholding...........................................395
Thresholding from texture................................. 400
Multiple thresholding criteria............................ 402
Textural orientation......................................... 403
Region boundaries......................................... 406
Noise and overlaps........................................ 408
Selecting smooth boundaries............................ 411
Conditional histograms..................................... 413
Boundary lines................................................ 415
Contours........................................................420
Cluster analysis...............................................423
More segmentation methods............................ 430
Image representation.......................................434

8 Processing Binary Images. . . . . . . . 439

Boolean operations.........................................439
Combining Boolean operations.........................442

viii Contents
© 2016 by Taylor & Francis Group, LLC
 Masks............................................................447
From pixels to features.................................... 449
Filling holes................................................... 450
Measurement grids..........................................452
Boolean logic with features...............................454
Selecting features by location...........................458
Double thresholding.........................................462
Erosion and dilation.........................................463
Opening and closing.......................................465
Isotropy..........................................................469
Measurements using erosion and dilation...........470
Extension to grayscale images...........................473
Neighborhood parameters............................... 474
Examples of use..............................................476
Euclidean distance map...................................479
Watershed segmentation................................. 482
Ultimate eroded points.................................... 488
Skeletons....................................................... 490
Topology........................................................492
Boundary lines................................................497
Combining skeleton and Euclidean distance
map............................................................ 500

9 Image Measurements . . . . . . . . . . . 505

Photogrammetry............................................. 505
Comparisons.................................................. 511
Global measurements...................................... 518
Volume..........................................................520
Surface area...................................................525
Grain size......................................................530
Multiple surfaces.............................................533
Length............................................................535
Thickness........................................................538
Sampling strategies.........................................539
Determining number........................................543
Curvature, connectivity, and the Disector............546
Anisotropy and gradients.................................551
Size distribution...............................................554
Classical stereology (unfolding).........................555

10 Feature Measurements . . . . . . . . . . 559

Brightness measurements..................................559
Density...........................................................561
Brightness profiles............................................563
Color values...................................................566
Determining location........................................568
Orientation.....................................................573
Neighbor relationships.....................................575

Contents ix
© 2016 by Taylor & Francis Group, LLC
 Separation distance.........................................579
Alignment.......................................................582
The linear Hough transform...............................586
The circular Hough transform............................588
Counting........................................................592
Special counting procedures.............................597
Feature size................................................... 602
Circles and ellipses..........................................607
Caliper dimensions......................................... 609
Perimeter........................................................ 612

11 Characterizing Shape . . . . . . . . . . . 619

Describing shape............................................ 619
Dimensionless ratios.........................................624
Effects of orientation........................................628
“Like a circle”................................................. 630
An example: Leaves.........................................633
Topology and the skeleton............................... 640
Boundaries.....................................................647
Shock graphs..................................................649
Fractal dimension............................................651
Measurement techniques..................................653
Harmonic analysis.......................................... 660
Chain code....................................................662
An example: Arrow points................................665
Wavelets........................................................667
Moments........................................................672
An example: Dandelion...................................676
Zernike moments.............................................677
Landmarks......................................................679

12 Correlation, Classification,
Identification, and Matching . . . . . . 683
A variety of purposes...................................... 683
Matching.......................................................685
Cross- ­correlation.............................................689
Curvature scale space......................................692
Classification..................................................696
Distributions and decision points........................697
Linear discriminant analysis (LDA) and principal
component analysis (PCA)............................. 700
Class definition...............................................702
Unsupervised learning......................................709
Are groups different?....................................... 712
Neural nets..................................................... 716
k-­Nearest neighbors.........................................720
Parametric description......................................722
Bayesian statistics............................................724

x Contents
© 2016 by Taylor & Francis Group, LLC
 A comparison.................................................726
Harmonic analysis and invariant moments..........729
Species examples............................................732
Correlation.....................................................736
Landmark data................................................743

13 3D Imaging. . . . . . . . . . . . . . . . . . . 749
More than two dimensions................................749
Volume imaging versus sections.........................750
Serial sections.................................................754
Removing layers..............................................756
Reconstruction.................................................760
Confocal microscopy.......................................762
Stereo viewing................................................764
Tomography................................................... 767
Tomographic reconstruction..............................770
Reconstruction artifacts.....................................774
Algebraic reconstruction...................................776
Maximum entropy...........................................780
Imaging geometries......................................... 781
Other signals..................................................783
Beam hardening and other issues......................786
3D tomography...............................................790
Dual energy methods.......................................795
Microtomography............................................799
3D reconstruction and visualization................... 803
Slices and surfaces......................................... 806
Marching cubes.............................................. 810
Volumetric displays.......................................... 813
Ray tracing..................................................... 815

14 3D Processing and Measurement. . . 819

Processing voxel arrays.................................... 819
When the z-­axis is different...............................824
Multiple image sets..........................................828
Thresholding and segmentation........................ 830
Morphological operations and structural
measurements...............................................837
Skeletons....................................................... 840
Surface and volume.........................................847
Quantitative use of reconstructions.....................849
Methods for object measurements.....................852
Size...............................................................854
Examples of object measurements.....................857
Other object measurements..............................861
Limitations......................................................867
Industrial applications..................................... 868
Comparison to stereological measurements.........878

Contents xi
© 2016 by Taylor & Francis Group, LLC
 Spherical harmonics, wavelets, and fractal
dimension.....................................................881
Other applications and future possibilities.......... 888

15 Imaging Surfaces . . . . . . . . . . . . . . 891

Producing surfaces...........................................891
Imaging by physical contact.............................894
Noncontacting measurements...........................898
Shape from shading and polynomial texture
map............................................................ 902
Microscopy of surfaces................................... 906
Stereoscopy................................................... 908
Matching points.............................................. 913
Composition imaging....................................... 917
Processing of range images.............................. 919
Processing of composition maps........................923
Data presentation and visualization...................924
Surface rendering............................................928
Measurements.................................................931
Profiles...........................................................933
Representing elevation data..............................937
The surface measurement suite..........................940
Hybrid properties............................................943
Topographic analysis.......................................945
Fractal dimensions...........................................949

References . . . . . . . . . . . . . . . . . . . . . . 957
Index . . . . . . . . . . . . . . . . . . . . . . . . . 1013

xii Contents
© 2016 by Taylor & Francis Group, LLC
Introduction

About this text

Image processing is used in a wide variety of applications, primarily for two somewhat
different purposes:

1. improving the visual appearance of images for a human observer, including their
printing and transmission, and
2. preparing images for the measurement and analysis of the features and struc-
tures that they reveal.

The techniques appropriate for each of these tasks are not always the same, but there
is considerable overlap, and this book explains, illustrates, and compares methods used
for both purposes. To get the best possible results, it is important to know about the
intended uses of the processed images. For visual enhancement, this means having some
familiarity with the human visual process and an appreciation of what cues the viewer
responds to or normally overlooks in images. The chapter on human vision addresses
those issues. It also is useful to know about printing and storage methods, since many
images are processed in the context of reproduction, storage, or transmission.
This handbook presents and illustrates an extensive collection of image processing tools
to help the user and prospective user of computer-based systems understand the meth-
ods provided in various software packages and determine those steps that may be best
suited for particular applications. Comparisons are presented for different algorithms
that may be employed for similar purposes, using a selection of representative pictures
from various microscopy techniques, as well as macroscopic, forensic, remote sensing,
and astronomical images. Throughout the text, a conscious effort has been made to
include examples of image processing and analysis from a wide variety of disciplines,
and at all scales, from the nano- to astro-, including real-world macroscopic images.
It is very important to emphasize that the scale of an image matters very little to the
techniques used to process or analyze it. Microscopes that have a resolution of nano-
meters and telescopes that produce images covering light years produce images that

xiii
© 2016 by Taylor & Francis Group, LLC
require many of the same algorithms. People trying to use image processing and analysis
methods for their particular area of interest should understand that the same basic tools
are useful at other scales and in other disciplines, and that a solution to their problems
may already exist just down the street. It may also 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 may be more appealing to the senses, and
some may make the information more accessible, but these two goals might not call for
identical methods.
The measurement of images is often a principal method for acquiring scientific and
forensic data, and generally requires that objects or structure be well defined, either
by edges or by unique brightness, color, texture, or some combination of these factors.
The types of measurements 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 are
subjects that humans generally understand, although human vision is not quantitative
and is easily fooled. Shape is a more subtle concept, dealt with in a separate chapter.
Measurement data may be used for classification or recognition of objects. There are sev-
eral different strategies that can be applied, and examples are shown. The topics 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 mas-
tery of these techniques as being much like becoming a skilled journeyman carpenter.
The number of distinct tools (saws, planes, drills, etc.) is relatively small, and although
there are some variations (slotted or Phillips-head screwdrivers, or saw blades with fine
or coarse teeth, for example) knowing how to use each type of tool is closely linked
to understanding and visualizing what it can do. With a set of these tools, the skilled
craftsman can produce a house, a boat, or a piece of furniture. So it is with image pro-
cessing tools, which are conveniently grouped into only a few classes, such as histogram
modification, neighborhood operations, Fourier-space processing, and so on, that can be
used to accomplish a broad range of purposes. Visiting your local hardware store and
purchasing the appropriate woodworking tools does not provide the skills to use them.
Understanding their use requires practice, which develops the ability to visualize before-
hand what each will do. The same is true of the tools for image processing.
The emphasis throughout this seventh edition continues to be explaining and illustrating
methods so that they can be clearly understood, rather than providing dense mathemat-
ics and derivations. There are excellent texts on Fourier transforms, image compression,
mathematical morphology, stereology, and so on that provide all of the equations and
rigor that may be desired; many of them, as well as original publications, are referenced
here. But the thrust of this book remains teaching by example. Few people learn the
principles of image processing from equations. Just as we use images to communicate
ideas and to “do science,” so most of us rely on images to learn about 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 should
need to look up the mathematical foundations or the computer code, they will be easier
to understand.
This edition includes a greater range of “high end” or computationally intensive algo-
rithms than previous versions. With the continuing increase in the power and speed of
desktop and laptop computers, more of these methods are now practical for most users,

xiv Introduction
© 2016 by Taylor & Francis Group, LLC
and consequently more of the available software programs tend to include them. The
algorithms themselves are not necessarily new, but they have become more accessible.
However, the simpler tools that have been available to most users for decades are still
viable, and sometimes give equal or even superior results.

A word of caution
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 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 relevant 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.
The word “photoshopping” has become an everyday expression, with generally nega-
tive connotations. Most scientific publications and the editors who review submitted
papers have become very aware of the ease with which images can be processed or cre-
ated, and the dangers of inadequate documentation. For example, see M. Rossner and
K. M. Yamada’s “What’s in a Picture?” ( J. Cell Biology 166:11–15, 2004) for the Journal
of Cell Biology’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). This 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 to a nontechnical jury. This frequently requires showing
that the details obtained from the image are present in the original but become visually
more evident and measurable 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 potentially misleading and usually wrong. Some such as using copy-and-paste to
insert something into 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 in the first place) can create a false impression. A general guideline to be consid-
ered 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 existing details more
accessible, either visually for presentation and communication, or to facilitate measure-
ment. Of course, the steps used must be documented and reported, and it is better to use
an algorithmic procedure than to manually tweak settings until the results “look good.”
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 understanding 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.

Introduction xv
© 2016 by Taylor & Francis Group, LLC
The scientist who does not understand his or her instruments and what they do risks
making serious errors. But after acquiring an image, some are content to use software,
possibly downloaded free from the web, to apply algorithms they do not understand and
have not tested. The results can be misleading or worse. An important but often over-
looked concern is the importance of avoiding the use of programs that alter the image
without the user being aware of it. For example, placing an image into a slide presenta-
tion, web page, or word processing document may alter colors, discard pixels, and intro-
duce unwanted compression. Saving an image with a lossy compression method such
as jpeg will discard potentially important information that cannot be recovered and is
strongly discouraged.
This seventh edition brings many more references, plus new examples and images
throughout. The characterization of shape and the statistical analysis of data are covered
in greater detail, and examples of forensic applications of imaging have been expanded.
A major expansion is in the area of 3D imaging. The availability of instruments such as
synchrotron and microfocus X-ray tomography, and the ongoing development of com-
puter software that can present visualizations as well as make measurements with 3D
voxel arrays offer exciting possibilities for many research disciplines as well as routine
industrial and medical applications, and I try to present a current overview of these new
capabilities and their uses.
A great many scientists, worldwide and representing a broad range of disciplines, are
represented by example images from their work, and many have provided assistance,
raw data, references, and other information. They are acknowledged throughout, and
their help is gratefully appreciated. Many thanks to you all. Particular thanks are due
to Dr. Brian Metscher, at the University of Vienna, who performed microtomography on
specimens; to Dr. Loes Brabant, at Inside Matters, who provided data and the Octopus
software (based on the Morpho+ programs developed at the University of Gent) for pro-
cessing and measurement of 3D data sets; to Dr. Andreas Wiegmann, at Math2Market
GmbH, who provided valuable references, examples, and reconstructions; to Patrick
Barthelemy, at FEI Visualization Science, who loaned their Amira 3D visualization soft-
ware; and to Dr. Paul Shearing, at the Department of Chemical Engineering, University
College, London, who provided access to data from ongoing research projects. Access to
advanced 2D algorithms was provided by Andy Thé at The MathWorks. Christian Russ
at Ocean Systems provided valuable input on video technology and image compression.

A personal note
A brief personal and historical comment seems appropriate for this seventh edition.
Nearly 60 years ago I undertook my first serious foray into image processing—and it
was analog, not digital. As a student project, I tried to design an optical solution to a
problem that seemed to be beyond the capability of then-available (at least to me) com-
puter solution. The problem was to obtain dimensionally corrected images of the land
surface using aerial photography for surveying (with satellite imagery unimaginably far
in the future, the intended vehicle was a small plane flying at 12,000 feet, which tied in
nicely with my interest in flying). My solution was a continuously moving strip recorder
using photographic film and a wide angle slit lens system that “morphed” the image
projected onto the film one line at a time so that the geometric distortion was can-
celed. It worked on paper. In practice, the physical optics were impractical to fabricate
and specific to a fixed elevation and speed, and many difficulties—such as maintaining

xvi Introduction
© 2016 by Taylor & Francis Group, LLC
the stability and knowing the exact position of the airplane—would have frustrated its
actual use. The project succeeded in achieving its immediate purpose of fulfilling the
course requirements.
Since then I have been continuously involved in many ways with imaging throughout
my career, both in industry and academia. I’ve used and in some cases helped design
light and electron microscopes, surface metrology instruments, X-ray, gamma ray and
neutron tomography; have mounted digital cameras onto telescopes as a hobby; and
have programmed computers (ranging from a Cray to the early Apple IIs) to implement
most of the known, and some novel, processing and measurement algorithms. My name
is on several patents, and I’ve been involved in forensic cases and court trials in the
United States, Canada, and England. Most satisfying to me personally is that I’ve taught
image processing and measurement to several thousands of students, in settings rang-
ing from formal semester-long university courses, to 3- or 4-day intensive workshops
under the auspices of professional societies for their members and corporations for their
own employees.
As I approach the start of my ninth decade, and marvel at the progress in the capabili-
ties to extract useful information from images, I am especially pleased to work with the
next generation of researchers who will carry on this work. I have previously published
a book (Introduction to Image Processing and Analysis) and several papers with my
son, Christian Russ, who is involved in software developments for imaging. My present
coauthor, Brent Neal, was once a student who worked with me at North Carolina State
University and is now an accomplished researcher and the leader of the core instrumen-
tation facility at Milliken Research, where his expertise in polymer characterization and
measurement science uses a broad spectrum of 2D and 3D imaging technologies. I’ve
worked with Brent on a previous book (Measuring Shape), as well as several technical
publications, and a set of materials science teaching aids (“Visualizations in Materials
Science,” sponsored by the National Science Foundation), and have enormous respect
for his knowledge, intelligence, wide-ranging curiosity, and sense of humor. Putting
this volume together with him has been fun, and it is a much stronger text because of
his involvement.
The encouragement, support, and patience of our “girls”—Helen, Sarah, Meg, and Claire—
has also been a vital and most appreciated assistance for this effort.

Introduction xvii
© 2016 by Taylor & Francis Group, LLC
 1
Acquiring Images

Human reliance on images

Humans 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 3). For example, we expect to recognize familiar people and sur-
roundings in photographs, and television and movies are major sources of entertainment.
Tests with college students indicate that they are predominantly visual learners (in spite of
which, lectures are still the most common teaching mode). In a more specific instance, crime
scene photographs are often critical in explaining physical evidence to a jury.
Our expectations in everyday life extend to how we pursue more technical goals as well.
Scientific instruments commonly produce images to communicate their results to the user,
rather than generating audible tones or emitting 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 their range of
natural vision. In order to best understand the methods used in image processing and analy-
sis, it is helpful to begin with the types of images that are used and the devices that are used
to obtain them.
Optical devices such as microscopes and telescopes allow us to see things that are vastly smaller,
larger, or farther away than we could otherwise. Beyond the visible portion of the electro-
magnetic 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.
Figure 1.1 and Figure 1.2 show examples. One combines visible light with infrared and X-ray
data, obtained with different orbiting telescopes, and the second is an image presenting radio
telescope data in which color represents the Doppler shift in the radio signal. Such devices
and presentations are used to further extend imaging capability. Closer to home, ultraviolet
light is used in forensic investigations to cause latent fingerprints (Figure 1.3) and various
body fluids to fluoresce, enabling their detection at crime scenes. Infrared light also has been
used to reveal an identifying tattoo hidden beneath a more recent, elaborate tattoo (Asheville,
NC, Citizen-Times, August 7, 2012).
Signals other than electromagnetic radiation can be used to produce images, too. New types
of microscopes that use atomic-scale probes to “feel” the specimen surface also present their

1
© 2016 by Taylor & Francis Group, LLC
Figure 1.1 The pinwheel galaxy (M101) imaged in visible (green, Hubble image), infrared (red, Spitzer
image), and X-ray (blue, Chandra image) wavelengths. (Image courtesy of NASA and ESA.)

Figure 1.2 Combining visible light and radio astronomy produces images such as this view of HH 46/47
(generated with data from the Atacama Large Millimeter Array). These are often displayed with false
colors to emphasize subtle variations in signal strength or, as in this example, Doppler shift. The red
and orange colors identify the jet moving away from us, and the blue and magenta show the jet mov-
ing toward us. (Courtesy of European Space Agency.)

data as images (Figure 1.4). The data so collected may represent the surface elevation and
topography, but other signals, such as surface compliance or drag force on the probe, may
also be used. Acoustic waves at low frequency produce sonar images, while at gigahertz fre-
quencies 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 1.5 shows an acoustic microscope
image of a subsurface defect, and Figure 1.6 shows a sonogram of a baby in the womb.

2 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
 Figure 1.3 Latent fingerprint fluorescing
under ultraviolet light.

Figure 1.4 Atomic force microscope image

of human chromosomes. (Data cour-
tesy of S. Thalhammer and F. Jamitzky,
Helmholtz Zentrum, München, Germany.)

Figure 1.5 Microscope image of voids in solder bond beneath a GaAs die: (a) die surface (visible light);
(b) acoustic image showing strong signal reflections (white areas) from the surface of the voids.
(Courtesy of J. E. Semmens, Sonoscan Inc., Elk Grove Village, Illinois).

Acquiring Images 3
© 2016 by Taylor & Francis Group, LLC
Figure 1.6 Surface reconstruction sonogram,
showing a 26-week-old fetus in utero.

Extracting information
Some images such as holograms or electron diffraction patterns record brightness as a func-
tion of position or angle, but are unfamiliar to most observers. Figure 1.7 shows an image
of an electron diffraction pattern from a transmission electron microscope, in which the
atomic structure of the specimen is revealed (but only by measurement and to those who
know how to interpret the data). Other kinds of data including weather maps with special-
ized symbols, graphs of business profit and expenses, and charts with axes representing time,
family income, cholesterol level, or even more obscure parameters have become part of daily
life as illustrated in Figure 1.8. The latest developments in computer interfaces and displays
make extensive use of graphics to take advantage of the large bandwidth of the human

Figure 1.7 A convergent beam electron diffraction (CBED) pattern from an oxide microcrystal, which
can be indexed and measured to determine the atomic unit cell dimensions.

4 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
 Figure 1.8 Typical graphics used to communicate news
information include one-dimensional plots such as
stock market reports and two-dimensional presenta-
tions such as weather maps.

visual pathway. Tufte (1990, 1997, 2001) in particular demonstrates the power of well-designed
graphics to communicate complex information.
There are some important differences between the kind of information that human vision
extracts from images, and the ways in which it seems to do so, as compared to methods based
on computers that are applied for scientific, technical, or forensic purposes. Humans are espe-
cially poor at judging color or brightness of objects and features within images unless they can
be compared by making them adjacent. Human vision is inherently comparative 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 measuring scale. Chapter 3
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 human vision, but the overview in Chapter 3 can help the
reader to better understand how humans see things so that we become more skilled observ-
ers. Computer-based image processing and analysis use algorithms based on our understand-
ing of human vision methods in some cases, but also employ other methods that seem not to
have direct counterparts in human vision. Some image processing methods are based directly
on the physics of the image formation and detection process (Sharma, 2005).
Several types of imaging applications offer the most straightforward types of images for
analysis. These include microscopy, aerial and satellite imagery, and industrial quality control.
That is because in those situations there is the greatest knowledge of and 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 determining size, may have dif-
ferent lighting, which alters their color, and may partially be obscured by other objects. Crime
scene and accident photographs are often taken under difficult conditions, from less than
optimum points of view, and with variable lighting, so that their analysis can be challenging.
Surveillance cameras are becoming ubiquitous, but their placement, lighting variations, and
the variable location and orientation of subjects, complicate identification efforts. Examples
from a broad variety of image types, sources, and applications are used in the following chap-
ters to illustrate methods and algorithms.

Acquiring Images 5
© 2016 by Taylor & Francis Group, LLC
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 extract-
ing the details, and performing measurements utilize many of the same algorithms and tech-
niques, as set out in the following chapters. The interpretation of the measurements, as
presented in later chapters, does require some specialization for different viewing geometries
but is fundamentally independent of magnification.

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 copy stand, 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 simplest type of video camera is the vidicon, illustrated in Figure 1.9.
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 that the current flowing 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 1.10, conforms to standards of voltage and timing
(slightly different in Europe than in the United States, 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 8 guesses to arrive at a 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 (for example) 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

Figure 1.9 Functional diagram of a vidicon tube. Light striking the phosphor coating changes its local
resistance and hence the current that flows to the anode as the electron beam scans in a raster pattern.

6 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
Figure 1.10 Standard RS-170 video signal shows the brightness variation along one scan line (ranging
between 0 volts = black and 0.7 volts = white).

Figure 1.11 Examples of (a) pincushion and (b) barrel distortion, as well as vignetting at the edges
and corners.

distortion, or more complicated situations) and is subject to degradation by stray fields from
wiring or instrumentation. Figure 1.11 shows examples of pincushion and barrel distortion,
as well as vignetting and loss of focus. Maintaining the focus of the electron beam in the cor-
ners of the image requires 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
lights produce spots that spread laterally in the coating and appear larger than the features
represented, with “comet tails” in the scan direction.
There are, however, some advantages for the tube-type camera. The spatial resolution is 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,

Acquiring Images 7
© 2016 by Taylor & Francis Group, LLC
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 intensifica-
tion are capable of acquiring images in very dim light (e.g., astronomy, night surveillance,
fluorescence microscopy).

CCD cameras
The tube-type camera has now been largely supplanted by the solid-state chip camera, the
classic form of which is the CCD (charge-coupled device). The camera chip contains an array
of diodes that function as light buckets. Light photons entering the semiconductor raise elec-
trons from the valence to the conduction band, so the number of electrons is a linear measure
of the light intensity. The diodes are formed by photolithography, so they have a perfectly
regular pattern with no image distortion or sensitivity to the presence of stray fields. The
devices are also inexpensive, compact, and rugged compared to tube cameras. CCDs were first
invented and patented at Bell Labs in 1969 (George Smith and Willard Boyle were awarded the
2009 Nobel Prize in Physics for this invention), and have now almost entirely displaced film
in consumer and professional still and movie cameras.
The basic operation of a CCD is illustrated in Figure 1.12. Each bucket represents one “pixel”
in the camera (this word has different meanings in different contexts, as explained later, so
it must be used with some care). With 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 is measured to produce the numeric output from a digital camera.
Shifting the electrons is illustrated in Figure 1.13. Every set of three electrodes 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. This does not reduce the area sensitive to incoming photons,
because electrons generated in the high field regions migrate to the low field bucket where

Figure 1.12 The principle of CCD operation, illustrated as a set of buckets and conveyors. (After J. R.
Janesick, 2001, Scientific Charge-Coupled Devices, SPIE Press, Bellingham, WA.)

8 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
Figure 1.13 Varying the voltages on a set of three electrodes shifts electrons from one pixel to another
in a CCD.

they are held. By changing the voltage applied to the regions in phases, as shown in the fig-
ure, 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 dimension. Repeating the process acts like a conveyor
belt and is the reason for the name “charge-coupled device.”
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, infra-
red light penetrates easily and these cameras have red and especially 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 would
produce blurred or fogged images). Figure 1.14 shows this spectral response, which can be
further tailored and extended by using materials other than silicon. Applying an organic coat-
ing to convert ultraviolet (UV) radiation to visible wavelengths extends the spectral response
to shorter wavelengths. The chip can reach a high total efficiency when antireflective coatings

Figure 1.14 Spectral response: (a) silicon-based chip; (b) color sensors in the human eye, which are
commonly identified as red, green, and blue sensitive but cover a range of wavelengths.

Acquiring Images 9
© 2016 by Taylor & Francis Group, LLC
are applied, limited primarily by the “fill factor,” the area fraction of the chip that contains
active devices between the narrow ditches that maintain electrical 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 photo­
graphic film, which are all approximately logarithmic.
Human vision typically notices brightness differences of a few percent, that is, 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 1.15. 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 maxi-
mum optical density, based on the maximum physical density of silver particles or dye mol-
ecules. 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 difference 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. Some consumer
cameras and most professional cameras include the possibility to read the “raw” linear data
as well as the converted image.
When film is exposed directly to electrons, as in the transmission electron microscope 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 devel-
opment, but only a single electron. 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); this 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 have large sensors (6 cm wide in the Hubble

Figure 1.15 Response of photographic film. The central portion of the curve shows a linear increase in
density (defined as the base-ten logarithm of the fraction of incident light that is transmitted) with the
logarithm of exposure. High (“hard”) contrast corresponds to a steep curve, and low (“soft”) contrast
gives a less steep curve and a greater dynamic range.

10 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
telescope, for example). But for consumer devices, making each chip one-third, one-quarter,
or even two-tenths of an inch in 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 later) makes the
individual detectors small, but the gaps between must remain broad enough to prevent elec-
trons from diffusing laterally. The result is a reduction in total efficiency. Some devices place
small lenses over the diodes to capture light that would otherwise fall into the ditches (Holst &
Clerico, 2012), but these add cost and also are not so uniform as the diodes themselves (which
are typically within 1% across the entire chip).
The most important effect of making the detectors small is to reduce their capacity for elec-
trons, called the well capacity. A 15 µm pixel in the Hubble CCD has a capacity of about
500,000 electrons, which with low readout noise (as can be achieved by slower electron trans-
fers, for example) of a few electrons gives a dynamic range much greater than photographic
film. Even larger well capacity and dynamic range can be achieved by combining (binning)
more detectors for each stored pixel and 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.
The size of the detectors in many consumer pocket cameras and cell phones, typically about
1 µm, is close to the theoretical limits set by the number of photons collected and electrons
generated, and the resulting statistical noise (Schöbert et al., 2012).
Increased noise, for instance, due to reading out the signal at video rates (each horizontal line in
52 µs for U.S. standard definition video), dramatically reduces the dynamic range so that a typi-
cal consumer grade video camera has no more than about 50 distinguishable brightness levels
(expensive studio cameras meet the broadcast video specification of 100 levels). Since chip cam-
eras are linear with brightness, they produce fewer viewable gray levels as shown in Figure 1.16.
This performance is inferior to film, which can distinguish thousands of brightness levels.

Figure 1.16 Comparison of gray level steps from linear (equal increments) and logarithmic (equal
ratios) detectors.

Acquiring Images 11
© 2016 by Taylor & Francis Group, LLC
CMOS detectors
CMOS (complementary metal-oxide semiconductors) 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 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 several transistors immediately adjacent
to each 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 1.17). 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 may be compensated by placing lenses over each detec-
tor 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 make the sensors in a CMOS detector less uniform than
those in the CCD array, producing a fixed pattern that can be compensated for in software
(requiring recording a reference image with uniform illumination). Some newer chip designs
now “stack” the control circuits and amplifier underneath the detector.
In addition to the fixed pattern noise, CMOS detectors usually have a greater amount of ran-
dom noise 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, especially in the smaller chips used in pocket cameras and phones as the
pixel counts have risen to more than 10 million, have small well capacities, resulting in limited
dynamic range for the images. The images are usually stored with 8 bits per channel because

Figure 1.17 Schematic diagram of a typical CMOS detector. Each active light sensor (green) has addi-
tional transistors that are connected to addressing and output lines.

12 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
of the way memory is traditionally organized, but may not have that much actual brightness
resolution. Lossy compression further reduces the image quality as described in Chapter 2.
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 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 count. This approach, com-
bined with the use of larger sensors to achieve greater sensitivity and dynamic range, has led
some manufacturers to use CMOS detectors as large as a traditional film negative in digital
single-lens reflex cameras.
The advantages of CMOS sensors lie primarily in three areas: they consume 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 similar to 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 customized fabrication, for greater uniformity,
and fewer noise-producing imperfections than can be tolerated in other devices. Although
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 over-
all trend has been for CMOS sensors to continue to improve in quality and performance. The
advantages of the CCD sensor are still important for most technical applications, and the trade-
offs can be considered on a case-by-case basis (Nakamura, 2006; Holst & Lomheim, 2007).

Camera artifacts and limitations

The technology and function of the CCD and CMOS chip cameras described earlier are funda-
mentally the same, whether the camera is a still camera or a video camera (and many current
models can function in either mode). The differences are primarily concerned with the way
the image data is handled after the detector. Still cameras store the image in local memory,
whereas video cameras are—at least in principle—intended for live transmission or recording
of the sequence of pictures.
There are several problems with video that contribute to the specific types of defects present
in the images that must be dealt with by subsequent processing. One is the fact that video
signals may be interlaced (Figure 1.18). In some HDTV modes and with digital still cameras
the image is scanned progressively, line by line. Interlacing is a clever trick to minimize visual
flicker in broadcast television images, accomplished with tube cameras by scanning the elec-
tron beam at twice the frame rate (60 times per second in North America, 50 in Europe) in
the same interlace pattern as the display television set. With a CCD chip, it requires that the
array be read out twice for every 1/30th of a second frame, once to collect the odd numbered
lines and again for the even numbered lines. Many cameras 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, and so on in the other. This reduces vertical resolution, but for casual
viewing purposes it is not noticeable.
Motion can cause the even and odd fields of a full frame to be offset from each other, pro-
ducing a significant degradation of the image as shown in the Figure 1.18. A similar effect
occurs with stationary images if the horizontal retrace signal is imprecise or difficult for the

Acquiring Images 13
© 2016 by Taylor & Francis Group, LLC
Figure 1.18 Interlace scan covers odd-numbered lines in 1/60th-second field, and even-numbered lines
in a second field. When motion is present (either in the scene or caused by camera movement), this
produces an offset in the image.

Figure 1.19 Distortion of the rotating propellers produced by progressive scan makes the blades
appear curved.

electronics to lock onto; this is a particular problem with signals played back from consumer
videotape recorders. Moving images can also be distorted with progressive scan cameras due
to the time required to read from the top of the image to the bottom (Figure 1.19). Brightness
variations from fluorescent light sources may also be present. Practically all scientific imaging
systems, such as digital cameras, direct-scan microscopes (the scanning electron microscope
or SEM, the atomic force microscope or AFM, etc.), flatbed scanners, film or slide digitizers,
and similar devices, use progressive scan.
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 background
signal that further degrades dynamic range and may produce blurring. Electronic shuttering is

14 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
usually done one line at a time, which distorts moving images. 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,
halving sensitivity.
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, 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 bottom, and can be visually detected in images
if there is not a lot of other detail or motion to obscure it.
The transfers of electrons from one detector to another occurring during readout of a chip
account for some of the noise in the signal. Inherent statistical variations in the production
and collection of charge are a smaller effect. The conversion of the tiny charge to a voltage and
its subsequent amplification are the greatest sources of noise in most systems. Readout and
amplifier noise can be reduced by slowing the transfer process so that fewer electrons are lost
in the shifting process and the amplifier time constant can integrate out more of the random
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 nonvideo applications, of course. Digital still cameras use the same chip technology
(but much higher numbers of detectors) as solid-state video cameras, and produce higher
quality images because of slower readout. Janesick (2001) discusses the various sources of
noise and their control in scientific CCDs of the type used in space probes and astronomical
imaging, where they have entirely replaced film. Holst (1998) has comprehensive details on
the design and characteristics of solid-state detectors and camera systems.

Color cameras
Color cameras are designed in four principal ways, shown in Figure 1.20. For stationary
scenes (which includes some scientific applications, 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 (Figure 1.20a), which are then combined for viewing. The advan-
tages of this scheme include lower cost and the ability to use different exposure times for the
different color filters, which can compensate for the poorer sensitivity of the silicon chip to
short wavelength (blue) light.
Some high-end consumer and most professional video cameras use three sensors (Figure 1.20b).
A prism and filter array split the incoming light into red, green, and blue components, which
are recorded by three different sensors whose outputs are electronically combined to pro-
duce 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-definition
video camera has many fewer pixels than a typical digital still camera). The optics and hard-
ware to keep everything in alignment add some cost, and the depth of the prism optics makes
it impractical to use short focal length (wide angle) lenses. The sensitivity for low light levels
is also reduced. This design is rarely used in digital still cameras.

Acquiring Images 15
© 2016 by Taylor & Francis Group, LLC
Figure 1.20 Schematic diagrams of color cameras: (a) color wheel camera with red, green, and blue
filters (the fourth filter position is empty, allowing the camera to be used as a monochrome detector
with greater sensitivity for dim images); (b) prisms and dichroic filters for a three-chip color camera;
(c) stacking the blue, green, and red sensitive detectors on top of one another; (d) Bayer and Sony
RGBE patterns used in single-chip cameras (the “E” or “emerald” filter is cyan).

Standard video images are often digitized into a 640 × 480 array of stored pixels (the dimen-
sions of the VGA [video graphics array] display that was once standard for personal c­ omputers),
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 direc-
tion for the brightness (luminance) signal, and about half that for the color information, as
explained later.
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. (A very common
problem with forensic examination of surveillance videotapes is that they 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). Fortunately, most video
recording now uses digital storage.
Video images are not very high resolution, although HDTV (high-definition television) has
improved things somewhat. The highest current standard resolution for HDTV is 1080p,
meaning 1080 horizontal lines with progressive scanning, and 1920 samples per line (a total
of just over 2 million points, whereas many current digital still cameras store 10 times as
many pixels). Interest is slowly building for “4K” video, with 3840 × 2160 display pixels. But
the actual image resolution is not that high because the images are compressed for transmis-
sion, and much of the information (especially the color values) is discarded. Video technology
is usually a poor choice for scientific imaging unless there is some special need to capture
real-time images (i.e., 25 to 30 frames per second) to record changes or motion. Digital still
cameras produce much higher resolution images with greater dynamic range, and many can
also record video-rate sequences when that is needed.
Another approach to color camera design, developed by Foveon Corp. and used in a few
cameras, employs three transistors at each pixel location, stacked on top of each other using
CMOS technology (Figure 1.20c). 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 informa-
tion. This approach does not suffer from loss of spatial resolution due to interpolation but has
problems with consistent or accurate color fidelity.

16 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
 Figure 1.21 Fragment of an image showing a “zipper”
pattern from a single-chip digital camera.

Most digital cameras use a single pixel array, often of very high pixel (detector) count, with
a color filter array (CFA) that allows red, green, and blue light to reach specific detectors.
Different patterns may be used (Figure 1.20d), 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 mimics to some extent the human eye’s greater sensitivity to green. Some
cameras have a stripe pattern instead, and a variety of other patterns are also used including
detectors with different filter colors, intended to obtain better color discrimination particu-
larly in the green range of wavelengths. The original Bayer patent describes the use of cyan,
magenta, and yellow filters. The problem with the single-chip camera is that the image resolu-
tion in each color channel is reduced. With the Bayer pattern, the red intensity at three-fourths
of the 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 herringbone pattern (often referred to as a “zipper”) and color fringes appear along
contrast edges in the picture, as shown in Figure 1.21.
Interpolation reduces the image resolution compared to the number of individual detectors in
the camera (which is generally the specification advertised by the manufacturer). Inherently,
this “demosaicing” (or “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 requirements. A comparison of several meth-
ods can be found in (Kimmel, 1999; Ramanath, 2000; Ramanath et al., 2002; Shao et al., 2005;
Miao & Qi, 2006; K.-L. Chung et al., 2010; Guarnera et al., 2010; Huang et al., 2010; D. Zhou
et al., 2014). 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 forensic cases to
identify the model of camera used to photograph a scene, even to identify specific fixed pat-
tern noise from an individual camera or to detect alterations made to the image later (Bayram
et al., 2006; Swaminathan et al., 2007; McKay et al., 2008; Farid, 2008, 2009).
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, especially when CMOS detectors are used. In a three-chip system these
produce different variations in the red, green, and blue output that increases the color noise
in the images.
A practical caution in the use of all these types of cameras is the need to avoid automatic
gain or brightness compensation circuits. These can change the image contrast or linearity
in response to bright or dark regions that may not even lie within the recorded portion of

Acquiring Images 17
© 2016 by Taylor & Francis Group, LLC
the image, and increase the gain and noise for a dim signal, making comparison between
images difficult.

Camera resolution
The electrical signal coming from the silicon detector is analog, regardless of whether digitiza-
tion takes place within the camera body or even on the same chip. In most cases, the actual
image resolution with a single-­chip camera and filter arrangement is one-­half to two-­thirds
the value that might be expected from the advertised number of pixels (detectors) in the cam-
era, because of interpolation and amplifier characteristics. Some cameras record images with
many more stored pixels than the chip resolution warrants. Such interpolation and empty
magnification contributes no additional information in the image.
Comparing cameras based on actual resolution rather than the stated number of recorded pix-
els 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, or for reducing the effects of camera shake
during long exposures). 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 reduced 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 number of discrete points
across the image that can be distinguished from each other. It is often specified in terms of
the number of line pairs that can be resolved, which 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 (more are needed if the lines do not lie exactly on the detectors). Resolution
also depends on the amount of brightness contrast between the lines and the spaces, and the
amount of noise (random variation) present in the image. Lines are much easier to resolve
than individual points, since averaging along each line improves the signal-­to-­noise ratio.
The situation is even more complicated with some digital still cameras that shift the detector
array to capture multiple samples of the image. The usual 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 resolution 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 resolution approaching that corresponding to
the detector spacing. Some studio cameras displace the entire sensor array to different regions
of the film plane to collect tiles that are subsequently 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, and these designs are not common.
Rather than a two-­dimensional array of detectors, it is also possible to use a linear array
(­usually 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 desktop scanners, which for many applications
are excellent image acquisition devices. It has also been used in studio cameras mounted on a
tripod, and some light microscopes accomplish the same thing by moving the stage and speci-
men 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

18 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
access the stored array (Bacus & Bacus, 2000, 2002) and to interactively deliver a selected por-
tion 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 amplification. In
some cases, processing software that reduces image noise using averaging or median filters
may be applied separately to each color channel, using different parameters according to the
actual or expected noise content. Chapter 4 shows better approaches.
Digital cameras using the same chip technology as that in a video camera can produce much
better image quality. This is due in part to the longer exposure times, which collect more elec-
trons and so reduce noise due to statistics and amplification. Furthermore, the slower read-
out of the data from the chip, which may take up to a second instead of 1/60th of a second,
reduces readout noise. Digital still cameras read out the data in one single pass (progressive
scan). By cooling the chip and amplifier circuitry to reduce dark currents, integration (long
exposures up to tens of seconds or for some astronomical applications many minutes) can be
used to an advantage because of the high dynamic range (large well size and large number of
bits in the digitizer) of some chip designs. Most uncooled camera chips begin to show unac-
ceptable pixel noise due to dark current with integration times of more than a few seconds.
Also, 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 more than the
resolution of the video signal, which as noted earlier is rather poor. In a digital still camera,
very high pixel counts can give rise to extremely high resolution. Digital camera backs with
60 million to 80 million detectors are commercially available for professional photographers.
In addition to dedicated cameras for attachment to microscopes, telescopes, or other special-
ized optics, single-­lens reflex cameras are available with high resolution (more than 20 million
pixels at this writing) and large area detectors that can be used in technical applications, with
simple optical attachments that make it easy to connect them to instruments (and of course
the camera may also be removed and used for other purposes). The camera may be tethered
directly to a computer for control and storage, but in many cases it is more practical to record
the images to memory cards whose contents are later downloaded. Professional digital cam-
eras with large, high-­resolution detector arrays, interchangeable lenses, and so forth provide
capabilities comparable to traditional 35 mm and larger film cameras. Manufacturers of cam-
eras have converted from film to digital recording, with a wide variety of designs available and
new developments 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 a bench microscope, with a 10× (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 the field of view becomes 160 µm. For a camera
with 4800 × 3200 sensors (15 megapixels) the low magnification image is recorded at about
3 pixels per µm, more than adequate for the resolution of the optics (about 2 µm). The high
magnification image is recorded with 300 pixels per µm. Since the optical resolution of the
microscope under optimum conditions is about 0.5 µm with the 100× lens, this produces a vast
and unnecessary 150× oversampling. At low magnification, or for viewing fine detail in large
scenes (such as aerial and satellite imagery), a high pixel count makes sense. When the limita-
tion on resolution lies with the optics, it may not make sense (Swift & Trinkle-­Mulcahy, 2012).

Acquiring Images 19
© 2016 by Taylor & Francis Group, LLC
Electronics and bandwidth limitations
Video cameras (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
brightness, as shown earlier in Figure 1.10. The scan is nominally 525 lines per full frame,
composed of 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 earlier, and
this value is reduced if the signal is degraded by the electronics. Even the best system can be
degraded in performance by such simple things as cables, connectors, or incorrect termina-
tion impedance.
HDTV modes include many differences from conventional television. The pixel density is
much higher, with a wider aspect ratio of 16:9 (instead of the 4:3 used in an NTSC [National
Television Standards Committee] television) with square pixels. A high-­quality HDTV mode
presents 1920 × 1080 pixel images at the rate of 60 scans per second, for a total data rate
exceeding 2 gigabits per second, several hundred times as much data as analog broadcast
television. To transmit this data, either over the air or via cable, compression techniques are
required (discussed in Chapter 2).
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 using successive approximation (described earlier) 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 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 the next reading made, so that a series of bright-
ness values is obtained along each scan line. Figure 1.22 illustrates the digitization of a signal

Figure 1.22 Digitization of an analog voltage signal

along one line in an image (blue) produces a series
of values in steps (red) that are equal in time and
rounded to integral multiples of the smallest measur-
able increment.

20 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
into equal steps in time and in value. Digitizing several thousand points along each scan line,
repeating the process for each line, and transmitting or storing the values in memory produces
a digitized image for further processing (including ­compression) or analysis.

Handling color data

Most real-­world images are color rather than monochrome. The most common type of color
image is that produced by recording the intensity in three different wavelength bands of vis-
ible light. It is instructive to begin by considering video, since standard definition broadcast
television uses color in a way that has been accepted as visually satisfactory. The NTSC color-­
encoding scheme used in the United States was developed in the early 1950s as a compatible
add-­on to existing monochrome television broadcasts. It adds the color information within the
same, already narrow bandwidth limitation. The result is that the color has even less lateral
resolution than the brightness information.
Color subsampling reduces the amount of data in an image by representing color data with
lower resolution than luminance (brightness) data. This is done in color spaces such as YCbCr,
explained next (often called YUV, which is an analog rather than digital color representation).
Y is the brightness or luminance component of the signal, and the other two axes give the
color or chrominance data. Uncompressed color is represented as 4:4:4. The numbers describe
the number (and spacing) of the sampling of the brightness and color values. The common
subsampling options are 4:2:2, 4:2:0, 4:1:1, and YUV-9.

4:2:2. Full sampling of Y and 2:1 horizontal sampling of U and V. This is the analog
sampling scheme most commonly used in professional and broadcast video and in
tape formats such as D-1 and Digital Betacam. It looks good, but the data compression
ratio is only 33%.
4:2:0. Full sampling of Y and 2:1 sampling of the two color components in both the hori-
zontal and vertical dimensions. That is, for every four luminance samples, there are
two chrominance samples every other line. This yields a 50% reduction in data. 4:2:0
is the color space used in MPEG compression, used in most DVD recordings.
4:1:1. Full sampling of Y and 4:1 horizontal sampling of color components. This is the
color space of the digital video (DV) formats used in many digital camcorders. It uses
color samples four pixels wide and one pixel tall, so color bleeding is much worse in
the horizontal than in the vertical direction.
YUV-9. This is the color format used in most of the video compression on the internet.
For every 16 luminance Y samples in a 4 × 4 pixel block, there is only one U and one
V sample, producing smaller files with correspondingly lower color fidelity. YUV-9
subsampling often results in noticeable color artifacts around the edges of brightly
colored objects, especially red.

The point of this list is that all of the color representations, both analog and digital, discard
some, even most, of the color information, compromising subsequent efforts to use color in
analysis (for example, determining the boundaries of regions for measurement). These limita-
tions are judged to be acceptable for television pictures, since the viewer tolerates colors that
are less sharply bounded and judges the edges of features to be defined by the brightness
component of the image wherever the colors do not exactly correspond. The same tolerance
has been used effectively by artists and may be familiar to parents whose young children have
not yet learned to color inside the lines.

Acquiring Images 21
© 2016 by Taylor & Francis Group, LLC
The poor spatial sharpness of NTSC color is matched by its poor consistency in representing
the actual color values (a common joke is that NTSC means “Never The Same Color”). Video
recordings of color images are often less useful for analysis than monochrome ones.
Better fidelity in the image can be preserved by not mixing together the color and bright-
ness information. Instead of the composite signal carried on a single wire, some cameras and
recorders separate the chrominance (color) and luminance (brightness) signals onto separate
wires. This is the so-­called Y-­C or S-­video format used in some camcorders (Hi-8 and S-­V HS
formats). Some computer interfaces accept this format, which yields an improvement in the
quality of digitized images. Some color cameras intended for technical purposes bring out the
red, green, and blue signals separately so that they can be individually digitized. Recording
the image in computer memory then involves treating each signal as a monochrome one, con-
verting it to a set of numbers, and storing them in memory. If the signals have first been com-
bined, the encoding scheme used is likely to be YIQ (used in U.S. analog broadcast ­television)
or YUV (used in European PAL broadcast TV).
YIQ assigns 2.5 MHz (out of the total 4.2 MHz bandwidth) to the Y or brightness signal,
1.3 MHz for the I color information (approximately red to cyan), and just 0.4 MHz for the
Q signal (approximately magenta to green, which is thought to be less important for human
vision). YUV assigns equal bandwidth to the two color channels, which correspond to a
slightly rotated set of axes in color space.
Another important development in video is digital video (DV) recording. DV records up to
13.5 MHz for the luminance (brightness) information, and one-­fourth that for the chrominance
(color) information. This produces much sharper edge delineation for the brightness data in
the image and improves the usefulness of the resulting images. The digital video signal is
fully preserved, unlike conventional analog recorders that impose a further bandwidth reduc-
tion on the data, and are susceptible to quality loss due to dirty heads, worn tape, or making
successive copies from an original. Current model DV video cameras have replaced the tape
cartridge with solid-­state memory to store the data, but the same data format is used.
The result is that DV images have approximately 500 × 500 brightness resolution (but less
for the color information, as noted earlier for the 4:1:1 format) and nearly 8 bits of contrast,
and may be read into the computer without a separate digitizer board since they are already
digital in format. The IEEE 1394 standard protocol for digital video (also known as Firewire)
establishes a standard serial interface convention that is available on some consumer-­priced
cameras and decks, and is supported by some computer manufacturers. Most of these digi-
tal cameras can be controlled by the computer to select individual frames for transfer and can
be used to record single frames annotated with date and time.

Color encoding
Conversion from RGB (the brightness of the individual red, green, and blue signals, as cap-
tured by the camera and ultimately stored in the computer) to the color encoding schemes
used in television is based on color differences. Y, the “luminance” signal, is just the brightness
of a panchromatic monochrome image as displayed by a black-­and-­white television receiver.
It combines the red, green, and blue signals in approximate proportion to the human eye’s
sensitivity to them. The color components of the signal were chosen in the 1960s for compat-
ibility with existing black-­and-­white sets, which simply ignored the color information. The
YUV or YIQ components are obtained using analog circuits, while the YCbCr transformation
(sometimes abbreviated as YCC) is used for digital transmission.

22 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
 Figure 1.23 Gamma adjustment, showing the
curves used to encode and decode bright-
ness values.

Transforming RGB values to the YCbCr system begins by applying a gamma adjustment to the
individual RGB values (Balasubramanian et al., 1994). This is a nonlinear operation defined
by the power-­law expression:
γ
Vadjusted = Vinitial (1.1)

where the V values range from 0 (black) to 1 (maximum). There are two complementary
transformations involved, for encoding and decoding. For instance, in most computer displays
images are encoded and decoded with a gamma of 2.2, as shown in Figure 1.23.
The gamma-­adjusted RGB values (shown as R′G′B′) are then used to calculate Y, Cb, and Cr as:
Y ′ = K R i R ′ + K G ·G ′ + K B · B ′

B′ − Y ′
Cb = 0.5 (1.2)
1 − KB
R′ − Y ′
Cr = 0.5
1 − KR

where the K values vary with the type of monitor and with various television standards
(Table 1.1).
The YCbCr axes are shown schematically in Figure 1.24. Although the equations for the con-
version from RGB do not degrade the data, the use of limited precision (e.g., 8 bit integers)
to represent the values can result in some loss. Furthermore, it was shown earlier that color
information is only a coarse sample from the actual pixels, and that the color data from the

Table 1.1 K Values in Equation 1.2

KR KG KB
Broadcast TV 0.299 0.587 0.114
Computer monitor 0.2126 0.7152 0.0722

Acquiring Images 23
© 2016 by Taylor & Francis Group, LLC
Figure 1.24 The YCbCr space used to encode
image colors.

camera depend on the demosaicing and interpolation of data from the individual sensors. In
general, even before the MPEG compression described in Chapter 2, video imagery has many
problems affecting resolution and color. Nevertheless, there are times when these images must
be processed and analyzed as a source of information. This is a very brief overview of video
color technology, and numerous publications, especially ones published by IEEE and SPIE, go
much deeper into the various aspects.
Another approach to color compression may be encountered with digitized images. Reduction
of color values to 256 colors using a lookup table, specifying the “best” 256 colors to match
the contents of the image, may be used for computer images in order to reduce the file size.
The resulting images are described as “Indexed Color.” The lookup table itself requires only
3 × 256 bytes to specify the R, G, B values for the 256 colors, which are written to the display
hardware. Only a single byte per pixel is needed to select from this palette, reducing the file
size by two-­thirds. The most commonly used technique for selecting the optimum palette is
the median cut algorithm (Heckbert, 1982), which subdivides color space based on the RGB
values of the pixels in the original image. In some applications, a default color lookup table
is predefined (e.g., consisting of only the 216 standard colors used by internet browsers). For
visual purposes such a reduction may provide a displayed image that is satisfactory, but for
image analysis purposes this should be avoided.

Other image sources

Many types of scientific apparatus produce images or data that can be viewed as an image
without using a camera. Medical imaging with X-­ray computed tomography (CT) or magnetic
resonance imagers (MRI), for example, are discussed in Chapter 13. The scanning electron
microscope (SEM) or scanned probe microscopes such as the atomic force microscope (AFM)
typically form an image in a raster pattern with up to about 1000 scan lines. The signals are
digitized using 8 or 12 bits, and so are similar in image size and brightness resolution to many
camera systems. Figure 1.25 schematically shows the function of an SEM. The focused beam
of electrons is scanned over the sample surface in a raster pattern while various signals gen-
erated by the electrons are detected. These include secondary and backscattered electrons,
characteristic X-­rays, visible light, and electronic effects in the sample.

24 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
 Figure 1.25 The scanning electron microscope
(SEM) focuses a fine beam of electrons on
the specimen, producing various signals
that may be used for imaging as the beam is
scanned in a raster pattern.

Other point-­scanning microscopes, such as the AFM, the confocal scanning light microscope
(CSLM), and contact profilometers, produce very different signals and information. All provide
a time-­varying signal that is related to spatial locations on the sample by knowing the scan-
ning speed and parameters, which allows storing the data as an image. Many of these devices
have noise and resolution characteristics that are different in directions parallel and perpen-
dicular to the scanning direction.
Significantly larger arrays of pixels are available from flatbed scanners. These devices use a
linear solid-­state detector array and can typically scan areas at least 8 inches × 10 inches and
sometimes up to several times that size. Although primarily intended for the desktop publish-
ing market, they are readily adapted to scan many types of specimens that are flat or nearly
flat, as well as electrophoresis gels used for protein separation, or photographic prints and
negatives. Specialized scanners are used for digitizing photographic negatives such as medical
X-­rays and 35 mm slides. A high-­quality photographic negative can record several thousand
distinguishable brightness levels and several thousand points per inch (both values are much
better than prints). Commercial scanners are used in publishing and to convert films to digital
values for storage. At the consumer level, scanners with 2400 pixels per inch are common for
large-­area reflective originals, and up to 4000 pixels per inch for film. Most digitize full-­color
red, green, and blue (RGB) images.
Scanners are inexpensive, but rather slow, taking tens of seconds to digitize the scan area.
Another characteristic problem they exhibit is pattern noise in the sensors: if all of the detec-
tors along the row are not identical in performance a “striping” effect appears in the image
as the sensor is scanned across the picture. If the scan motion is not perfectly smooth, it can
produce striping in the other direction. One difficulty with such scanners arises from the illu-
mination. There may be a drop-­off in intensity near the edges of the scanned field because of
the short length of the light source. More troublesome is the warm-­up time for the light source
to become stable, if fluorescent illumination is used rather than LEDs.

Pixels
Using an array of discrete pixels to represent a continuous image is not something new that just
developed as a result of chip-­based cameras and digital computing. Figure 1.26a shows a frag-
ment of a 1st century Roman mosaic from El Djem, Tunisia. Individual pieces of colored glass,
mostly square in shape, are arranged so that at a normal viewing distance they are perceived

Acquiring Images 25
© 2016 by Taylor & Francis Group, LLC
Figure 1.26 (a) Mosaic images and (b) cross-stitch samplers are physical forms of pixelation.

as an image, including subtle shading. The viewer easily fills in the gaps and interprets this as
a continuous-­tone representation. The 19th century cross-­stitch sampler in Figure 1.26b has
a more regular grid, defined by the woven support cloth, with various colored threads used to
form the image pixels. With modern digital cameras, the individual pixels are generally smaller
and more numerous, and for convenience in manufacture and data processing are ideally
square and arranged in a regular Cartesian grid, but the principle remains the same.
It is desirable to have the spacing of the pixel values be the same in the horizontal and verti-
cal directions (i.e., square pixels), as this simplifies many processing and measurement opera-
tions. There are some theoretical advantages to having pixels arranged as a hexagonal grid
(Staunton, 1989; Bell et al., 1989; Barnard & Boreman, 1991; Middleton & Sivaswamy, 2005;
X. He & Ja, 2005; X. Li, 2013), and indeed nature uses that arrangement (Figure 1.27). But
because of the way that acquisition hardware actually functions, and to simplify the address-
ing of pixels in computer memory, this is rarely done.
Since pixels have a finite area, those that straddle a boundary in the scene average the bright-
ness levels of two regions and have an intermediate brightness that depends on how the

Figure 1.27 The hexagonal arrangement of cones in the (a) foveal area of the human eye and (b)
ommatidia in the eye of a fly.

26 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
Figure 1.28 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, and 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.

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 1.28 shows several examples of an image with varying numbers of pixels across its
width (spatial resolution), and Figure 1.29 shows the same image with varying numbers of
gray levels (tonal resolution). For a color image, reducing the number of brightness levels in
each RGB channel is shown in Figure 1.30.
For the most common types of image acquisition devices, such as digital cameras, the pixels
represent an averaging of the signal across a finite area of the scene or specimen. However,
there are some situations in which this is not so. At low magnification, for example, the scan-
ning 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 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 locations 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 be used to locate it to a fraction of the pixel spacing. These methods are discussed in
Chapter 10 on image measurements.

Acquiring Images 27
© 2016 by Taylor & Francis Group, LLC
Figure 1.29 Four representations of the same image, with variation in the number of gray levels used.
From the upper left: 32, 16, 8, and 4. In all cases, a full 256 × 256 array of pixels are retained. Each
step in the coarsening of the image is accomplished by rounding the brightness values.

Figure 1.30 Four representations of the same image with variation in the number of brightness levels
in each of the RGB channels: (a) 256; (b) 16; (c) 4; (d) 2.

28 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
Tonal resolution
In addition to defining the number of sampled points along each scan line, and hence the res-
olution of the image, the design of the ADC also controls the precision of each measurement.
High-­speed flash analog-­to-­digital converters may 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, if it does
not vary from full black to white. Also, the quality of most analog video cameras and their
associated electronics rarely produces voltages that are free enough from electronic noise
to justify full 8-bit digitization. A typical camera specification of 49 dB signal-­to-­noise ratio
means 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 computer memory into
bytes, so that one byte of storage can hold the brightness value from one pixel in the image.
Many digital still cameras and most scanners produce more than 256 distinguishable bright-
ness values, and for these it is common to store the data for R, G, and B each in 2 bytes or
16 bits, giving a possible range of 65536:1, which exceeds the capability of any current imag-
ing device (but not some other sources of data that may be displayed as images, such as sur-
face elevation, discussed in Chapter 15). For a camera with a 10 or 12 bit output, the values
are usually 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 sufficient, 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 can be
represented by fractional floating point values. That is the convention used in this book.
Not all programs use the 0..255 convention for representing brightness values. For example, the
0..1 range (using floating point values) has the advantage that the data are directly useable in
many arithmetic operations, and also normalizes the differences between 8, 12, 16 bit or other
digitization ranges. However, the values must still be scaled to 0.255 integers for display purposes.
When the stored image is displayed from computer memory, the numbers are used in a digital-­
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). A common technique in digital pro-
jectors uses a rapidly moving mirror for each pixel to vary the fraction of time that the light
is sent to the screen, producing a variation in perceived brightness. These processes are com-
paratively noise-­free and high resolution, since computer display technology has been devel-
oped 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 these displays,
nor can they be successfully recorded or printed using ink jet or laser printers (discussed in
Chapter 2). About 20 to 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 when the purpose of the imaging is to produce printed output.
A somewhat different 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 in interstellar space. If only 256 brightness

Acquiring Images 29
© 2016 by Taylor & Francis Group, LLC
levels are stretched to cover this entire range, there is not enough sensitivity to small varia-
tions to reveal detail in either the 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.
Medical imaging display systems usually are capable of a greater bit depth, such as 1024
shades of gray. This is based on an assessment that between 720 and 900 shades of gray
(Kimpe & Tuytschaever, 2007) are required to cover the full range from black to white with
fine enough steps corresponding to the just-­noticeable difference of the human visual system
when the eye is fully adapted to a luminance value of a uniform background. Furthermore,
the brightness steps on the display are linear but ability to distinguish contrast is worse for the
bright and dark ends of the luminance range (Peli et al., 1996; Barten, 1999). The use of 1024
display brightness levels ensures that the full range of the data is available to the viewer but
does not imply that they can all be differentiated in the display of an actual image.

The image contents

Figure 1.31 shows an image 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.
The histogram shown is well spread out over the available 256 brightness levels, with peaks
corresponding to each of the areas. If a bright light falls on a portion of the detector in the

Figure 1.31 A monochrome (grayscale)

image and its brightness histogram,
which plots the number of pixels with
each brightness value. The counts in
each peak correspond to the areas of
the regions of the image.

30 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
Figure 1.32 An image taken with filtered sunlight and an effective color temperature of 5000C, but
stored as a raw file and opened using different assumed color temperatures. From left to right:
3500C, 4500C, 5500C, and 6500C.

solid-­state camera that is 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 relation-
ship 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.
Color correction and calibration are dealt with in Chapter 4, but obtaining absolute color
information from video cameras is not possible because of the broad range of wavelengths
passed through each filter, variations in illumination color (for instance, even with slight volt-
age changes on an incandescent bulb, or outdoor lighting at different times of the day) and
the way the color information is encoded.
The color temperature of the illumination used is critical to matching colors in images.
Figure 1.32 shows an image recorded using filtered sunlight, with an effective color tempera-
ture (described more fully in Chapter 2) of approximately 5000C, 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.

Camera limitations
The development of the digital still-­frame cameras for the consumer market involves practi-
cally every traditional maker of film cameras and photographic supplies, as well as manu-
facturers of smart phones and others. Many of these cameras are unsuitable for technical
applications because of limitations in the optics (fixed focus lenses with geometric d
­ istortions)
and limited resolution. But even the camera chip in a cell phone may offer adequate spatial
and tonal resolution for some purposes, and has the added virtue of always being available,
which can have vital importance in some forensic situations.
It is the use of image compression that creates the most important problem. In an effort to
pack many images into the smallest possible memory or minimize file sizes to be sent via
the internet, JPEG and other forms of image compression are used. Chapter 2 explains that
these are lossy techniques, which discard information from the image. The discarded data is
selected to minimally impact human interpretation and recognition of familiar images (such as

Acquiring Images 31
© 2016 by Taylor & Francis Group, LLC
snapshots of the kids and the summer vacation), but the effect on quantitative image analysis
can be severe. Edges are broken up and shifted, color and density values are altered, and fine
details can be eliminated or moved. Many consumer cameras also apply in-­camera process-
ing to the image, for example, to visually sharpen edges using methods shown in Chapter 5,
and to convert the linear output from the detector to the logarithmic response provided
by film.
The important advantage of the higher-­end consumer and professional cameras is that they
offer storage or transfer to the computer without any lossy compression or other processing,
and this is much preferred for technical applications. There are a number of different “raw”
formats used by the various camera manufacturers. Many computer programs (e.g., Adobe
Photoshop®) offer routines to read the raw data from the camera and convert it with adjust-
ments for lighting, vignetting, and contrast.
Most methods for defining and isolating objects to be analyzed, described in later chapters,
depend upon the assumption that the same object will have the same appearance, wherever
it happens to lie in the image area. Basically, that means the brightness and color should not
depend on position. In real-­world photographs, that is not often the case. Achieving uniform
illumination of the area to be imaged is a difficult task. In satellite imagery, for example, the
curvature of the planet, combined with the position of the sun and any irregularities in
the local surface slope, or clouds, produce inherent variations.
In a microscope, proper alignment and adjustment of the lenses (Köhler illumination) pro-
duces parallel, uniform illumination at the specimen. On a copy stand, or in a studio setting,
the careful use of extended light sources such as diffusers and umbrella reflectors can provide
uniform illumination. But even if uniform lighting can be achieved, it does not ensure that the
resulting image will have uniform brightness from center to edge.
Most photographs exhibit a gradual drop-­off of brightness with distance from the center,
called vignetting, as illustrated in Figure 1.33. There are several reasons for this, and in many
cases such as portraiture it is considered desirable, as it focuses interest on the person or
object at the center. A principle cause of this darkening is the mechanical result of light enter-
ing a lens aperture from the periphery of the image; the circular aperture of the lens becomes
an ellipse with reduced area, admitting less light. Also, light entering at an angle passes
through greater thicknesses of glass and filters, reducing intensity. The light toward the edges
of the image that does reach the film or transistors in the sensor impinges at an angle, which

Figure 1.33 Vignetting and its removal: (a) original image of crumpled paper with a plot of the average
brightness versus radius; (b) after subtracting blurred copy, the average radial brightness is nearly
flat, and the contrast range of the details is increased.

32 The Image Processing Handbook, Seventh Edition

© 2016 by Taylor & Francis Group, LLC
reduces the effective area of the transistor and also reduces the depth of penetration and thus
the absorption of the photons and the generated signals. All of these effects are increased
by the use of wide-­angle lenses and large apertures.
Several methods for removing or compensating for the variation in brightness due to vignett-
ing are shown in Chapter 4. One of the simplest methods, although not often perfect, is to
heavily blur a copy of the image to eliminate the fine details, for example, by defocusing the
lens and recording another image. Then either divide or subtract the blurred copy from the
original (depending on whether the original image has been acquired by a device with a linear
response, like the raw output from CCD, or a logarithmic one, like film). Subtraction is used
in the example in Figure 1.33; the process also removes the color information.

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 introduced in the collection of electrons from 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 in the image that is due to differences present in the scene to the noise
level is the signal-­to-­noise ratio (S/N). When this is low, the features present may be invisible
to the observer. Figure 1.34 shows an example in which several features of different size
and shape are superimposed 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 or mini-
mum dimension.
Acquiring images at video rates of 25 to 30 frames per second is sometimes referred to as
“real-­time” imaging, but 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 image to be recorded at periods of many minutes,
hours, or even days. For others, very short exposures and high rates are needed. Special cam-
eras that do not use video frame rates or bandwidths can achieve rates up to 10 times that of
a standard video camera for full frames, and even higher for small image dimensions. Very
high speed cameras (up to 1011 images per second) 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 capture rate does not need to be high. Either stroboscopic imaging
or a fast shutter speed may be enough to stop the important motion to provide a sharp image.

Figure 1.34 Features on a noisy background: (a) signal to noise ratio 1:1; (b) signal to noise ratio 1:3;
(c) signal to noise ratio 1:7.

Acquiring Images 33
© 2016 by Taylor & Francis Group, LLC
Another random document with
no related content on Scribd:
 . . . . . . . . . . . . . . . . . . . .

Le parfum des gardénias entre dans ma chambre avec le soleil ; il

domine l’odeur des jasmins, des chèvrefeuilles et des roses sous
lesquels la maison est enfouie.
C’est une vieille maison, avec une véranda tout autour et un puits
à margelle de pierre dans le patio… Un oiseau-mouche fait vibrer ses
petites ailes brillantes au-dessus d’une grande fleur blanche…
Avant que Marthe s’éveille, nous sommes déjà loin, Pío, Navarro
et moi ; une voiture légère nous emmène au hasard, eux, le fusil sur
l’épaule, moi, une ombrelle à la main, ivres de lumière et de liberté.
J’avais protesté contre les fusils, mais j’en ai compris l’utilité quand
j’ai vu se lever presque sous les pieds du cheval un nombre
incroyable de lièvres.
— Un imprudent, m’expliqua Carlos, un imprudent qui était aussi
sans doute un chasseur enragé, a introduit dans le pays le lièvre qui
y était inconnu, et nous sommes maintenant obligés de défendre
nos cultures contre ce rongeur infatigable qui pullule dans les
champs. C’est à le chasser que sont employés ces beaux lévriers que
vous avez caressés ce matin, et c’est vraiment en cas de légitime
défense que nous le tuons, car sa chair ne vaut pas grand’chose, et
on ne sait pas encore chez nous utiliser sa peau.
Nous avons gagné la grand’route… ou plutôt l’espace séparant
les terres cultivées qui sert de chemin… La voiture bondit, les
ressorts plient, nous sommes jetés les uns contre les autres… Voici
que nous nous trouvons face à face avec un immense troupeau de
bœufs et de vaches… J’ai un petit frisson… Le cocher, de deux coups
de fouet, s’ouvre un passage et les bêtes s’écartent pour nous laisser
passer…
— Les bœufs ne sont dangereux que pour les piétons, me dit Pío,
les hommes qui les gardent sont des « gauchos », toujours à cheval,
et la silhouette d’une voiture leur est relativement familière, mais
l’homme à pied est pour eux un animal inconnu, contre lequel ils
foncent volontiers à coups de corne.
 Le troupeau disparaît dans la poussière qu’il a soulevée, de gros
rats sans queue traversent la route, des oiseaux chantent à plein
gosier sur les arbustes qui la bordent, nous allons toujours… Il est
onze heures, lorsque nous revenons à l’estancia, ivres de soleil,
poussiéreux, affamés, contents. Une demi-douzaine de lièvres sont
envoyés à la cuisine, et Mamita fait la grimace en les regardant. Elle
gronde Pío et Carlos de m’avoir emmenée sans sa permission.
— Regardez-la un peu, leur dit-elle, elle a au moins un kilo de
poussière sur ses vêtements ! Et moi qui voulais lui montrer la basse-
cour, et aller avec elle chercher des œufs ce matin… il y a au moins
deux cents poussins de plus qu’hier !…
Je console la chère vieille dame, et je lui promets de
l’accompagner l’après-midi, après la sieste.
— Nous irons voir les poulets, les dindons, les oies, les canards,
tout, Mamita, tout !

. . . . . . . . . . . . . . . . . . . .

J’ai tout vu, je n’en puis plus. Mais Mamita est si contente et si
fière ! Pío me remercie d’un regard et nous nous asseyons tous deux
sur le banc, près des lys…

. . . . . . . . . . . . . . . . . . . .

Pendant trois jours un vent furieux a secoué la maison, les

champs se moiraient sous son souffle, les pétales de fleurs
s’éparpillaient dans les allées, et les saules qui bordent la petite
rivière se pliaient échevelés jusqu’à caresser l’eau de leurs feuilles.
Nous connaissions ce vent, le « Pampero », pour l’avoir subi à
Buenos-Aires, mais là, dans ce bouquet d’arbres isolé au milieu de la
plaine, sa violence se décuple. Je suis sortie souvent pour la sentir
passer, cette grande haleine fraîche de la pampa. Elle est pure,
impétueuse, on sent qu’elle a passé sur des neiges immaculées et
frôlé des terres vierges et lorsqu’elle se tait, elle laisse le ciel plus
bleu, et la nature plus vigoureuse.
Notre demi-emprisonnement a été délicieux, et notre intimité
avec nos compagnons de captivité s’est accrue ; j’ai découvert dans
Mamita des trésors de bonté, et dans ses hôtes mille ressources
intellectuelles. La facilité d’assimilation de mes futurs compatriotes
est prodigieuse : ils savent tout ce qu’ils ont envie de savoir !
Lorsque les méthodes d’enseignement auront progressé, que l’on
élaborera des programmes d’éducation moins touffus, et qu’on se
donnera la peine de faire approfondir aux jeunes gens les matières
qu’on leur enseigne, les Universités et les Facultés argentines
rivaliseront facilement avec celles du vieux monde, comme le fait
déjà leur Faculté de Médecine.
Pío m’a montré la bibliothèque, commencée par son aïeul, et que
son père et lui ont accrue avec soin et orgueil. Que de belles et rares
éditions, en espagnol, en allemand, en italien, en français ! Et quelle
variété de sujets ! Depuis les romans et les poèmes, jusqu’aux traités
d’agriculture et de géologie !
— Presque tous mes amis ont des bibliothèques mieux fournies
que la mienne, m’a dit mon fiancé, nous lisons beaucoup ici, mais
aucun n’aime ses livres comme j’aime les miens !… les nôtres, a-t-il
ajouté… car il a vu avec quelle avidité mes yeux parcouraient les
titres des volumes, et avec quelles précautions je touchais ces
reliures précieuses, nous viendrons lire souvent ici tous deux et
parler de la littérature de votre pays pour laquelle vous voyez mon
admiration, et qui est familière à toute la classe supérieure en
Argentine.
Pío me tend un beau volume habillé de maroquin havane, qui
porte en lettres d’or ces mots : La Gloria de Don Ramiro.
— Voilà un de mes livres favoris, et je voudrais que vous l’aimiez,
il est merveilleusement écrit et noblement conçu. L’auteur se nomme
Enrique Rodriguez-Larreta…
— Le ministre d’Argentine à Paris, interrompis-je, je connais son
livre.
 — Il a mis des années à finir ce roman, il en a ciselé le style, il
est l’espoir et l’orgueil de notre littérature.
— D’autres vont suivre son exemple, mon cher Pío, il a montré à
vos jeunes hommes de lettres que rien de beau ne s’improvise, c’est
une grande et utile leçon qu’il leur donne !

. . . . . . . . . . . . . . . . . . . .

Le pampero s’est tu, nous voici délivrés. Je suis allée ce matin à

cheval avec Pío, faire un peu la connaissance des « gauchos » qui
gardent et soignent les troupeaux de l’estancia, et nous sommes
entrés dans les petites maisons de brique crue, semées de loin en
loin dans la plaine. En général, la maison était vide, et la porte
cependant grande ouverte… Malgré des antipathies et même des
haines qui s’élèvent entre ces hommes violents et passionnés, la
méfiance n’existe pas, et le vol est inconnu. Et puis, que volerait-on ?
Un lit de camp, qu’on appelle « catre », une selle couverte de peaux
de mouton, le « recado », une calebasse pour faire le maté sans
lequel le paysan argentin ne peut vivre, une guitare, et c’est tout ce
que contient la petite maison solitaire ! Dès l’aube, le gaucho la
quitte à cheval, et revient à la nuit tombante, drapé dans son
« poncho » et brisé par d’interminables galops dans la pampa… C’est
alors, qu’assis devant sa porte, sa guitare sur les genoux, regardant
la plaine qui perd peu à peu ses couleurs, il invente ces chants
languissants, au rythme arabe, qui montent dans le soir comme le
soupir même de la campagne exténuée par une journée d’ardent
soleil ; il chante aussi sa fiancée, celle qu’il aimera uniquement, qui
viendra lui apporter le sourire de son visage brun, et les petits qu’il
verra se rouler dans l’herbe, se hisser sur les chevaux, qui lui
crieront au retour de ses épuisantes chevauchées des mots tendres
et rauques, et qui, plus tard, chanteront à leur tour, devant la
pampa, obscurcie par le soir, leur fiancée brune et l’avenir de leurs
enfants.

. . . . . . . . . . . . . . . . . . . .
 Nous sommes allés jusqu’au petit village, le « pueblo » qui
dépend de la station « Dos Ombuës » : quelques petites maisons de
briques recouvertes de ciment, dont les toits plats servent de
terrasse, une école d’où s’échappent en troupe de beaux enfants
bronzés et turbulents, l’église sans clocher, et l’inévitable, multiple,
indispensable « almacen ». Le mot « almacen », mot arabe, veut dire
« magasin », et dans l’esprit du paysan argentin, il représente
l’endroit où tout se vend et tout s’achète. Que ne trouve-t-on pas
dans l’almacen d’un village ? Des toupies et des harnais, des
espadrilles et de la bière, des rubans et des fromages, du vin et des
fers à friser, sans compter les graines, l’encre, les pièges à mouche,
les conserves de viande et de légumes, les confitures et le pétrole…
La vente des étoffes est réservée aux colporteurs turcs qui vont
d’estancia à estancia, infatigables, sachant dix mots d’espagnol, et
qui quelquefois sont aussi riches que l’estanciero aux domestiques
duquel ils vendent leurs cotonnades bariolées.
— Vous devez trouver mon pays bien sauvage, me disait Pío,
tandis que nos chevaux prenaient la route du retour.
— Non, Pío, je l’aime, votre pays si jeune et si riche… Voyez quel
attrait il possède : l’almacenero est espagnol, le maçon qui construit
les maisons est italien, des Russes courbent le dos dans ce champ
labouré, et tous y sont venus de tous les coins du monde, attirés par
l’appât d’une fortune facile et d’une vie indépendante… ils travaillent
rudement et peinent souvent, mais riches ou pauvres, ils resteront !
Ils oublient leur langue pour parler votre espagnol doux et zézayant,
leurs enfants vénèrent les héros de l’indépendance argentine, et
dans deux générations, leurs descendants n’auront qu’un nom
étranger pour leur rappeler qu’un des leurs est venu de si loin… Et
vous le voyez, moi aussi, je reste !…
— Dieu en soit loué, ma chérie !

. . . . . . . . . . . . . . . . . . . .
 Hier après le dîner, nous sommes retournés au village, entendre
les chants des « gauchos » et voir leurs danses. Sauf la maman de
Pío, un peu fatiguée, tous les hôtes de l’estancia nous ont
accompagnés. Nous nous sommes assis sur des bancs, autour d’une
place carrée plantée de poivriers géants au feuillage floconneux, et
le spectacle a commencé. Un payador, — un improvisateur, — s’est
détaché du groupe que formaient les gens du village, et se plaçant
sur une chaise, il a commencé à accorder sa guitare.
Un silence profond s’était établi… La guitare bourdonnait comme
un essaim d’abeilles d’argent et il a lancé les premières notes
tendres et plaintives d’un « Triste »… J’étais suspendue à cette voix
un peu gutturale, un peu voilée, qui chantait les peines d’amour du
gaucho solitaire, et son désir de mourir si celle qu’il aimait lui
refusait son cœur. Quelle passion criait dans cette mélodie primitive,
et quelle ardeur désolée ! Un autre nous dit une « vidalità », puis des
« milongas » lui succédèrent, et je ne me lassais pas d’écouter,
d’écouter encore ces voix incultes et puissantes. Les chanteurs
improvisaient souvent, quelques couplets ironiques se mêlaient aux
lamentations passionnées et toute leur âme passait dans la
musique… Ils se turent et la danse commença.
Quelques-uns des danseurs venaient de loin, et arrivaient à
cheval, portant parfois en croupe leur « novia ». Les hommes étaient
tous revêtus du costume national : des pantalons blancs garnis de
dentelles et recouverts d’une pièce d’étoffe qui passe entre les
jambes : « le chiripá », une veste courte, un mouchoir de soie
éclatante au cou, et sur des bottes étroites, des éperons d’argent
travaillé comme les boucles de leur ceinture, comme leurs étriers et
comme le harnachement de leurs chevaux. Les femmes dont la
figure un peu plate a un grand charme, portaient des robes claires et
empesées et une longue natte de cheveux noirs s’allongeait
jusqu’aux genoux des plus jeunes…
Celui qui conduisait la danse frappa dans ses mains, la musique
commença, les couples se formèrent et nous vîmes se dérouler les
figures charmantes et harmonieuses de la danse nationale
argentine : le Péricon. Avec des attitudes nobles, avec des
mouvements pleins d’une grâce sauvage, les danseurs mettaient un
genou en terre, se fuyaient, se retrouvaient, passaient sous un bras
étendu et faisaient la chaîne, les mains séparées par les foulards
qu’ils avaient retirés de leur cou. Entre chaque figure, un d’entre eux
s’avançait et improvisait deux vers amoureux ou plaisants qu’il
adressait à une des danseuses : celle-ci ripostait aussitôt, et la danse
reprenait. La lune faisait briller l’argent des éperons et des ceintures,
les robes passaient et repassaient comme des nuages blancs, les
pieds touchaient doucement le sol, et lorsque la musique se tut,
lorsque, fatigués, les couples se réunirent aux gens du village assis
sous les grands arbres, je poussai un soupir de regret.
— Et dire qu’en Europe on croit que c’est l’infâme et dégradant
tango qu’on danse ici ! m’écriai-je. Comme je voudrais pouvoir
démentir ceux qui ont créé cette légende !
— Consolez-vous, ma chère amie, me répondit Carlos Navarro, la
mode stupide du tango a déjà disparu de l’autre côté de l’Océan ;
tous ceux qui l’ont acceptée en ont été quittes pour avoir imité
pendant quelques mois les voyous des villes argentines et leurs
compagnes…
Nous fîmes nos adieux aux gauchos que je remerciai et qui nous
accompagnèrent jusqu’aux voitures avec mille souhaits courtois de
bonheur et de santé.
Pendant le retour dans la nuit tiède, aucun de nous ne parla, Pío
serrait ma main dans la sienne et une infinie douceur nous
engourdissait…
Je regardais les étoiles de mon nouveau ciel, plus rares et plus
brillantes que celles de l’autre hémisphère, et la route me parut bien
courte, car mon cher fiancé murmurait à mon oreille les mots
immortels qui font oublier, qui font espérer…
Ce jour est celui qui précède notre départ. Marthe ne tient plus en
place, malgré l’affection et les distractions qu’elle a trouvées à
l’estancia. Elle ne dit pas une phrase sans y introduire le nom de
Georges, et lorsque Mamita a fait des efforts pour retarder le
voyage, les yeux de la pauvre enfant se sont remplis de larmes. Je
comprends si bien son désir de revoir celui dont elle est séparée
depuis tant de mois que j’ai résisté énergiquement aux objurgations
de Mamita et aux prières de Pío.
Nous partons demain… mais pas seules ! mon fiancé nous
accompagne ; il a décidé cela sans me le dire, et il a prié Marthe en
riant, d’être notre chaperon.
Cher Pío ! vous n’avez pas voulu me laisser retomber dans mon
isolement, avec le spectacle du bonheur des autres devant les yeux !
Je reconnais là votre amour attentif et la délicatesse de votre cœur…
et, égoïste que je suis, je me réjouis de vous faire quitter la maison,
les amis, la maman même, pour me suivre et me protéger !
Tout est prêt, et Georges est averti. Pío a un ami dans le pays où
nous allons, un vieil Anglais qui s’est épris d’un site, et qui veut
mourir devant les montagnes qu’il aime ; une dépêche a avisé cet
original de notre arrivée et sa maison sera une des étapes de notre
route vers la mine de « la Carlota », résidence de Georges. Notre
équipement est déjà à la gare, nos adieux sont faits et j’ai entendu
Mamita me murmurer à l’oreille : — Au retour… je vais faire arranger
votre appartement…
Oh ! embrassez-moi bien fort, ma nouvelle maman… dans vos
bras, je redeviens une petite fille… vous savez si bien aimer vos
enfants !

. . . . . . . . . . . . . . . . . . . .

Des heures, des heures encore, et de la poussière. J’en ai dans

les cheveux, dans la bouche, on en mange au restaurant pendant
que la plaine se déroule éternelle de chaque côté des rails. Pío nous
entoure d’attentions, mais Marthe ne voit rien, n’entend rien… elle
va rejoindre son mari, et voilà tout !
L’horizon commence à onduler, voici des ruisseaux, de grandes
pierres, et des collines bleuâtres au loin : la Cordillère commence…
Nous nous sommes arrêtés à Cordoba, où mille couvents font
retentir leurs cloches parmi des jardins fleuris. L’air est pur et frais,
c’est ici que les malades viennent trouver la santé.
Nous voici de nouveau en route vers le nord. Le train s’arrête
souvent, et les visages aperçus dans les gares pareilles à des
hangars sont basanés. Dans quelques heures, nous serons au village
où s’arrête le chemin de fer ; c’est à cheval, et par étapes, que nous
continuerons notre voyage.
Le paysage a changé, nous traversons d’immenses vallées ; la
locomotive, chauffée au bois, a ralenti sa marche, la poussière a
envahi les vitres, et je sors sur la plate-forme qui termine notre
wagon pour mieux voir la contrée dans laquelle nous sommes
entrés.
Des montagnes prodigieuses s’élèvent à droite et à gauche, et le
sol est d’un jaune clair, coupé de failles profondes et envahi par des
arbustes épineux. Voici des vols de perruches vertes, de petits
oiseaux blancs aux ailes bordées d’un liseré noir ; ces trous sont des
terriers de « zorrinos », les skungs du pays, ou de viscachas : de la
carcasse desséchée d’une mule s’élèvent de petits vautours…
 Les montagnes se font de plus en plus hautes… Nous arrivons :
le chemin de fer ne va pas plus loin…

. . . . . . . . . . . . . . . . . . . .

Le prévoyant, le sage Georges a envoyé toute une caravane à

notre rencontre : des hommes, des chevaux, des mulets.
Sachant que Pío nous accompagne, il est resté dans la montagne
où il fait double travail pour revenir plus vite avec nous à Buenos-
Aires. Marthe comprend et n’est pas trop déçue…

. . . . . . . . . . . . . . . . . . . .

C’est une auberge charmante qui nous abrite… Les servantes ont
le type indien, les pommettes saillantes et les cheveux plats, elles
nous servent en souriant de leurs belles dents égales…
Derrière la maison s’élève un bois d’orangers géants pleins de
fruits et de fleurs… toutes les maisons du « pueblo » sont en boue
grise…

. . . . . . . . . . . . . . . . . . . .

Le soleil se lève, les chevaux sont sellés, les mules chargées,

nous partons… Il nous faut une semaine pour arriver à « la Carlota »
par petites étapes, bien que le pauvre Georges ait fait la route en
quatre jours…
Comme cette première journée a été rude !
Marthe est rompue et je quitte la selle avec joie… Mais quelles
admirables et singulières routes nous suivons !… Toujours cette terre
pâle, maintenant couverte de cactus monstrueux, grands comme de
grands arbres et tous en fleur… la lumière rayonne…
Nous avons marché assez lentement, il nous a fallu coucher sur
des « catres », dans une cabane de berger… et d’un sommeil si
profond ! Je croyais, lorsque Pío nous a fait appeler pour repartir que
je venais à peine de fermer les yeux !…
Tous les chevaux sont réunis autour d’une jument qui porte une
clochette au cou, c’est la « madrina », elle est dressée à les conduire
à l’endroit où les « gauchos » veulent les brider et les seller. Pío
choisit des montures pour nous, puis pour lui. Ces chevaux à demi
sauvages sont d’une douceur et d’une indifférence déconcertantes ;
celui que je monte aujourd’hui est venu se ranger avec les autres
autour de la « madrina », on lui a jeté le lazzo autour de son
encolure fine, et il s’est laissé caresser et conduire comme s’il me
connaissait depuis toujours… et il ne me connaît pas, je le sais, il ne
me connaîtra jamais, même s’il m’obéit pendant des semaines ; ce
qu’il désire, c’est la course, l’espace, le ruisseau verdi d’herbes
fraîches, et dépasser les autres ! Quand je le laisserai libre, quand
mon poids aura cessé de meurtrir sa croupe, il se roulera sur la terre
dure, et courra vers ses compagnons en hennissant… il sera
débarrassé…
Il faut arriver au coucher du soleil à la maison de l’ami de Pío, cet
Anglais rêveur et excentrique, amoureux d’un paysage ; nous
partons au galop, un galop doux et cadencé.
Les gauchos et les muletiers nous suivent, allument le feu
pendant les haltes, et nous servent avec une dignité grave…
Les heures passent, chacune éclaire une beauté nouvelle, nous
approchons de la fin de l’étape, et à mon insu j’ai dépassé mes
compagnons…
Tout à coup, mon cheval hennit ; dans la splendeur du jour qui
m’éblouit, son instinct lui fait pressentir la nuit toute proche ; je hâte
son petit galop balancé ; avec une adresse merveilleuse, il évite les
épines des cactus, les trous des viscachas, les failles du terrain, et il
va, il va vers l’eau, vers l’herbe, vers le soir qui est le repos. Voici
déjà une maison de boue, une treille, des chiens… à droite, à
gauche sur mon passage, s’élèvent et retombent lourdement, les
dindons sauvages, beiges et bleus, des cris inconnus traversent l’air
léger et une odeur composée de tous les parfums m’arrive portée
par le vent, c’est l’odeur de l’espace… Je m’arrête, mon cœur bat
furieusement, mes yeux se fixent devant moi, ce que j’ai vu à cette
minute, je sais que je ne l’oublierai jamais. Par-delà la maison, de
l’autre côté de la vallée, s’élève une chaîne de montagnes, en pierre
rouge, d’un rouge sombre et calciné et dont la crête irrégulière mord
le ciel de toutes ses dents. Le soleil, en mourant, jette son or sur
cette pourpre qui le lui renvoie en poudre et en rubis ; deux grands
oiseaux planent, si haut qu’ils peuvent voir l’autre côté de la muraille
ardente. Ce doit être beau ce qu’ils voient, car ils ne redescendent
pas… Est-ce la lumière du bonheur dont je ne peux apercevoir qu’un
reflet, qui éclaire l’autre versant de ces montagnes ? Quel royaume
mystérieux a jeté tout son or de ce côté pour ne garder que
l’amour ? Ce sont des paillettes qui étincellent aux cassures du
rocher, et des pépites qui brillent dans les fissures… Est-ce El Dorado
qui existe puisqu’un homme l’a inventé ?… Et je reste sans pensée,
sans souffle, à regarder, jusqu’au moment où l’ombre démesurée qui
s’allongeait devant moi cesse de me montrer la route du royaume
secret… Ce ne sont que des pierres, des pierres éternelles qu’essaie
de ronger un lichen livide, il n’y a rien, rien… et cependant, je me
promets de revenir demain chercher le chemin de mon rêve guidée
par l’ombre qui n’a pas eu le temps aujourd’hui de me le montrer…

. . . . . . . . . . . . . . . . . . . .

La maison de briques crues est exquise, et notre hôte, M. Danley,

nous en fait les honneurs avec autant de cérémonieuse politesse que
s’il recevait à Londres des membres de la famille royale.
Un excellent dîner nous réunit à sa table, servi par des demi-
indiens qu’on appelle ici des « Chinos ». Après les premiers
compliments, mis en verve par notre présence, notre hôte
commence à taquiner Pío en lui faisant quelques critiques sur ses
compatriotes.
— Voyons, mon cher ami, lui dit-il dans un espagnol imparfait et
pittoresque, avouez que rien au monde n’est plus beau que ce
paysage, et que les Argentins sont fous d’aller de l’autre côté de la
terre contempler des sites qui pâlissent auprès de ceux que vous
avez traversés depuis votre départ de Buenos-Aires ?
— Mais, répond Pío, personne à Buenos-Aires ne connaît nos
provinces, et…
— C’est le tort qu’on a, riposta Danley, on devrait les connaître ! A
peine quelques « porteños » vont-ils en convalescence à Mendoza ou
à Córdoba, et ils reviennent dégoûtés des hôtels et des auberges, et
navrés parce qu’ils n’ont pas trouvé à quatre mille pieds d’altitude un
cinématographe où passer leur soirée ! Et pourtant, Mendoza est
admirable, la montagne y est sauvage et terrible…
— Elle est bienfaisante aussi, ajouta Pío, les sources médicinales
chaudes en jaillissent, et font des cures merveilleuses…
— C’est pourquoi vous allez à Carlsbad…
— Rosario de la Frontera aussi possède des eaux qui guérissent
les affections du foie, et des promenades splendides dans des forêts
millénaires…
— C’est pourquoi vous allez à Vichy !…
— Un peu de patience, attendez que nous ayons un siècle de
plus…
— J’attendrai !
Et il est bien capable d’attendre par curiosité…
Notre fatigue ne nous empêcha pas de sortir après le dîner. La
montagne, devenue mystérieuse, nous attirait et nous fascinait. Très
haut, très loin, des feux brillaient…
— Votre mari est là, dit M. Danley à Marthe, ces lumières
marquent l’entrée de la mine où il travaille, et le poste du chemin de
fer aérien qui descend le minerai dans la vallée…
Marthe trouve sûrement que ces petites lampes sont plus belles
que les étoiles, car ses yeux ne les quittent plus !
Le chemin que nous faisait suivre notre hôte monte
insensiblement, et surplombe son jardin. Des lucioles dansaient dans
la nuit claire, et de petits hiboux jetaient des cris qui ressemblaient à
un rire triste.
— Voici un nid de condors…
La lune se levait, sa lumière atteignit l’angle d’un rocher, et nous
vîmes s’agiter d’énormes masses grises, tandis qu’une sorte de râle
parvenait à notre oreille…
Un désir ardent me venait de contempler de plus près l’hôte
formidable des Andes… hélas ! des pentes abruptes et des précipices
me séparaient de ce nid de géants ailés… Il me fallut attendre au
lendemain pour voir de près, sous le soleil, un condor captif dans le
« patio » de la maison de M. Danley. J’appris que le grand oiseau de
proie dort à peine, — la légende lui attribue même une insomnie
perpétuelle, — et qu’il est presque impossible d’approcher son nid,
placé toujours sur des pics à peu près inaccessibles. Cependant,
quelquefois, un chasseur hardi, ou aidé par le hasard, découvre un
poussin que ses parents, qui chassent toujours ensemble, ont
abandonné momentanément, il s’en empare, et après lui avoir mutilé
une aile, il l’installe dans la cour de sa demeure. Là, le condor
grandit, vieillit et meurt sans avoir connu l’ivresse de planer et la joie
aiguë de fondre sur une proie vivante.

. . . . . . . . . . . . . . . . . . . .

Nous avons quitté l’hospitalière petite maison avec regret…

Reverrons-nous cet homme si fin, si clairvoyant, cet amoureux
fervent de la beauté ? Il nous a promis de venir nous voir à Buenos-
Aires, et nous a montré avec orgueil les fils télégraphiques qui
suivent la vallée :
— Je vous enverrai une dépêche, nous dit-il, les communications
sont si faciles !
Et c’est vrai ; je constate tout à coup combien ces régions à peu
près désertes sont facilement reliées à la capitale. M. Danley,
Georges même, ont eu en deux heures le télégramme annonçant
notre départ de l’estancia, et on sait au fond des mines les nouvelles
du vieux et du nouveau monde chaque matin !

. . . . . . . . . . . . . . . . . . . .

Enfin ! nous avons atteint le but de notre voyage, Marthe est

tombée en sanglotant dans les bras de son mari, qui, lui-même,
avait les yeux humides, pendant que Pío et moi les regardions tout
émus.
Après un déjeuner qui nous a été offert par les ingénieurs, nous
sommes allés visiter les galeries, voir transporter le minerai dans les
wagonnets qui le descendront le long d’un câble d’acier, jusqu’au
fond de la vallée.
La mine est à près de quatre mille mètres d’altitude. Au flanc
d’une montagne voisine, nous pouvons voir, petite et noire comme
l’entrée d’une fourmilière, l’ouverture béante d’une autre mine, elle
contient de l’or, celle-là.
Georges nous montre comment fonctionnent les machines et
s’enthousiasme en nous décrivant les incalculables richesses
minérales de l’Argentine. Son séjour ici n’a pas seulement raffermi sa
santé et bronzé son visage, il lui a donné la conviction absolue que
presque toutes ces montagnes contiennent des trésors, et des
trésors plus accessibles que ceux déjà exploités ; il a beaucoup causé
avec ses compagnons de travail et avec les géologues que la
compagnie leur a adjoints, tous sont de son avis, et plusieurs ont
déjà fait des explorations probantes.
— Il y a de tout dans ce pays, s’écrie-t-il, de l’or, de l’argent, du
cuivre, de l’antimoine, du fer, du charbon et même du pétrole. Mais il
faut des hommes pour exploiter tout cela… et on a mille peines à en
trouver ; les émigrants restent à la ville, et les « chinos » préfèrent le
travail au grand air. Ah ! si nous avions la main-d’œuvre nécessaire !
Georges a attrapé la maladie du pays, la « hâte », et il hausse les
épaules quand l’un de nous prononce le mot « patience » !
 Nous avons cependant patienté trois jours avant de reprendre le
chemin du retour, en nous émerveillant de la résistance dont mon
cousin a fait preuve, et de son courage au travail. Ceux qui restaient
nous regardaient avec envie, mais aucun ne songeait à déserter sa
tâche, car tous sentent qu’ils partiront riches et libres, et c’est une
perspective qui permet d’attendre !
La mission de Georges est terminée pour cette année, il l’a
accomplie à merveille, le voilà sur la route de la fortune…
Nous avons fait nos adieux aux mineurs, et regardé une dernière
fois le panorama merveilleux de la profonde vallée hérissée de
cactus fleuris, enserrée de murs de granit, et j’emporte au cœur le
désir de revenir, et un amour plus fort pour ma nouvelle patrie.

. . . . . . . . . . . . . . . . . . . .

Les montagnes se sont effacées à l’horizon, la plaine s’étend de

nouveau devant nos yeux, presque dorée maintenant : voici la
banlieue et ses villas, voici les faubourgs, et c’est Buenos-Aires.
Mamita est à la gare, elle s’extasie sur notre bonne mine, et ne
me cède qu’à regret à nos amis qui l’ont accompagnée. On lui
présente Georges, et nous regagnons la petite maison où je suis
arrivée avec tant de mélancolie et si peu d’espoir, et dans laquelle je
reviens étourdie de bonheur et fière d’un amour partagé…

. . . . . . . . . . . . . . . . . . . .

En attendant l’heure de m’appeler « la Señora de Valdez » je vis

presque chez Mamita, elle n’a pas d’autre enfant que Pío, et nous
gâte tous deux tant qu’elle le peut ; elle nous a cédé un étage de sa
grande maison, car selon la coutume argentine nous ne quitterons
pas le toit des parents ; la sœur de Mamita continuera aussi à vivre
avec elle ; c’est une vieille fille qui est restée célibataire par goût. Ce
type est assez rare dans un pays où toutes les femmes trouvent un
mari, mais il existe, et la tante Victoria en est un exemple : un
exemple doux, aimable et effacé. Elle ne se manifeste à nous que
par des envois de cadeaux et des apparitions furtives. Son
appartement tranquille, presque monacal, est aussi fermé que
possible aux bruits multiples de la rue, et je m’y réfugie parfois, à sa
grande joie, lorsque quelque cousine éloignée vient faire une visite
accompagnée de sept ou huit enfants turbulents… Mamita supporte
les cris de ces petits diables déchaînés avec une patience que
j’admire, et arrive à les faire taire en les gavant de gâteaux et de
« dulce de leche » qu’ils dévorent voracement, imités souvent par
leur maman. Ce goût pour les sucreries et l’absence d’exercice —
une femme comme il faut ne se promène guère à pied —
m’expliquent l’embonpoint que prennent si tôt les Argentines, au
détriment de leur beauté. On m’a dit que depuis quelques années,
les femmes avaient pris l’habitude de la marche, et j’en ai en effet
rencontré quelques-unes au Bois de Palermo, mais je crains que
l’amour des gâteaux et des bonbons ne leur passe pas avant bien
longtemps…
Carmen vient souvent, soit chez nous, soit chez Mamita, toujours
affectueuse et charmante ; elle m’amène presque toujours son petit
garçon : Carlitos, avec lequel je parle français. Le délicieux enfant ! Il
a toute la précocité des petits porteños, mais ses parents ont
contenu l’exubérance de sa nature sans en retirer la spontanéité, et
Carmen m’assure que la méthode d’éducation qu’elle emploie est
désormais en vigueur dans beaucoup de familles, et que l’excessive
liberté qu’on laissait aux enfants tend heureusement à se
restreindre.
Je vois peu Pío tout occupé des préparatifs de notre mariage, il
aura comme témoins, comme « padrinos », Daniel Cruz et Carlos
Navarro ; mes témoins à moi seront Georges et M. Roy. Mamita est
folle d’émotion et fait une exposition des cadeaux qu’on nous envoie
chaque jour… Et il y en a plein une chambre. La générosité argentine
n’est pas un vain mot : des bijoux, des dentelles, des meubles, des
porcelaines, de l’argenterie ! Marthe range tout cela sous l’œil
attendri de Mamita qui approuve ou fait la moue selon les objets
qu’on apporte. Rien n’est assez beau pour ses chers enfants, et ce
mariage de Pío qu’elle désirait depuis si longtemps qu’elle avait
cessé de l’espérer, lui semble un événement destiné à bouleverser le
monde.
Comme toutes les femmes argentines, Mamita est tout amour, et
pour elle, celui ou celle qui n’aime pas est au ban de l’humanité !

. . . . . . . . . . . . . . . . . . . .

Mon fiancé m’a apporté tantôt une bague, un diamant

magnifique, et ma joie doubla visiblement le plaisir qu’il avait à
m’apporter cette belle pierre.
— Deviendriez-vous coquette, ma chérie ? me demanda-t-il en
souriant.
— Vous faites tout ce qu’il faut pour que je le devienne, lui
répondis-je, et puis, nous ne sommes guère habituées en France à
de pareilles prodigalités. Nos jeunes filles ne portent pas de bijoux et
s’habillent avec une extrême simplicité, aussi lorsqu’elles sont
mariées, chaque cadeau que leur fait leur mari les enchante ; elles
ne grossissent leurs écrins que peu à peu, et c’est lorsqu’il faut
cacher leur cou et leurs épaules qu’elles mettent des colliers… En
voyant ici des enfants de seize ans porter des perles et des brillants,
je me suis souvent demandée de quelles pierreries et de quels
joyaux elles pourront se parer plus tard, et quelles dépenses celui
qui les aura épousées devra faire pour leur offrir un présent qui ne
leur paraisse pas médiocre !… Et elles sont si charmantes, les jeunes
Argentines, avec leur éclat de fleur librement poussée, leur teint
merveilleux, leur intelligence ouverte et avide de savoir ! Elles n’ont
pas besoin de surcharger tant de grâce par des ornements qu’excuse
le déclin de la beauté. J’en ai rencontré beaucoup à Paris, surtout
aux conférences des Annales où les conduit leur désir d’apprendre,
et je les ai comparées aux petites Françaises qui les entouraient, et
qui certes, étaient pour la plupart moins jolies qu’elles. Eh ! bien, je
ne pouvais jamais distinguer une jeune fille argentine de sa sœur
mariée, et je lui donnais toujours, j’en suis sûre, cinq ou six ans de
plus que son âge, grâce à son excès d’élégance dans la toilette, et
aux bijoux dont elle était couverte.
— Mais nos jeunes femmes ?
— Ah ! les jeunes femmes, c’est autre chose ! Là, il n’y a rien à
redire, rien à critiquer ! Elles sont vêtues avec le goût le plus délicat
et le plus sûr, et affectionnent les teintes discrètes aussi bien que les
robes de la coupe la plus distingué et la plus raffinée. Lorsque je suis
arrivée, faut-il vous l’avouer, Pío ? j’apportais quelques préjugés
contre les femmes argentines, je les croyais un peu ignorantes et
très frivoles, je confondais leur dévotion sincère et traditionnelle en
même temps, avec une religiosité machinale… Vous voyez que je
vous dis tout ! Je vous confesse avec la même sincérité que ces
préjugés absurdes ont fondu comme la neige au soleil au contact de
mes nouvelles amies, et à mesure que je vivais parmi elles. J’ai
trouvé dans Carmen une créature d’élite, un cœur tendre, un
cerveau ardent, Lucia Iturri de Hambourg est une merveille de tact,
de finesse et de fidélité dans l’amitié, Délia Ortiz, notre chère Délia,
n’est-elle pas le modèle des mères et des filles, et n’a-t-elle pas au
suprême degré, l’art de grouper les intelligences et de faire briller
l’art sous toutes ses formes ? Et Léonor Cruz dont le père est
propriétaire de votre grand journal ! Y a-t-il un esprit plus alerte, une
charité plus admirable, et une âme plus élevée que la sienne ? Votre
Gloria Villalba, belle comme une jeune déesse et savante comme un
recteur d’Académie, ne serait-elle pas admirée en Europe par les
juges les plus difficiles ?… Et tant d’autres ! Et Mamita ! chère
Mamita, qui ne vit que pour ceux qu’elle aime et qui s’oublie toujours
en pensant aux autres !… Quelles mères sont ou seront toutes ces
femmes ! Comment n’aurais-je pas été conquise ? Il faut être aveugle
ou fou pour ne pas voir que les Argentines préparent des
générations dignes d’un grand pays ! Et c’est parce que je les
connaissais déjà que j’ai été si surprise par mon bonheur, mon cher
fiancé, et par votre choix !
— Oh ! ma trop modeste novia, vous voulez seulement, n’est-ce
pas, me faire répéter que je vous aime ?
 — Mais dites-moi, Pío, n’aviez-vous pas, vous aussi, quelques
préjugés contre les Français ?… contre les Françaises ? ajoutai-je plus
bas.
— Si, j’en avais ! Pourquoi vous le cacher, ma chérie ? Comme
ceux que vous venez de m’avouer, ils se sont dissipés lorsque je
vous ai connue, aimée, et lorsque je suis entré dans l’intimité de vos
cousins. J’ai fait avec de jeunes amis, un premier voyage en Europe
lorsque j’avais dix-huit ans, et tout comme eux, j’ai connu le pire de
la vie de Paris ; nous avions beaucoup d’argent à notre disposition,
une liberté illimitée, et aucun de nous n’était majeur… vous voyez
que toutes ces circonstances concordaient à merveille pour nous
faire faire des sottises… nous n’y avons pas manqué, hélas ! Et je ne
compte pas l’esprit d’indépendance que nous possédons si fort, et le
fait qu’une lettre d’avertissement et de remontrances devait mettre
vingt jours pour nous atteindre ! Nous étions déchaînés, tout
simplement, et ce que nous avons connu de Français, et de
Françaises surtout, à cette époque, n’était pas fait pour nous donner
une opinion très bienveillante de nos frères latins ! Plus tard, lorsque
je suis retourné à Paris, j’ai évité les erreurs passées, mais sans
pénétrer dans la société parisienne, plus fermée encore que la nôtre.
Je n’ose vous parler des Parisiennes, ou soi-disant telles, qui hantent
les cafés-concerts de Buenos-Aires, mais je puis vous assurer
qu’elles ne font rien pour soutenir le bon renom de la nation à
laquelle elles disent appartenir, et malheureusement, ce sont celles-
là qui s’expatrient ; les autres les « vraies » voyagent trop peu pour
se hasarder jusqu’en Amérique du Sud. Vous vous rendez compte
combien j’étais excusable de si mal connaître dans ses nationaux, un
pays que j’aime et que j’admire, mais qui ne m’apparaissait qu’à
travers les cafés de Paris, les music-halls de notre capitale, et
quelques magasins où des commerçants trop avides nous exploitent
sans merci. Désormais, outre l’admiration que j’ai depuis toujours
pour les artistes, les hommes de lettres et les savants de votre
patrie, je me suis fait la conviction que les hommes qu’elle produit
sont honnêtes, droits et courageux, que leurs femmes aiment et
respectent leur foyer, tout en le rendant attrayant par leur charme et
leur culture. Si notre mariage vous fait Argentine, mon amour me
rapproche de la France, et je voudrais voir beaucoup d’unions
comme la nôtre, basées non seulement sur l’amour, mais encore sur
une parfaite franchise et une confiance réciproque. Décidément,
Cupidon a tort de porter un bandeau sur les yeux : Il faut beaucoup
se connaître pour bien s’aimer…
Tout est prêt, je me marie demain ! Marthe, la chère Marthe, émue
et agitée, vient de me montrer la robe que je dois porter pour la
cérémonie, elle la trouve jolie. Tant mieux !… mais j’avoue que je
serais incapable de la décrire ! Nos bagages sont déjà partis au
« Tigre » où nous passerons huit jours seuls, puis Mamita, Georges
et Marthe viendront nous retrouver…
Georges termine ses affaires avec la Compagnie minière qui l’a
engagé, et repartira en Europe quand son nouveau contrat sera
signé, pour revenir dans deux ans.
Mon Dieu ! est-ce bien moi qui suis là, entourée, choyée, aimée ?
Et Pío qui craignait que son pays ne me plaise pas ! Mais comment
peut-on ne pas aimer la terre où on est heureux ? Je me sens la fille
adoptive et chère de cette patrie que mon instinct a choisie, je la
sens forte, grande dans l’avenir, et je me donne à elle avec
confiance, avec abandon… L’amour seul de mon Pío ne m’aurait pas
conquise si l’Argentine n’avait commencé la conquête !

. . . . . . . . . . . . . . . . . . . .

Nous sommes unis pour la vie… Ce matin, j’ai dit adieu à la petite
maison où restent encore mes cousins, je l’ai parcourue tout entière,
le cœur un peu serré… Elle a abrité pendant des mois mon amour
commençant et ma joie renaissante… J’ai pris un petit bouquet de
jasmin dans le patio et je l’ai fixé à ma ceinture, son parfum
persistant se mêlait à l’odeur de l’encens pendant que je m’inclinais
sous la bénédiction de Mgr Morera, cousin de Mamita, qui a marié
Carmen et qui a voulu nous bénir. Pío tremblait d’émotion, et de
grosses larmes coulaient sur mes joues, larmes heureuses cette fois !
Il est dix heures et demie du soir, l’orchestre bourdonne encore à
mes oreilles, car Mamita reçoit en bas, rayonnante. Pío fait charger
les sacs sur la voiture, et je me suis réfugiée dans mon boudoir ;
mon bonheur est encore timide…
Des baisers, des souhaits, et nous voici sur la route qui va au
Tigre…
— Nous ne pourrions pas aller plus loin, m’a dit Pío, la route
praticable aux automobiles s’arrête là…
La voiture roule mollement entre des rangées d’eucalyptus et de
« paraisos » dont les fleurs violettes ont un parfum de lilas et
d’amande, les phares font de grands cercles lumineux et tremblants
dans lesquels tourbillonnent des insectes affolés.
Nous nous taisons, Pío a appuyé ma tête contre son cœur que je
sens battre à grands coups profonds…
Pourvu que ce soit bien loin, le Tigre !…

. . . . . . . . . . . . . . . . . . . .

Hélas ! non, l’approche de l’eau rafraîchit la nuit, nous arrivons…

Une barque nous attend et nous mène dans une petite île ; des
lumières brillent aux fenêtres de la villa toute vêtue de verdure,
j’entends le clapotis de l’eau, et le bruit des rames qui s’affaiblit au
loin… Le fleuve réfléchit les étoiles… La voix de mon mari murmure
tendrement mon nom… Je suis heureuse.

. . . . . . . . . . . . . . . . . . . .
 Le Tigre ! Venise verte, Venise éplorée dans la chevelure bleuâtre
des saules, petites îles pleines de fleurs et de fruits où des maisons
invisibles laissent échapper des rires plus frais que l’eau et l’herbe,
jardins gardés par de hauts peupliers, mystérieux comme des
murailles, élancés comme des prières !…
Oh ! paysage unique et apaisé, comme vous bercez doucement
notre bonheur !
Le grand fleuve Paraná, le Río Paraná, divisé en innombrables
cours d’eau, enserre plusieurs centaines d’îlots fleuris que nous
parcourons, Pío et moi, à l’heure où le soleil descend vers l’horizon…
Je commence à reconnaître les canaux et nous sommes allés
l’autre jour jusqu’au fleuve immense. Notre bateau dansait comme
sur la mer, et frôlait en passant de petites îles flottantes où croît une
fleur bleue, et qui vont jusqu’à l’Océan, emportant parmi des herbes
géantes, des animaux effarés.
Quelquefois, pour passer entre les rives de deux îles, il nous faut
écarter de la main les branches pendantes des saules-pleureurs.
Des poissons brillent une seconde dans l’air, puis replongent en
faisant jaillir l’eau en perles scintillantes… Le soir, des barques, des
canots, de petits yachts sillonnent les principaux « ríos » où leurs
feux se reflètent, on cause d’un bateau à l’autre, des femmes en
toilette de soirée, des hommes en frac, escaladent les escaliers des
embarcadères pour aller à une réunion mondaine, et des musiques
s’alanguissent dans les feuillages humides…
Mais nous fuyons le bruit, notre embarcation s’enfonce parmi le
dédale des îlots, et nous emporte vers les petits canaux cachés sous
les branches entrecroisées où nous sommes seuls, seuls avec notre
amour.

. . . . . . . . . . . . . . . . . . . .

La maison s’est animée, les domestiques vont et viennent, j’ai

mis des fleurs dans toutes les chambres… Mamita et mes cousins
sont arrivés. Sous le prétexte de faire voir les Régates à Georges,
Mamita a abrégé tant qu’elle a pu notre solitude ! La vérité est
qu’elle mourait d’envie de nous rejoindre ; nous lui en voulions un
peu de cette hâte, Pío et moi, mais quand nous avons vu la joie qui
éclairait sa figure lorsqu’elle nous a aperçus, lorsque nous avons
senti ses bras maternels nous serrer contre son cœur, notre légère
mauvaise humeur a fondu en tendresse et nous nous sommes écriés
ensemble :
— Comme vous avez bien fait de venir si vite, Mamita !
En attendant ces fameuses Régates, Georges et Marthe explorent
à leur tour les ríos et les îles ; nous ne nous retrouvons guère qu’à
table, quand quelque excursion un peu lointaine ne nous réunit pas.
Mamita voit ses vieilles amies et ses jeunes parentes, fait des
vêtements pour les pauvres, et avec une entente merveilleuse mène
la maison et les domestiques. Presque toutes les familles avec
lesquelles nous sommes en relations à Buenos-Aires sont au
« Tigre », il y fait presque frais tandis que la ville gémit sous une
chaleur torride, et c’est un va-et-vient continuel de barques et de
canots sur les ríos, tout le jour, et tard dans la nuit.
Notre jardin est plein de roses, les pêches gonflent au soleil leurs
petites joues veloutées, et les fleurs du jasmin plient sous le poids
vibrant des abeilles…
Georges et Mamita sont les meilleurs amis du monde, il la fait rire
aux larmes en lui racontant, avec un accent français prodigieux, des
histoires extraordinaires et lui dépeint Paris, dont le seul nom la
terrifiait, sous des couleurs si attrayantes qu’elle finit par nous dire
un jour :
— Que diriez-vous, mes enfants, si j’allais faire un petit tour en
Europe avec vous ?
Pío n’en croyait pas ses oreilles.
— Oh ! pas tout de suite, a ajouté Mamita, mais, dans deux ans,
par exemple ; nous pourrions aller chercher Marthe et Georges…
Je félicitai Mamita de la bonne idée qu’elle avait de nous suivre
en Europe, et je lui exprimai ma joie de la voir disposée à aimer mon
pays. Elle me caressa doucement les cheveux :
 — Vous me le faites tous aimer, ma fille chérie, Marthe est une
petite femme modèle, et ce grand garçon, elle désigna Georges,
arrive à me faire croire tout ce qu’il veut ! Et c’est un réprouvé,
cependant ! N’a-t-il pas essayé de me faire l’apologie du divorce ?
Georges protesta, l’apologie en question n’était que des
explications données à Mamita sur une coutume qui la choquait si
fort…
Toutes les femmes argentines, et surtout celles de la génération
de notre chère maman, ont en abomination le divorce ; elles ne
l’admettent pas, même dans les cas extrêmes, et se bouchent les
oreilles lorsqu’un audacieux essaie de leur prouver que quelquefois
le divorce est une délivrance et permet à un être de refaire sa vie
compromise par un mariage malheureux. Le sentiment religieux est
pour beaucoup dans cette horreur de la rupture d’une union
consacrée par Dieu, mais il s’y ajoute un sentiment particulier du
devoir maternel, et une soumission presque orientale aux volontés
de l’époux et puis, comment une femme qui a cinq ou six enfants
pourrait-elle quitter le foyer conjugal, et refaire une vie qui
appartient déjà à ces petits êtres ? Il y a de mauvais ménages sans
doute, en Argentine, mais sûrement moins que dans les pays où la
femme est trop mêlée à la vie de son mari et où, par conséquent,
les heurts sont plus fréquents entre deux caractères opposés ou
deux natures différentes. On a proposé plusieurs fois à la Chambre
une loi établissant le divorce, elle n’a jamais pu passer, et il
s’écoulera vraisemblablement de bien longues années avant qu’elle
obtienne un vote favorable… si tant est qu’elle l’obtienne jamais.

. . . . . . . . . . . . . . . . . . . .

Tout le « Tigre » est pavoisé sur le parcours des yachts qui vont
prendre part aux régates. Pas de vent, pas un nuage, la journée est
magnifique, mille embarcations se pressent sur l’eau miroitante et
les courses commencent…
 On encourage de cris sympathiques les concurrents dont les
parents et les amis s’entassent dans des barques ou sur les rives, on
acclame les vainqueurs, on console les vaincus, de belles jeunes
filles trépignent de joie parce que leur frère ou leur fiancé a gagné
un prix… Les oriflammes flottent dans une brise tiède, les robes
claires et les ombrelles se dorent au soleil couchant…
Dans une demi-heure tout sera silencieux ; le bruit des moteurs à
pétrole sera remplacé par le coassement des petites grenouilles qui
se cachent dans les roseaux, et les cris d’enthousiasme par le
murmure de l’eau que caressent les rameaux tombants des saules…
Mais la nature n’a qu’un court répit ; ce calme profond est de
nouveau troublé par une fête de nuit : une fête vénitienne. Des
bateaux illuminés conduits par les beaux jeunes gens qui ont prouvé
tantôt ce que peuvent les sports pour l’embellissement d’une race,
suivent le fil de l’eau, doucement, si nombreux et si pressés que
leurs feux paraissent être le reflet mouvant des étoiles. Dans une île,
un orchestre joue une musique italienne, qui nous parvient légère,
lointaine, faite pour accompagner un rêve. On ne crie plus ce soir, la
beauté de la nuit paralyse les gaietés bruyantes, et c’est de la poésie
et de la douceur qui flottent dans l’air et sur les eaux…

. . . . . . . . . . . . . . . . . . . .

Pío s’est mis en tête de me conduire à Mar-del-Plata. Mamita a

protesté, j’ai résisté, mais mes cousins ont soutenu Pío, et mon cher
tyran, obéi et triomphant, nous emmène tous trois passer une
semaine dans le port de mer à la mode, le Deauville Argentin :
Mamita restera au « Tigre » jusqu’à notre retour.
Sur les conseils de Pío, nous emportons beaucoup de robes, et
tous nos bijoux… Il paraît que c’est la « Grande Semaine » de Mar-
del-Plata, et qu’on y fait assaut d’élégance.
En douze heures de voyage, nous avons atteint le but de notre
voyage. Nos chambres sont retenues, heureusement, car les hôtels
et les villas n’ont plus un coin à offrir au voyageur, fût-ce à prix d’or !
L’argent a peu de valeur à Buenos-Aires, à Mar-del-Plata, il n’en a
plus du tout ! Nous étions, mes cousins et moi, stupéfaits de ce qu’il
fallait payer à l’hôtel pour un modeste appartement et une
nourriture médiocre. Pío riait de notre étonnement, lui qui ne conçoit
pas plus que ses compatriotes ce qu’est l’économie, et encore moins
l’épargne ; il se moque gentiment de nous, lorsque nous essayons de
lui parler de « prévoyance… »
— Prévoyance de quoi ? demande-t-il.
— Mais de l’avenir.
— De mon avenir ? Je fais de bonnes affaires, mes terres me
rapportent dix ou douze pour cent, bon an mal an, et se valorisent
chaque jour… Alors ?
— Les temps de disette peuvent venir !
— Les temps d’abondance peuvent durer !
— Mais si vous avez des enfants ?
— Ils travailleront…
— Des filles ?
— Elles se marieront !
— Et dire que vous êtes tous comme ça ! s’écria Georges
abandonnant la discussion, c’est effrayant !
— Pourquoi effrayant ? Cette insouciance, cette libéralité que vous
blâmez, nous rend audacieux ; à cause d’elle, nous nous lançons
dans les affaires les plus risquées avec une confiance qui les fait
réussir.
— Mais quand un homme meurt sans fortune, que devient sa
veuve ? Qui élève ses enfants ?
— La veuve est toujours recueillie ou aidée par quelqu’un de sa
famille ou de celle de son mari, et jusqu’aux amis du disparu, tous
s’unissent pour donner aux enfants une éducation et une instruction
suffisantes pour marcher à la conquête de la fortune ! Les enfants
élevés ainsi ne sont pas moins bien armés pour la lutte que ceux
dont les parents sont devenus ou restés riches. N’oubliez pas qu’une
de nos forces est cette possibilité donnée à tous, sans distinction de
classe ou de nationalité, d’arriver à la richesse par le travail : notre
terre offre ses trésors à qui possède la force de les prendre, et elle
est si prodigue elle-même que ses fils seraient sans excuses s’ils
osaient être avares…

. . . . . . . . . . . . . . . . . . . .

Pío avait raison : il faut avoir vu Mar-del-Plata pendant sa

semaine élégante, mais je me promets bien tout bas de n’y pas
revenir !… C’est un luxe étourdissant : on se promène sur la Rambla,
on fait les cent pas sur la plage, on déjeune, on joue, on va écouter
de la musique au Club, on dîne, on danse, on fait un bridge, et
chacune de ces actions est précédée d’un changement de toilette…
et quelle toilette ! C’est à se demander si les costumes de bain ne
viennent pas de chez Paquin… car on se baigne de temps en temps,
quand on a un moment de loisir… entre les courses et le thé…
J’oubliais les fiançailles, les brouilles, les raccommodements et les
bavardages qui semblent s’accumuler pendant cette courte saison.
Je suis péniblement ce mouvement effréné, et je vois bien que
mon mari ne reste que pour ne pas abandonner nos cousins. Eux
s’amusent franchement, entraînés par d’autres jeunes couples, et
voient s’écouler les jours avec peine…
Nous avons pourtant pu nous isoler assez souvent, Pío et moi,
pour faire quelques promenades le long de la côte. Partout, des
rochers sauvages, des falaises monstrueuses, un chaos de pierres
battues par des vagues furieuses, démesurées. L’Océan est sans
douceur, ici, il apporte une houle que rien n’a brisée pendant des
milliers de lieues, un vent qui a traversé la moitié du monde sans
perdre sa violence, et les plages, magnifiques, s’arrondissent à perte
de vue… C’est une beauté sévère, grandiose, impressionnante : la
hauteur des falaises écrase, et la vue de l’Atlantique tumultueux
éveille la terreur, et fait naître la nostalgie des mers calmes… La ville
elle-même est somptueuse, mais encore inachevée, on y bâtit
toujours et on rêve d’y faire un port plus facile à atteindre pour les
navires de fort tonnage, que celui de Buenos-Aires, rendu d’accès
difficile par le fond mouvant du Rio. Mar-del-Plata ou plutôt le
« Club » de Mar-del-Plata, a un ravissant petit théâtre, assez peu
fréquenté d’ailleurs, et possède l’inévitable cinématographe où les
familles envoient les enfants, les jours de pluie ou les soirs de vent.
Des estancias sont éparpillées à quelque distance du bord de la
mer ; nous en avons visité une, propriété d’un ami de Pío, et nous
avons constaté non sans étonnement que les produits de cette terre
si soumise au climat marin, dépassent en qualité et en abondance
tout ce que nous avons vu jusqu’à ce jour, et que le bétail y est
splendide.
Des réunions mondaines, des thés, des bridges, la roulette… et le
dernier jour de la Grande Semaine, qui est aussi celle du Carnaval,
s’est achevé par un bal masqué à l’hôtel Windsor, lieu de réunion de
la société élégante ; ce bal réunissait tant de jolies femmes, tant
d’adorables jeunes filles et tant d’hommes beaux et distingués
qu’aucun lieu du monde ne pourrait en grouper davantage.
Malgré cette dernière, cette éblouissante vision, c’est sans
regrets que nous reprenons le train, emmenant Georges et Marthe
qui poussent quelques soupirs, et c’est avec joie que nous
retrouvons notre chère petite île verte du Tigre et l’active tendresse
de Mamita, dont la joie de nous revoir est émouvante.

. . . . . . . . . . . . . . . . . . . .

J’ai dû aller ces jours-ci à Buenos-Aires pour y faire quelques

achats ; la journée que j’y ai passée a été rude. Il y a des semaines
qu’il ne pleut pas, les rues sont poussiéreuses, les maisons sont
fermées et comme barricadées contre le soleil ennemi, les pieds des
chevaux s’enfoncent dans l’asphalte amolli des chaussées, les rares
passants se glissent le long des maisons, cherchant l’ombre, et le
soir, pour trouver un peu de fraîcheur ou plutôt l’illusion de la
fraîcheur, des milliers de gens vont au bois de Palermo où semble se
concentrer toute l’animation de la ville. On attend l’orage, on guette