![tinkle, to proclaim that Herr Roth was coming with a Fable of Gellert,
or a chapter from Vater Pestalozzi's serious novel, Gumal und Lina,
to read, and expound, and catechise upon. This last lesson before
dinner was always accompanied by frequent yawns and other
unrepressed symptoms of fatigue; and at its conclusion we all rose
with a shout, and rushed into the corridors.
On resuming work in the afternoon, there was even less attention
and method observed than before. The classes were then broken up,
and private lessons were given in accomplishments, or in some of
the useful arts. Drawing dogs and cows, with a master to look after
the trees and the hedges; whistling and spitting through a flute;
playing on the patience of a violin; turning at a lathe; or fencing with
a powerful maître d'armes;—such were the general occupations. It
was then, however, that we English withdrew to our Greek and
Latin; and, under a kind master, Dr M——, acquired (with the
exception of a love for natural history, and a very unambitious turn
of mind) all that really could deserve the name of education.
We have now described the sedentary life at the chateau. In the
next paper the reader shall be carried to the gymnasium; the drill
ground behind the lake; to our small menageries of kids, guinea
pigs, and rabbits; be present at our annual ball and skating bouts in
winter, and at our bathings, fishings, frog-spearings, and rambles
over the Jura in summer.
FOOTNOTES:
[14] Cicero, De Fin., ii. 1.](https://image.slidesharecdn.com/1112650-250601081100-a94026f3/85/The-Image-Processing-Handbook-Sixth-Edition-6th-Edition-John-C-Russ-74-320.jpg)
![in the Continent of Hindostan, at the distance of ten thousand miles
from the parent state; they had made themselves felt alike, and at
the same moment, at Nankin, the ancient capital of the Celestial
Empire, and at Cabool, the cradle of Mahommedan power.
Conquering all who resisted, blessing all who submitted, securing
the allegiance of the subjects by the justice and experienced
advantages of their government, they had realised the boasted
maxim of Roman administration—
Parcere subjectis et debellare superbos,
and steadily advanced through a hundred years of effort and glory,
not unmixed with disaster, from the banks of the Hoogley to the
shores of the Indus—from the black hole of Calcutta to the throne of
Aurengzebe.
Nulla magna civitas, said Hannibal, diu quiescere potest—si foris
hostem non habet, domi invenit: ut praevalida corpora ab externis
causis tuta videntur, suis ipsis viribus conficiuntur.[15] When the
Carthaginian hero made this mournful reflection on the infatuated
spirit which had seized his own countrymen, and threatened to
destroy their once powerful dominion, he little thought what a
marvellous confirmation of it a future empire of far greater extent
and celebrity was to afford. That the system of free trade—that is,
the universal preference of foreigners, for the sake of the smallest
reduction of price, to your own subjects—must, if persisted in, lead
to the dismemberment and overthrow of the British empire, cannot
admit of a moment's doubt, and will be amply proved to every
unbiassed reader in the sequel of this paper. Yet the moment chosen
for carrying this principle into effect was precisely that, when the
good effects of the opposite system had been most decisively
demonstrated, and an empire unprecedented in magnitude and
magnificence had reached its acme under its shadow. It would be
impossible to explain so strange an anomaly, if we did not recollect
how wayward and irreconcilable are the changes of the human
mind: that action and reaction is the law not less of the moral than](https://image.slidesharecdn.com/1112650-250601081100-a94026f3/85/The-Image-Processing-Handbook-Sixth-Edition-6th-Edition-John-C-Russ-77-320.jpg)
The Image Processing Handbook Sixth Edition 6th Edition John C Russ
- 1. The Image Processing Handbook Sixth Edition 6th Edition John C Russ download https://ebookbell.com/product/the-image-processing-handbook- sixth-edition-6th-edition-john-c-russ-2225300 Explore and download more ebooks at ebookbell.com
- 2. Here are some recommended products that we believe you will be interested in. You can click the link to download. The Image Processing Handbook 7th Edition Russ John Cneal F Brent https://ebookbell.com/product/the-image-processing-handbook-7th- edition-russ-john-cneal-f-brent-22131752 The Image Processing Handbook Fourth Edition 4th Edition John C Russ https://ebookbell.com/product/the-image-processing-handbook-fourth- edition-4th-edition-john-c-russ-2325970 The Image Processing Handbook 6ed Russ Jc https://ebookbell.com/product/the-image-processing-handbook-6ed-russ- jc-2598084 The Image Processing Handbook 7th Edition John C Russ F Brent Neal https://ebookbell.com/product/the-image-processing-handbook-7th- edition-john-c-russ-f-brent-neal-5245774
- 3. The Image Processing Handbook 5th Edition John C Russ https://ebookbell.com/product/the-image-processing-handbook-5th- edition-john-c-russ-981882 The Handbook Of Astronomical Image Processing Richard Berry James Burnell https://ebookbell.com/product/the-handbook-of-astronomical-image- processing-richard-berry-james-burnell-52165134 The Electrical Engineering Handbook Circuits Signals Speech And Image Processing Richard C Dorf https://ebookbell.com/product/the-electrical-engineering-handbook- circuits-signals-speech-and-image-processing-richard-c-dorf-4719780 The Image Processing Cookbook 4th Edition John C Russ https://ebookbell.com/product/the-image-processing-cookbook-4th- edition-john-c-russ-5848668 The Art Of Image Processing Digital Camera Processing 1st Edition Chris Bore https://ebookbell.com/product/the-art-of-image-processing-digital- camera-processing-1st-edition-chris-bore-46665556
- 5. IMAGE PROCESSING Handbook The Sixth Edition John C. Russ IMAGE PROCESSING
- 8. IMAGE PROCESSING Handbook The Sixth Edition John C. Russ North Carolina State University Materials Science and Engineering Department Raleigh, North Carolina CRC Press is an imprint of the Taylor & Francis Group, an informa business Boca Raton London New York
- 9. CRC Press Taylor & Francis Group 6000 Broken Sound Parkway NW, Suite 300 Boca Raton, FL 33487-2742 © 2011 by Taylor and Francis Group, LLC CRC Press is an imprint of Taylor & Francis Group, an Informa business No claim to original U.S. Government works Printed in the United States of America on acid-free paper 10 9 8 7 6 5 4 3 2 1 International Standard Book Number-13: 978-1-4398-4063-4 (Ebook-PDF) This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been made to publish reliable data and information, but the author and publisher cannot assume responsibility for the valid- ity of all materials or the consequences of their use. The authors and publishers have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if permission to publish in this form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may rectify in any future reprint. Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or uti- lized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopy- ing, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers. For permission to photocopy or use material electronically from this work, please access www.copyright.com (http:// www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com and the CRC Press Web site at http://www.crcpress.com
- 10. v © 2011 by Taylor & Francis Group, LLC Contents Acknowledgments. . . . . . . . . . . . . . . . . . . xi Introduction . . . . . . . . . . . . . . . . . . . . . . xiii Author . . . . . . . . . . . . . . . . . . . . . . . . . xvii 1 Acquiring Images . . . . . . . . . . . . . . . . 1 Human reliance on images for information............ 1 Video cameras. .................................................. 6 CCD cameras.................................................... 7 Camera artifacts and limitations..........................13 Color cameras..................................................15 Camera resolution.............................................18 Focusing. ......................................................... 20 Electronics and bandwidth limitations. ..................21 Pixels...............................................................24 Gray scale resolution........................................ 26 Noise............................................................. 28 High depth images........................................... 30 Color imaging. ..................................................34 Digital camera limitations.................................. 42 Color spaces................................................... 42 Color correction................................................52 Color displays. ..................................................54 Image types. .................................................... 56 Range imaging................................................ 58 Multiple images............................................... 64 Stereoscopy. .................................................... 69 Imaging requirements. ....................................... 77
- 11. vi Contents © 2011 by Taylor & Francis Group, LLC 2 Human Vision. . . . . . . . . . . . . . . . . . 85 What we see and why...................................... 85 Recognition..................................................... 88 Technical specs................................................ 92 Acuity..............................................................97 What the eye tells the brain..............................101 Spatial comparisons........................................103 Local to global hierarchies................................107 It’s about time................................................. 113 The third dimension. ......................................... 118 How versus what............................................. 121 Seeing what isn’t there, and vice versa...............122 Image compression. .........................................125 A world of light...............................................126 Size matters....................................................129 Shape (whatever that means)............................132 Context..........................................................133 Arrangements must be made. ............................135 Seeing is believing..........................................137 So in conclusion..............................................139 3 Printing and Storage. . . . . . . . . . . . 141 Printing.......................................................... 141 Dots on paper. ................................................145 Color printing.................................................150 Printing hardware............................................156 Film recorders.................................................161 Other presentation tools...................................162 File storage. ....................................................163 Storage media................................................164 Magnetic recording.........................................166 Databases for images......................................167 Browsing and thumbnails. ................................. 174 Lossless coding. ...............................................178 Reduced color palettes. ....................................183 JPEG compression...........................................184 Wavelet compression. ......................................187 Fractal compression.........................................192 Digital movies.................................................194 4 Correcting Imaging Defects. . . . . . . 199 Contrast expansion......................................... 200 Noisy images................................................ 205 Neighborhood averaging. ............................... 208 Neighborhood ranking.................................... 214 Other neighborhood noise reduction methods. ....226
- 12. vii Contents © 2011 by Taylor & Francis Group, LLC Defect removal, maximum entropy, and maximum likelihood.........................................232 Nonuniform illumination...................................235 Fitting a background function............................238 Rank leveling..................................................244 Color images..................................................248 Non-planar views. ...........................................250 Computer graphics..........................................252 Geometric distortion........................................254 Alignment. ......................................................256 Interpolation...................................................261 Morphing.......................................................265 5 Image Enhancement in the Spatial Domain. . . . . . . . . . . . . . . . . . . . . . 269 Contrast manipulation......................................270 Histogram equalization....................................274 Local equalization...........................................279 Laplacian.......................................................283 Derivatives. .....................................................293 Finding edges with gradients............................296 More edge detectors...................................... 306 Texture...........................................................312 Fractal analysis............................................... 317 Implementation notes.......................................319 Image math....................................................319 Subtracting images..........................................320 Multiplication and division. ...............................323 Principal components analysis...........................325 Other image combinations...............................331 6 Processing Images in Frequency Space. . . . . . . . . . . . . . . . . . . . . . . 337 About frequency space....................................337 The Fourier transform.......................................338 Fourier transforms of simple functions.................341 Frequencies and orientations. ............................345 Preferred orientation........................................350 Texture and fractals.........................................351 Isolating periodic noise....................................356 Selective masks and filters................................361 Selection of periodic information.......................364 Convolution....................................................370 Deconvolution.................................................372 Noise and Wiener deconvolution......................378 Template matching and correlation....................385 Autocorrelation...............................................391
- 13. viii Contents © 2011 by Taylor & Francis Group, LLC 7 Segmentation and Thresholding. . . .395 Thresholding...................................................395 Automatic settings...........................................398 Multiband images.......................................... 403 Two-dimensional thresholds. ............................. 405 Multiband thresholding................................... 408 Thresholding from texture. ................................. 411 Multiple thresholding criteria. ............................ 414 Textural orientation..........................................415 Region boundaries..........................................419 Conditional histograms. ....................................426 Boundary lines................................................427 Contours....................................................... 430 Image representation.......................................432 Other segmentation methods............................436 The general classification problem.................... 440 8 Processing Binary Images. . . . . . . . 443 Boolean operations........................................ 443 Combining Boolean operations........................ 446 Masks. .......................................................... 450 From pixels to features.....................................452 Boolean logic with features. ..............................457 Selecting features by location...........................461 Double thresholding. ....................................... 466 Erosion and dilation. ....................................... 468 Opening and closing.......................................471 Isotropy. .........................................................473 Measurements using erosion and dilation...........478 Extension to gray scale images.........................481 Morphology neighborhood parameters............. 482 Examples of use..............................................484 Euclidean distance map.................................. 488 Watershed segmentation. .................................491 Ultimate eroded points.....................................494 Skeletons. .......................................................498 Boundary lines and thickening......................... 503 Combining skeleton and EDM. ......................... 506 9 Global Image Measurements. . . . . . 511 Global measurements and stereology................ 511 Surface area. ..................................................516 ASTM Grain Size. ............................................521 Multiple types of surfaces.................................523 Length. ...........................................................525 Thickness. .......................................................527 Sampling strategies.........................................530
- 14. ix Contents © 2011 by Taylor & Francis Group, LLC Determining number........................................532 Curvature, connectivity, and the Disector............535 Anisotropy and gradients.................................538 Size distributions.............................................542 Classical stereology (unfolding).........................543 10 Feature-Specific Measurements. . . . 547 Brightness measurements..................................547 Determining location. .......................................556 Orientation. ....................................................559 Neighbor relationships. ....................................562 Alignment. ......................................................567 Counting........................................................574 Special counting procedures.............................579 Feature size....................................................584 Circles and ellipses. .........................................587 Caliper dimensions..........................................589 Perimeter........................................................592 11 Characterizing Shape . . . . . . . . . . . 597 Describing shape............................................597 Dimensionless ratios. ........................................599 Fractal dimension........................................... 604 Harmonic analysis...........................................610 Topology. .......................................................620 Three dimensions. ............................................623 12 Feature Recognition and Classification. . . . . . . . . . . . . . . . . . 627 Template matching and cross-correlation............628 Parametric description. .....................................631 Decision points...............................................635 Multidimensional classification..........................639 Learning systems. .............................................646 kNN and cluster analysis.................................652 Expert systems. ................................................655 Neural nets. ....................................................657 Syntactical models. ..........................................659 13 Tomographic Imaging. . . . . . . . . . . 661 More than two dimensions. ...............................661 Volume imaging vs. sections.............................664 Basics of reconstruction....................................670 Algebraic reconstruction methods......................676 Maximum entropy...........................................679 Defects in reconstructed images........................681
- 15. x Contents © 2011 by Taylor & Francis Group, LLC Beam hardening. ............................................ 686 Imaging geometries.........................................691 Three-dimensional tomography. .........................695 High-resolution tomography..............................701 14 3D Visualization . . . . . . . . . . . . . . . 707 Sources of 3D data. .........................................707 Serial sections. ................................................709 Optical sectioning...........................................713 Sequential removal..........................................715 Stereo measurement........................................ 717 3D data sets...................................................722 Slicing the data set..........................................724 Arbitrary section planes...................................727 The use of color..............................................731 Volumetric display...........................................732 Stereo viewing................................................736 Special display hardware.................................739 Ray tracing..................................................... 741 Reflection.......................................................746 Surfaces.........................................................750 Multiply connected surfaces. .............................754 Image processing in 3D...................................759 Measurements on 3D images. ...........................763 15 Imaging Surfaces . . . . . . . . . . . . . . 767 Producing surfaces. ..........................................767 Imaging surfaces by physical contact.................770 Noncontacting measurements...........................773 Microscopy of surfaces....................................777 Surface composition imaging............................782 Processing of range images..............................783 Processing of composition maps........................787 Data presentation and visualization...................788 Rendering and visualization..............................791 Analysis of surface data...................................796 Profile measurements. ...................................... 800 The Birmingham measurement suite. .................. 803 Topographic analysis and fractal dimensions..... 809 References . . . . . . . . . . . . . . . . . . . . . . 817
- 16. xi © 2011 by Taylor & Francis Group, LLC Acknowledgments All of the image processing and the creation of the resulting figures included in this book were performed on an Apple Macintosh® and/or a Sony VAIO® computer, using Adobe Photoshop® CS4 with the Fovea Pro plug-ins. Many of the images were acquired directly from various microscopes and other sources that provided digital output directly to the computer. Others were captured using a variety of digital cameras (Sony, Nikon, Canon, and others), and some were obtained using flat-bed and slide scanners (Nikon and Epson), often from images supplied by colleagues and researchers. These are acknowl- edged wherever the origin of an image could be determined. A few examples, taken from the literature, are individually referenced. The book was delivered to the publisher in digital form (on a writable DVD), without intermediate hard copy, negatives or prints of the images, etc. Among other things, this means that the author must bear full responsibility for typographical errors or problems with the figures. Every effort has been made to show enlarged image frag- ments that will reveal pixel-level detail when it is important. The process has also forced me to learn more than I ever hoped to know about some aspects of publish- ing technology! However, going directly from disk file to print also shortens the time needed in production and helps to keep costs down, while preserving the full quality of the images. Grateful acknowledgment is made of the efforts by the excellent edi- tors at CRC Press to educate me and to accommodate the unusually large number of illustrations in this book (more than 2000 figures and more than a quarter of a mil- lion words). Special thanks are due to Chris Russ (Reindeer Graphics Inc., Asheville, NC) who has helped to program many of these algorithms and contributed invaluable comments, and especially to Helen Adams, who has proofread many pages, endured many discussions about ways to present information effectively, and provided the support (and the occa- sional glass of wine) that make writing projects such as this possible. John C. Russ Raleigh, NC
- 18. xiii © 2011 by Taylor & Francis Group, LLC Introduction Image processing is used in a wide variety of applications, for two somewhat differ- ent purposes: 1. improving the visual appearance of images to a human observer, including their printing and transmission, and 2. preparing images for the measurement of the features and structures which they reveal. The techniques that are appropriate for each of these tasks are not always the same, but there is considerable overlap. This book covers methods that are used for both tasks. To do the best possible job, it is important to know about the uses to which the pro- cessed images will be put. For visual enhancement, this means having some familiarity with the human visual process and an appreciation of what cues the viewer responds to in images. A chapter on human vision addresses those issues. It also is useful to know about the printing or display process, since many images are processed in the context of reproduction or transmission. Printing technology for images has advanced significantly with the consumer impact of digital cameras, and up-to-date information is provided. The measurement of images is often a principal method for acquiring scientific data and generally requires that features or structure be well defined, either by edges or unique brightness, color, texture, or some combination of these factors. The types of measure- ments that can be performed on entire scenes or on individual features are important in determining the appropriate processing steps. Several chapters deal with measurement in detail. Measurements of size, position, and brightness deal with topics that humans largely understand, although human vision is not quantitative and is easily fooled. Shape is a more difficult concept, and a separate chapter added in this edition summarizes a variety of ways that shape may be described by numbers. Measurement data may be used for classification or recognition of objects. There are several different strategies that can be applied, and examples are shown. It may help to recall that image processing, like food processing or word processing, does not reduce the amount of data present but simply rearranges it. Some arrangements
- 19. xiv Introduction © 2011 by Taylor & Francis Group, LLC may be more appealing to the senses, and some may convey more meaning, but these two criteria may not be identical nor call for identical methods. This handbook presents an extensive collection of image processing tools, so that the user of computer-based systems can both understand those methods provided in pack- aged software and program those additions which may be needed for particular applica- tions. Comparisons are presented for different algorithms that may be used for similar purposes, using a selection of representative pictures from various microscopy tech- niques, as well as macroscopic, remote sensing, and astronomical images. It is very important to emphasize that the scale of the image matters very little to the techniques used to process or analyze it. Microscopes that have a resolution of nm and telescopes that produce images covering light years produce images that require many of the same algorithms. The emphasis throughout the book continues to be on explaining and illustrating meth- ods so that they can be clearly understood, rather than providing dense mathematics. With the advances in computer speed and power, tricks and approximations in search of efficiency are less important, so that examples based on exact implementation of methods with full precision can generally be implemented on desktop systems. The top- ics covered are generally presented in the same order in which the methods would be applied in a typical workflow. For many years, in teaching this material to students I have described achieving mastery of these techniques as being much like becoming a skilled journeyman carpenter. The num- ber of distinct woodworking tools — saws, planes, drills, etc. — is relatively small, and although there are some variations — slotted vs. Phillips-head screwdrivers, for example — knowing how to use each type of tool is closely linked to understanding what it does. With a set of these tools, the skilled carpenter can produce a house, a boat, or a piece of furniture. So it is with image processing tools, which are conveniently grouped into only a few classes, such as histogram modification, neighborhood operations, Fourier-space processing, and so on, and can be used to accomplish a broad range of purposes. Visiting your local hardware store and purchasing the appropriate tools do not provide the skills to use them. Understanding their use requires practice, which develops the ability to visu- alize beforehand what each will do. The same is true of the tools for image processing. In revising the book for this new edition, I have again tried to respond to some of the comments and requests of readers and reviewers. New chapters on the measurement of images and the subsequent interpretation of the data were added in the second edi- tion, and a section on surface images in the third. The fourth edition added the stereo- logical interpretation of measurements on sections through three-dimensional structures and the various logical approaches to feature classification. The fifth edition brought expanded sections on deconvolution, extended dynamic range images, and multichannel imaging, including principal components analysis. In this sixth edition, a new chapter on the meaning of shape has been added, as well as additional material on imaging in more than two dimensions. The sections on the ever-advancing hardware for image capture and printing have been expanded and information added on the newest hardware and software technologies. As in past editions, I have resisted suggestions to put “more of the math” into the book. There are excellent texts on image processing, compression, mathematical morphol- ogy, etc., that provide as much rigor and as many derivations as may be needed. Many of them are referenced here. But the thrust of this book remains teaching by example.
- 20. xv Introduction © 2011 by Taylor & Francis Group, LLC Few people learn the principles of image processing from the equations. Just as we use images to communicate ideas and to “do science,” so most of us use images to learn about many things, including imaging itself. The hope is that by seeing and comparing what various operations do to representative images, you will discover how and why to use them. Then, if you need to look up the mathematical foundations, they will be easier to understand. A very real concern for everyone involved in imaging, particularly in scientific and foren- sic fields, is the question of what constitutes proper and appropriate processing and what constitutes unethical or even fraudulent manipulation. The short answer is that anything that alters an image so as to create a false impression on the part of the viewer is wrong. The problem with that answer is that it does not take into account the fact that different viewers will tend to see different things in the image anyway, and that what constitutes a false impression for one person may not for another. The first rule is always to store a permanent copy of the original image along with rel- evant data on its acquisition. The second rule is to carefully document whatever steps are taken to process the image and generally to report those steps when the processed image is published. Most scientific publications and the editors who review submitted papers have become more aware in recent years of the ease with which image process- ing can be performed and the dangers of inadequate documentation. For example, see M. Rossner and K. M. Yamada (2004; J. Cell Biology) for that journal’s policy on image ethics and examples of improper manipulation. For forensic purposes, there is an additional responsibility to fully record the entire step- by-step procedures that are used and to make sure that those methods are acceptable in court according to the U.S. Supreme Court’s Daubert ruling (Daubert v. Merrell Dow Pharmaceuticals (92-102), 509 U.S. 579, 1993), which generally means that not only are the methods widely accepted by professionals, but also that they have been rigorously tested and have known performance outcomes. In a forensic setting, there will often be a need to explain a procedure, step by step, to a non-technical jury. This frequently requires showing that the details obtained from the image are really present in the origi- nal but only became visually evident with the processing. Some procedures, such as rearranging features or combining them within a single image, or differently adjusting the contrast of several images to make them appear more alike, are clearly misleading and generally wrong. Some, such as using copy-and-paste to dupli- cate a portion of an image, or selectively erasing portions of an image, are out-and-out fraudulent. Even selective cropping of an image (or choosing which field of view to record) can create a false impression. The general guideline to be considered is that it is never acceptable to add anything to an image, but it may be acceptable to suppress or remove some information if it makes the remaining details more accessible, either visually for presentation and communication or to facilitate measurement. Of course, the procedures used must be documented and reported. Any of the procedures shown here may be appropriate in a particular instance. But they can also be misused and should in any case never be used without understand- ing and careful documentation. The heart of the scientific method is replicability. If adequate information is provided on the processing steps applied and the original image data are preserved, then the validity of the results can be independently verified. An important but often overlooked concern is the need to avoid using programs that alter the image without the user being aware of it. For example, carefully correcting the
- 21. xvi Introduction © 2011 by Taylor & Francis Group, LLC colors in an image using Photoshop® and then placing it in PowerPoint® for presentation will cause changes even on the same computer screen (as well as discarding pixels and reducing resolution if copy-and-paste is used for the transfer). In addition, the image may appear different on another computer monitor or when using a projector. Pasting an image into Microsoft® Word will reduce the resolution and color or gray scale dynamic range. This may not affect the printed document, which has less gamut than the com- puter screen anyway, but the image cannot be subsequently retrieved from the document in its original form. Saving an image with a lossy compression method such as jpeg will discard potentially important information that cannot be recovered. The reader is encouraged to use this book in concert with a real source of images and a computer-based system and to freely experiment with different methods to deter- mine which are most appropriate for his or her particular needs. Selection of image processing tools to explore images when you don’t know the contents beforehand is a much more difficult task than using tools to make it easier for another viewer or a measurement program to see the same things you have discovered. It places greater demand on computing speed and the interactive nature of the interface. But it particu- larly requires that you become a very analytical observer of images. If you can learn to see what the computer sees and predict what various algorithms will do, you will become a better viewer and obtain the best possible images, suitable for further pro- cessing and analysis. To facilitate this hands-on learning, I have collaborated with my son, Chris Russ, to write a companion book, Introduction to Image Processing and Analysis, which teaches how to program these algorithms and create Adobe Photoshop compatible plug-ins that implement the methods. The downloadable solutions to the book’s worked problems can be used to apply the routines on either Macintosh or Windows computers. There are additional links to downloadable plug-ins and trial program packages on my Web site at http://www.DrJohnRuss.com.
- 22. xvii © 2011 by Taylor & Francis Group, LLC Author In his fifty-year career as scientist and educator, John Russ has used image processing and analysis as a principal tool for understanding and characterizing the structure and function of materials. Images from a wide variety of devices — including light and electron microscopes, x-ray and neutron tomography, and more — require computer processing and measurement to extract the important data. Much of Russ’ research work has been concerned with the micro- structure and surface topography of metals and ceramics. He has received funding for his research and teaching from government agencies and from industry. Although retired, Dr. Russ is currently assisting in the establish- ment of a new laboratory and program at North Carolina State University, which will be the first in the nation to offer advanced degrees in Forensic Science and Engineering. Familiarity with the algorithms and instruments led to Dr. Russ’ expertise being extended to a much broader range of images — from astronomy to biomedical research to food science to forensics. In addition to students in NCSU’s College of Engineering, Russ has been on graduate student commitees and collaborated with faculty in textiles, pulp and paper products, veterinary medicine, microbiology, food science, and archaeology, among others. Teaching the principles and methods involved to several thousand stu- dents and consulting for many industrial clients have further broadened Dr. Russ’ experi- ence and the scope of applications for image processing and analysis. After retirement, Dr. Russ was Research Director for Rank Taylor Hobson, a manu- facturer of precision instrumentation. He continues to write, to consult for a variety of companies (and to provide expert testimony in criminal and civil cases), to teach
- 23. xviii Author © 2011 by Taylor & Francis Group, LLC workshops worldwide on image processing and analysis, and to review publications and funding proposals. He is active in the Microscopy Society of America, the Microbeam Analysis Society, the Society of Photo-Optical Instrumentation Engineering (SPIE), the International Society for Stereology, is a board member of the Society for Quantitative Morphology, and a Fellow of the Royal Microscopical Society, and has presented invited lectures and workshops for these and other organizations. On November 16, 2006, the New York Microscopical Society awarded John Russ the Ernst Abbe Memorial Award for his contributions to the field of microscopy as a developer of computer-assisted micros- copy and image analysis.
- 24. 1 © 2011 by Taylor & Francis Group, LLC 1 Acquiring Images Human reliance on images for information H umans are primarily visual creatures. Not all animals depend on their eyes, as we do, for most of the information received about their surroundings (the characteristics of human vision are discussed in Chapter 2). This bias in everyday life extends to how we pursue more technical goals as well. Scientific instruments commonly produce images to communicate their results to the operator, rather than generating audible tones or emit- ting a smell. Space missions to other planets and equally arduous explorations of the ocean depths always include cameras as major components, and the success of those missions is often judged by the quality of the images returned. This suggests a few of the ways in which humans have extended the range of natural vision. Optical devices such as microscopes and telescopes allow us to see things that are vastly smaller or larger than we could otherwise. Beyond the visible portion of the electromagnetic spectrum (a narrow range of wavelengths between about 400 and 700 nanometers) there are sensors capable of detecting infrared and ultraviolet light, X-rays, and radio waves, and perhaps soon even gravity waves. Figure 1.1 shows an example, an image presenting radio telescope data in the form of an image in which color represents the Doppler shift in the radio signal. Such devices and presentations are used to further extend imaging capability. Signals other than electromagnetic radiation can be used to produce images, too. Novel new types of microscopes that use atomic-scale probes to “feel” the specimen surface present their data as images (Figure 1.2). The data so collected may represent the surface elevation and topography, but other signals, such as the lateral drag force on the probe, may also be used. Acoustic waves at low frequency produce sonar images, while at gigahertz frequencies the acoustic microscope produces images with resolution similar to that of the light microscope, but with image contrast that is produced by local variations in the attenuation and refraction of sound waves rather than light. Figure 1.3 shows an acoustic microscope image of a sub- surface defect, and Figure 1.4 shows a sonogram of a baby in the womb. Some images such as holograms or electron diffraction patterns record brightness as a func- tion of position, but are unfamiliar to the observer. Figure 1.5 shows an image of an electron diffraction pattern from a transmission electron microscope, in which the atomic structure of the samples is revealed (but only by measurement and to those who know how to interpret
- 25. 2 The Image Processing Handbook, Sixth Edition © 2011 by Taylor & Francis Group, LLC the data). Other kinds of data, including weather maps with specialized symbols, graphs of business profit and expenses, and charts with axes representing time, family income, choles- terol level, or even more obscure parameters, have become part of daily life, as illustrated in Figure 1.6. The latest developments in computer interfaces and displays make extensive use of graphics, to take advantage of the large bandwidth of the human visual pathway. Tufte (1990, 1997, 2001) in particular has demonstrated the power of appropriate graphics to com- municate complex information. There are some important differences between human vision, the kind of information it extracts from images, and the ways in which it seems to do so, as compared to the use of imaging devices based on computers for scientific, technical, or forensic purposes. Humans Figure 1.1 Radio astronomy pro- duces images such as this view of Messier 33 (generated with data from telescopes of the National Radio Astronomy Observatory, a National Science Foundation Facility managed by Associated Universities, Inc.). These are often displayed with false colors to emphasize subtle variations in signal strength or - as in this example - Doppler shift. Figure 1.2 Atomic force micro- scope image of human chromo- somes (courtesy S, Thalhammer, F. Jamitzky, Helmholtz Zentrum München, Germany).
- 26. 3 Acquiring Images © 2011 by Taylor & Francis Group, LLC are especially poor at judging color or brightness of objects and features within images unless they can be exactly compared by making them adjacent. Human vision is inherently com- parative rather than quantitative, responding to the relative size, angle, or position of several objects but unable to supply numeric measures unless one of the reference objects is a mea- suring scale. Overington (1976; 1992) disagrees with this widely accepted and documented conclusion but presents no compelling counter evidence. Chapter 2 illustrates some of the consequences of the characteristics of human vision as they affect what is perceived. This book’s purpose is not to study the human visual pathway directly, but the overview in Chapter 2 can help the reader to understand how humans see things so that we become bet- ter observers. Computer-based image processing and analysis use algorithms based on human vision methods in some cases, but also employ other methods that seem not to have direct counterparts in human vision. In particular, some image processing methods are based on the physics of the image formation and detection process (Sharma, 2005). Many of the examples and much of the analysis presented in this text involve images from various types of microscopes. The three classes of imaging applications that generally offer (a) (b) Figure 1.3 Acoustic microscope image of voids in solder bond beneath a GaAs die: (a) die surface; (b) acoustic image showing strong signal reflections (white areas) from the surface of the voids (courtesy J. E. Semmens, Sonoscan Inc).
- 27. 4 The Image Processing Handbook, Sixth Edition © 2011 by Taylor & Francis Group, LLC Figure 1.4 Surface reconstruction of sonogram imaging, showing a 26 week old fetus in the womb. Figure 1.5 A convergent beam electron diffraction (CBED) pattern from an oxide microcrystal, which can be indexed and measured to provide high accuracy values for the atomic unit cell dimensions.
- 28. 5 Acquiring Images © 2011 by Taylor & Francis Group, LLC the most straightforward types of images are microscopy, aerial (and satellite) imagery, and industrial quality control. That is because in those situations there is the greatest knowledge and/or control over the imaging geometry and the illumination of the scene. In more general “real world” cases the analysis and interpretation of the image contents can be much more difficult. Objects may lie at various distances from the camera, which complicates determin- ing size, may have different lighting, which alters their color, and may even partially obscure other objects. Crime scene and accident photographs are often taken under difficult condi- tions, from less than optimum points of view, and with variable lighting, so that their analysis can be challenging. The basic techniques for image processing and measurement are much the same for images regardless of their source or scale. Images ranging from microscopy to astronomy, images formed with light photons or sound waves, magnetic resonance or scanning profilometers, have much in common and the techniques for dealing with their imperfections, enhancing and extracting the details, and performing measurements utilize the same algorithms and techniques, which are set out in the following chapters. The interpretation of the measure- ments, as presented in later chapters, does require some specialization for different viewing geometries, but is fundamentally independent of magnification. Figure 1.6 Typical graphics used to communicate news information include one-dimensional plots such as stock market reports and two-dimensional presentations such as weather maps.
- 29. 6 The Image Processing Handbook, Sixth Edition © 2011 by Taylor & Francis Group, LLC Video cameras When the first edition of this book was published in 1990, the most common and affordable way of acquiring images for computer processing was with a video camera. Mounted onto a microscope or copystand, in a satellite or space probe, or using appropriate optics to view an experiment, the camera sent an analog signal to a separate “frame grabber” or analog-to- digital converter (ADC) interface board in the computer, which then stored numeric values in memory (Inoué, 1986; Inoué & Spring, 1997). The basic form of the original type of video camera is the vidicon, illustrated in Figure 1.7. It functions by scanning a focused beam of electrons across a phosphor coating applied to the inside of an evacuated glass tube. The light enters the camera through the front glass surface (and a thin metallic anode layer) and creates free electrons in the phosphor. These vary the local conductivity of the layer, so the amount of current that flows to the anode varies as the beam is scanned, according to the local light intensity. This analog (continuously varying) electrical signal is amplified and, as shown in Figure 1.8, conforms to standards of voltage and timing (the standards and timing are slightly different in Europe than the US, but the basic principles remain the same). Digitizing the voltage is accomplished by sampling it and generating a comparison voltage. The child’s game of “guess a number” illustrates that it takes only eight guesses to arrive at a 52 sec. picture width 63.5 sec. horizontal scan interval 0.7 volt range 0.3 volt sync pulse Figure 1.8 Standard RS-170 video signal shows the brightness variation along one scan line (ranging between 0 volts = black and 0.7 volts = white). Anode Phosphor Coating Electron Beam Grid Cathode Deflection and Focusing Coils Glass Tube Figure 1.7 Functional diagram of a vidicon tube. Light striking the phosphor coating changes its local resistance and hence the current that flows as the electron beam scans in a raster pattern.
- 30. 7 Acquiring Images © 2011 by Taylor & Francis Group, LLC value that defines the voltage to one part in 256 (the most widely used type of ADC). The first guess is 128, or half the voltage range. If this is (e.g.) too large, the second guess subtracts 64. Each successive approximation adds or subtracts a value half as large as the previous. In eight steps, the final (smallest) adjustment is made. The result is a number that is conveniently stored in the 8-bit memory of most modern computers. The tube-type camera has several advantages and quite a few drawbacks. Scanning the beam with electromagnetic or electrostatic fields can produce a distorted scan (pincushion or barrel distortion, or more complicated situations) and is subject to degradation by stray fields from wiring or instrumentation. Figure 1.9 shows an example of pincushion distortion, as well as vignetting and loss of focus. Maintaining focus in the corners of the image takes special circuitry, and the corners may also be darkened (vignetting) by the reduction in effective lens aperture and the additional thickness of glass through which the light must pass. The sealed vacuum systems tend to deteriorate with time, and the “getter” used to adsorb gas molecules may flake and fall onto the phosphor if the camera is used in a vertical orientation. The response of the camera (voltage vs. brightness) approximates the logarithmic response of film and the human eye, but this varies for bright and dark scenes. Recovery from bright scenes and bright spots is slow, and blooming can occur in which bright light produces spots that spread laterally in the coating and appear larger than the features really are, with “comet tails” in the scan direction. There are, however, some advantages of the tube-type camera. The spatial resolution is very high, limited only by the grain size of the phosphor and the size of the focused beam spot. Also, the phosphor has a spectral response that can be made similar to that of the human eye, which sees color from red (about 0.7 µm wavelength) to blue (about 0.4 µm). Adaptations of the basic camera design with intermediate cathode layers or special coatings for intensification are capable of acquiring images in very dim light (e.g., night scenes, fluorescence microscopy). CCD cameras The tube-type camera has now been largely supplanted by the solid-state chip camera, the oldest and simplest form of which is the CCD (charge-coupled device). The camera chip con- tains an array of diodes that function as light buckets. Light entering the semiconductor raises electrons from the valence to the conduction band, so the number of electrons is a direct lin- ear measure of the light intensity. The diodes are formed by photolithography, so they have a Figure 1.9 Example of an image show- ing pincushion distortion, as well as loss of focus and vignetting in the edges and corners.
- 31. 8 The Image Processing Handbook, Sixth Edition © 2011 by Taylor & Francis Group, LLC perfectly regular pattern with no image distortion or sensitivity to the presence of stray fields. The devices are also inexpensive and rugged compared to tube cameras. CCDs were first invented and patented at Bell Labs in 1969 (George Smith was awarded the 2009 Nobel Prize in Physics for this invention), and have now largely displaced film in consumer and profes- sional still and movie cameras. The basic operation of a CCD is illustrated in Figure 1.10. Each bucket represents one “pixel” in the camera (this word has a lot of different meanings in different contexts, as explained below, so it must be used with some care). With anywhere from a few hundred thousand to several million detectors on the chip, it is impractical to run wires directly to each one in order to read out the signal. Instead, the electrons that accumulate in each bucket due to incident photons are transferred, one line at a time, to a readout row. On a clock signal, each column of pixels shifts the charge by one location. This places the contents of the buckets into the readout row, and that row is then shifted, one pixel at a time but much more rapidly, to dump the electrons into an amplifier, which produces an analog voltage signal that may be sent out directly or measured to produce the numeric output from a digital camera. The simplest way of shifting the electrons is shown in Figure 1.11. Every set of three elec- trodes on the surface of the device constitutes one pixel. By applying a voltage to two of the electrodes, a field is set up in the semiconductor that acts like a bucket. Electrons are trapped in the central region by the high fields on either side. Note that this does not reduce the area sensitive to incoming photons, because electrons generated in the high field regions quickly migrate to the low field bucket where they are held. By changing the voltage applied to the regions in six steps or phases, as shown in the figure, the electrons are shifted by one pixel. First one field region is lowered and the electrons spread into the larger volume. Then the field on the other side is raised, and the electrons have been shifted by one-third of the pixel height. Repeating the process acts like a conveyor belt and is the reason for the name “charge- coupled device.” Figure 1.10 The basic principle of CCD operation, illustrated as a set of buckets and conveyors (after Janesick, 2001).
- 32. 9 Acquiring Images © 2011 by Taylor & Francis Group, LLC One significant problem with the chip camera is its spectral response. Even if the chip is reversed and thinned so that light enters from the side opposite the electrodes, very little blue light penetrates into the semiconductor to produce electrons. On the other hand, infrared light penetrates easily and these cameras have red and infrared (IR) sensitivity that far exceeds that of human vision, usually requiring the installation of a blocking filter to exclude it (because the IR light is not focused to the same plane as the visible light and thus produces blurred or fogged images). Figure 1.12 shows this spectral response, which can be further tailored and extended by using materials other than silicon. The chip can reach a high total efficiency when antireflective coatings are applied, limited primarily by the “fill factor” — the area frac- tion of the chip that contains active devices between the narrow ditches that maintain electri- cal separation. Also, the chip camera has an output that is linearly proportional to the incident light intensity, convenient for some measurement purposes but very different from human vision, the vidicon, and photographic film, which are all approximately logarithmic. Human vision notices brightness differences of a few percent, i.e., a constant ratio of change rather than a constant increment. Film is characterized by a response to light exposure which (after chemical development) produces a density vs. exposure curve such as that shown in Figure 1.13. The low end of this curve represents the fog level of the film, the density that is present even without exposure. At the high end, the film saturates to a maximum optical den- sity, for instance based on the maximum physical density of silver particles or dye molecules. In between, the curve is linear with a slope that represents the contrast of the film. A steep slope corresponds to a high-contrast film that exhibits a large change in optical density with a small change in exposure. Conversely, a low-contrast film has a broader latitude to record a scene with a greater range of brightnesses. The slope of the curve is usually called “gamma.” Many chip cameras include circuitry or processing that changes their output from linear to logarithmic so that the image contrast is more familiar to viewers. The more expensive con- sumer cameras and most professional cameras include the possibility to read the “raw” linear data as well as the converted image. t t t t t t t 1 2 3 4 5 6 1 φ φ φ 1 2 3 1 pixel period Semiconductor Insulator Electron Wells Transferring Charge Figure 1.11 Varying voltages on a set of three electrodes shifts electrons from one pixel to another in a CCD.
- 33. 10 The Image Processing Handbook, Sixth Edition © 2011 by Taylor & Francis Group, LLC When film is exposed directly to electrons, as in the transmission electron micrograph, rather than photons (visible light or X-rays), the response curve is linear rather than logarithmic. Many light photons are needed to completely expose a single silver halide particle for development, but only a single electron is needed. Consequently, electron image films and plates are often very high in density (values of optical density greater than 4, which means that 9999/10000 of incident light is absorbed), which creates difficulties for many scanners and requires more than 8 bits to record. The trend in camera chips has been to make them smaller and to increase the number of pixels or diodes present. Some scientific cameras, such as that used in the Hubble telescope, occupy an entire wafer. But for consumer devices, making each chip one-third, one-quarter, or even two-tenths of an inch in overall (diagonal) dimension places many devices on a single wafer and allows greater economic yield. It also requires smaller, less costly lenses. Putting more pixels into this reduced chip area (for more spatial resolution, as discussed below) makes the individual detectors small, but the ditches between then have to remain about the same size to prevent electrons from diffusing laterally. The result is to reduce the total effi- ciency markedly. Some devices place small lenses over the diodes to capture light that would Figure 1.12 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 long, medium and short wavelengths. (a) (b)
- 34. 11 Acquiring Images © 2011 by Taylor Francis Group, LLC otherwise fall into the ditches, but these add cost and also are not so uniform as the diodes themselves (which are typically within 1% across the entire chip). The other, and more important, effect of making the detectors small is to reduce their capacity for electrons, called the well capacity. A typical 15 µm pixel in a scientific grade CCD has a capacity of about 500,000 electrons, which with low readout noise (as can be achieved in spe- cial situations) of a few electrons gives a dynamic range greater than photographic film. Even larger well capacity and dynamic range can be achieved by combining (binning) more detec- tors for each stored pixel by using more steps in the phase shifting during readout. Reducing the area of the detector reduces the well size, and with it the dynamic range. Increasing the noise, for instance by reading out the signal at video rates (each horizontal line in 52 µs for US standard definition video), dramatically reduces the dynamic range so that a typical consumer grade video camera has no more than about 64 distinguishable brightness levels (expensive studio cameras meet the broadcast video specification of 100 levels). Since with the chip camera these are linear with brightness, they produce even fewer viewable gray levels, as shown in Figure 1.14. This performance is much inferior to film, which can distin- guish thousands of brightness levels. CMOS (Complementary Metal-Oxide Semiconductor) chips can also be used as image sensors, and in terms of sheer numbers are now more common than the original CCD devices. They are primarily used in relatively inexpensive consumer cameras and camera phones, although some have found their way into digital single lens reflex cameras. The conversion of light photons to electrons functions in the same way as in the CCD chip. The differences start with the way the signal is read out. In the CMOS designs there are from two to four transistors immediately adjacent to the light sensor which convert the charge to a voltage and amplify the signal. In principle, this means that any pixel in the array can be read out directly, addressing a pixel by row and column just as in a memory chip (Figure 1.15). This is different from the CCD method of “sweeping” the charge out to one corner of the array, reading all of the pixels in a fixed order. The space taken up by these control transistors reduces the “fill factor” or active area of the chip that is sensitive to light, but this is often compensated for by placing lenses over each detector to collect light from the dead areas and direct it to the active sensor. The lenses, and the use of individual amplifiers for each pixel, generally make the sensors in a CMOS detec- tor less uniform than those in the CCD array, producing a fixed pattern that can be compen- sated for in software (requiring recording an image with uniform illumination). In addition to the fixed pattern noise, the CMOS detectors usually have a greater amount of random noise Figure 1.13 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, while low (“soft”) contrast gives a less steep curve and films have a greater dynamic range. Saturation Fog Level Reciprocity Failure Linear Range (Slope = Gamma) Log (Exposure) Film Density
- 35. 12 The Image Processing Handbook, Sixth Edition © 2011 by Taylor Francis Group, LLC superimposed on the image signal because of the separate amplifiers, additional wiring and its associated capacitance and thermal noise, and greater dark current. The very small active regions (at least in the smaller chips used in pocket cameras and phones, and particularly as the pixel counts have risen to several million) have small well capacities, resulting in limited dynamic range for the images. The images are usually stored with 8 bits per channel, because of the way memory is traditionally organized, but often do not have that much actual bright- ness resolution. Larger area CMOS chips are also made which have larger detectors and consequently a greater well capacity and greater dynamic range. One advantage of the CMOS designs as used in more expensive cameras arises from the fact that the circuitry to access the pixels can be arranged along two adjacent sides of the array (addressing the rows and columns, respectively). That makes it possible to carefully trim away the chip on the other two sides, and arrange four of the chips together to produce a larger sensor with higher pixel counts. This approach, com- bined with the use of much larger sensors to achieve greater sensitivity and dynamic range, Figure 1.14 Comparison of visibility of gray level steps from linear (equal steps) and logarithmic (equal ratios) detectors: (a) Plots of intensity. (b) Display of the values from (a).
- 36. 13 Acquiring Images © 2011 by Taylor Francis Group, LLC has led some manufacturers to prefer CMOS detectors as large as a traditional film negative for digital single-lens reflex cameras. The advantages of CMOS sensors lie primarily in three areas: they consume much less power, and so give better battery life; the amplifier and digitization circuitry can be placed on the same chip to reduce size and perhaps increase ruggedness; and the production methods for the wafers are essentially the same as those used for other standard silicon devices such as memory and processors, whereas CCD wafers require unique processing. The latter advantage is somewhat offset by the fact that high quality CMOS sensors do require somewhat custom- ized fabrication for greater uniformity and fewer noise-producing imperfections than can be tolerated in other devices. While the cost to fabricate CMOS sensors is less than for CCD, the design costs are much higher. Of course, for devices that are to be produced in large quantity, this is a minor factor. The overall trend has been for CMOS sensors to continue to improve in quality and performance, and while the advantages of the CCD sensor are still important for most technical applications, it is wise to consider the trade-offs on a case-by-case basis (Nakamura, 2006; Holst Lomheim, 2007). Camera artifacts and limitations There are several problems with video cameras using chips which contribute to the specific types of defects present in the images that must be dealt with by subsequent processing. One is the fact that many video signals are interlaced (Figure 1.16). With high-definition video, and with digital still cameras, the image is scanned progressively. Interlacing is a clever trick to minimize visual flicker in broadcast television images, accomplished with tube cameras by scanning the electron beam in the same interlace pattern as the display television set. With a chip camera, it requires that the array be read out twice for every 30th of a second frame, once to collect the even numbered lines and again for the odd numbered lines. In fact, many cam- eras combine two lines to get better sensitivity, averaging lines 1 and 2, 3 and 4, 5 and 6, and so on, in one interlace field, and then 2 and 3, 4 and 5, 6 and 7, etc. in the other. This reduces Figure 1.15 Schematic diagram of a typical CMOS detector. Each active light sensor (green) has additional transistors that are connected to addressing and output lines.
- 37. 14 The Image Processing Handbook, Sixth Edition © 2011 by Taylor Francis Group, LLC vertical resolution but for casual viewing purposes is not noticeable. Motion can cause the even and odd fields of a full frame to be offset from each other, producing a significant deg- radation of the image, as shown in the figure. A similar effect occurs with stationary images if the horizontal retrace signal is imprecise or difficult for the electronics to lock onto; this is a particular problem with signals played back from consumer video tape recorders. (Moving images are also distorted with progressive scan cameras, due to the time required to read from the top of the image to the bottom.) During the transfer and readout process, unless the camera is shuttered either mechanically or electrically, photons continue to produce electrons in the chip. This produces a large 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Line Field 1 Field 2 Figure 1.16 (a) Interlace scan covers even numbered lines in one six- tieth-second field, and even numbered lines in a second field. (b) When motion is present (either in the scene or caused by camera motion), this pro- duces an offset in the com- plete image. (a) (b)
- 38. 15 Acquiring Images © 2011 by Taylor Francis Group, LLC background signal that further degrades dynamic range and may produce blurring. Electronic shuttering is usually done line-at-a-time so that moving images are distorted. Some designs avoid shuttering problems by doubling the number of pixels, with half of them opaque to incoming light. A single transfer shifts the electrons from the active detectors to the hidden ones, from which they can be read out. Of course, this reduces the active area (fill factor) of devices on the chip, reducing sensitivity by 50%. The high speed of horizontal line readout can produce horizontal blurring of the signal, again reducing image resolution. This is partially due to inadequate time for the electrons to diffuse along with the shifting fields, to the time needed to recover electrons from traps (impurities in the silicon lattice), and partially to the inadequate frequency response of the amplifier, which is a trade-off with amplifier noise. Even though the individual electron transfers are very efficient, better than 99.999% in most cases, the result of being passed through many such transfers before being collected and amplified increases the noise. This varies from one side of the chip to the other, and from the top to the bottom, and can be visually detected in images if there is not a lot of other detail or motion to obscure it. Many transfers of electrons from one detector to another occur during readout of a chip, and this accounts for some of the noise in the signal. Purely statistical variations in the produc- tion and collection of charge is a relatively smaller effect. The conversion of the tiny charge to a voltage and its subsequent amplification is the greatest source of noise in most systems. Readout and amplifier noise can be reduced by slowing the transfer process so that fewer elec- trons are lost in the shifting process and the amplifier time constant can integrate out more of the noise, producing a cleaner signal. Cooling the chip to about –40° also reduces the noise from these sources and from dark current, or thermal electrons. Slow readout and cooling are used only in non-video applications, of course. Digital still cameras use the same chip technol- ogy (but much higher numbers of detectors) as solid state video cameras, and produce higher quality images because of the slower readout. Janesick (2001) discusses the various sources of noise and their control in scientific CCDs of the type used in astronomical imaging (where they have almost entirely replaced film) and in space probes. Color cameras Color cameras can be designed in three principal ways, as shown in Figures 1.17, 1.18, and 1.19. For stationary images (which includes many scientific applications such as microscopy, but excludes “real-time” applications such as video), a single detector array can be used to acquire three sequential exposures through red, green and blue filters, respectively (Figure 1.17), which are then combined for viewing. The advantages of this scheme include low cost and the ability to use different exposure times for the different color bands, which can compensate for the poorer sensitivity of the silicon chip to short wavelength (blue) light. Many high-end consumer and most professional grade video cameras use three sensors (Figure 1.18). A prism array splits the incoming light into red, green, and blue components, which are recorded by three different sensors whose outputs are combined electronically to produce a standard video image. This approach is more costly, since three chips are needed, but for video applications they need not be of particularly high resolution (even a high-defini- tion video camera has many fewer pixels than a digital still camera). The optics and hardware to keep everything in alignment add some cost, and the depth of the prism optics makes it
- 39. 16 The Image Processing Handbook, Sixth Edition © 2011 by Taylor Francis Group, LLC impractical to use short focal length (wide angle) lenses. This design is rarely used in digital still cameras. Video images are often digitized into a 640 × 480 array of stored pixels (the dimensions of the VGA display that was once standard for personal computers), but this is not the actual resolution of the image. The broadcast bandwidth limits the high frequencies and eliminates any rapid variations in brightness and color. A standard definition video image has no more than 330 actual elements of resolution in the horizontal direction for the brightness (lumi- nance) signal, and about half that for the color (chrominance) information. Color information is intentionally reduced in resolution because human vision is not very sensitive to blurring of color beyond boundary lines. Of course, video signals can be further degraded by poor equipment. Recording video on consumer-grade tape machines can reduce the resolution by another 50% or more, particularly if the record head is dirty or the tape has been used many times before (an unfortunately very Figure 1.17 Schematic diagram of a color wheel camera with red, green and blue filters. The fourth filter position is empty, allowing the camera to be used as a mono- chrome detector with greater sensitivity for dim images (e.g., fluorescence microscopy). Figure 1.18 Schematic diagram of the prisms and dichroic filters for a three chip color camera.
- 40. 17 Acquiring Images © 2011 by Taylor Francis Group, LLC common problem with forensic examination of surveillance video is that the tapes are played — over and over — for visual examination by local police so that by the time professionals get them, the oxide coating — and the information — has been damaged or even removed). Video images are not very high resolution, although HDTV (high definition television) has improved things somewhat. Consequently, video technology is usually a poor choice for scientific imag- ing unless there is some special need to capture “real time” images (i.e., 25–30 frames per second) to record changes or motion. Digital still cameras have largely replaced them, as they produce much higher resolution images with greater dynamic range. Most digital cameras use a single pixel array, often of very high pixel (detector) count, with a color filter that allows red, green, and blue light to reach specific detectors. Different patterns may be used (Figure 1.19), with the Bayer pattern being very common (invented by Kodak researcher Bryce Bayer and the basis for U.S. Patent 3,971,065 “Color Imaging Array,” issued in 1976). Notice that it assigns twice as many detectors for green as for red or blue, which mim- ics to some extent the human eye’s greater sensitivity to green. The problem with the single- chip camera, of course, is that the image resolution in each color channel is reduced. The red intensity at some locations must be interpolated from nearby sensors, for example. It is also necessary to design the filters to give the same brightness sensitivity in each channel. If this is not done well, a herring-bone pattern (often referred to as a “zipper”) appears in images of a uniform gray test card and color fringes appear along contrast edges in the picture, as shown in Figure 1.20. Interpolation techniques for Bayer pattern color filters reduce the image resolution as com- pared to the number of individual detectors in the camera (which is generally the speci- fication advertised by the manufacturer). Inherently, this “demosaicking” process involves trade-offs between image sharpness, details, noise, processing time and conversion artifacts. The quality of the result, judged by its ability to preserve sharp boundaries in brightness while minimizing the introduction of color artifacts, varies inversely with the computational (a) (b) Figure 1.19 (a) Stripe and (b) Bayer filter patterns used in single chip cameras.
- 41. 18 The Image Processing Handbook, Sixth Edition © 2011 by Taylor Francis Group, LLC requirements. A comparison of several patented methods can be found in Ramanath (2000) and Shao et al. (2005), Tamburino et al. (2010), and Guanara et al. (2010). The combination of the specific color filter array arrangement and the camera’s interpolation firmware leaves a signature in images that can be used in some cases to identify the model of camera used to photograph a scene, even to identify specific fixed pattern noise from an individual camera, and to detect alterations made to the image later (Bayram et al., 2006; Swaminathan et al., 2007; Farid, 2008). Pattern noise is not unique to single-chip cameras with a color filter array. Three-chip cameras also have potential problems because all chips have some slight variations in the output from individual transistors. In a three-chip system these produce different variations in the red, green, and blue output that increase the color variations in the images. Another approach to color camera design, developed by Foveon Corp. and used in a few cameras, creates three transistors at each pixel location, stacked on top of each other, using CMOS technology. Blue light penetrates the shortest distance in silicon and is detected in the topmost transistor. Green light penetrates to the second transistor and red light penetrates to the bottom one. The output signals are combined to produce the color information. This approach does not suffer from loss of spatial resolution due to interpolation, but has potential problems with consistent or accurate color fidelity. Camera resolution The signal coming from the silicon detector is analog, even if the digitization takes place within the camera housing or even on the same chip, so the interpolation is done in the amplifier stage. In most cases, the actual image resolution with a single chip camera and filter arrange- ment is one-half to two-thirds the value that might be expected from the advertised number of pixels in the camera, because of this interpolation. And some cameras record images with many more stored pixels than the chip resolution warrants in any case. Such interpolation and empty magnification contribute no additional information in the image. Figure 1.20 Example of “zipper” patterns resulting from poor interpolation in a single-chip digital camera.
- 42. 19 Acquiring Images © 2011 by Taylor Francis Group, LLC Comparing cameras based on actual resolution rather than the stated number of recorded pixels can be difficult. It is important to consider the multiple meanings of the word “pixel.” In some contexts, it refers to the number of light detectors in the camera (without regard to any color filtering, and sometimes including ones around the edges that do not contribute to the actual image but are used to measure dark current). In some contexts it describes the number of recorded brightness or color values stored in the computer, although these may represent empty magnification. In other situations it is used to describe the displayed points of color on the computer monitor, even if the image is shown in a compressed or enlarged size. It makes much more sense to separate these various meanings and to talk about resolution elements when considering real image resolution. This refers to the num- ber of discrete points across the image that can be distinguished from each other and is sometimes specified in terms of the number of line pairs that can be resolved. This is one- third to one-half the number of resolution elements, since at least one element is needed for the line and one for the space between lines. It depends on the amount of brightness contrast between the lines and the spaces, and the amount of noise (random variation) present in the image. The situation is even more complicated with some digital still cameras that shift the detec- tor array to capture multiple samples of the image. The most common method is to use a piezo device to offset the array by half the pixel spacing in the horizontal and vertical directions, capturing four images that can be combined to more or less double the resolu- tion of the image as data are acquired from the gaps between the original pixel positions. For an array with colored filters, additional shifts can produce color images with resolu- tion approaching that corresponding to the pixel spacing. Some studio cameras displace the entire sensor array to different regions of the film plane to collect tiles that are sub- sequently assembled into an image several times as large as the detector array. Of course, the multiple exposures required with these methods means that more time is required to acquire the image. Rather than a two-dimensional array of detectors, it is also possible to use a linear array (or sometimes three, one each with red, green, and blue filters) that is swept across the image plane to acquire the data. This method is common in desk-top scanners (which for many applications are perfectly usable image acquisition devices). It has also been used in studio cameras, and some light microscopes accomplish the same thing by moving the stage and specimen under the optics so that an image of an entire 1 × 3 inch slide can be obtained with high spatial resolution. The image file produced is huge; special software is required to efficiently access the stored array (Bacus Bacus, 2000, 2002) and to interactively deliver a selected portion of the image data as the user varies position and magnification. Network access to such stored images also presents bandwidth challenges, but facilitates collaboration and teaching. With either a single-chip or three-chip camera, the blue channel is typically the noisiest due to the low chip sensitivity to blue light and the consequent need for greater amplifi- cation. In many cases, processing software that reduces image noise using averaging or median filters (discussed in Chapter 4) can be applied separately to each color channel, using different parameters according to the actual noise content, to best improve image appearance. Digital cameras using the same chip technology as a video camera can produce much better image quality. This is due in part to the longer exposure times, which collect more electrons and so reduce noise due to statistics and amplification. Also, the slower readout of the data from the chip, which may take a second or more instead of 1/60th of a second, reduces
- 43. 20 The Image Processing Handbook, Sixth Edition © 2011 by Taylor Francis Group, LLC readout noise. Digital still cameras read out the data in one single pass (progressive scan), not with an interlace. By cooling the chip and amplifier circuitry to reduce dark currents, integra- tion (long exposures up to tens of seconds, or for some astronomical applications many min- utes) can be used to advantage because of the high dynamic range (large well size and large number of bits in the digitizer) of some chip designs. In addition, the ability to use a physical rather than electronic shutter simplifies chip circuitry and increases fill factor. The number of pixels in video cameras need not be any greater than the resolution of the video signal, which, as noted above, is rather poor. In a digital still camera, very high pixel counts can give rise to extremely high resolution, which rivals film in some cases. There is also an interesting cross-over occurring between high end consumer and professional scientific grade cameras. In addition to dedicated cameras for attachment to microscopes or other separate optics, manufacturers are producing consumer single-lens reflex cameras with enough resolution (15 to 20 million pixels at this writing) that it is becoming practical to use them in technical applications, and simple optical attachments make it easy to connect them to microscopes or other instruments (and of course the camera may also be removed and used for other purposes). The camera may be tethered directly to a computer, but in many cases it is more practical to record the images to memory chips that are later downloaded to the com- puter. Professional digital cameras with large, high resolution detector arrays, interchangeable lenses, etc., are providing capabilities that compete with traditional 35mm and larger film cameras. Every manufacturer of cameras has recognized the shift away from film and toward digital recording, and an incredibly wide variety of cameras is now available, with new devel- opments appearing frequently. The benefits of a camera with a large number of sensors (high pixel count), as well as large individual sensors (large well size and consequent high dynamic range), seem obvious and desirable. For some applications, high pixel counts are not so important. At high optical magnification, the important limitation is the optical resolution. In the rather typical setup of my bench microscope, with a 10x (low magnification) objective lens, the image projected onto the chip by the transfer optics covers about 1600 µm width on the specimen. With a 100× (high magnification) objective lens that becomes 160 µm. For a camera with 3600 × 2400 sensors (less than 10 megapixels) the low magnification image is recorded at about 1 pixel per micron, adequate for the resolution of the optics. The high magnification image is recorded with 90 pixels per micron. Since the optical resolution of the microscope under optimum conditions is about 0.5 µm with the 100× lens, this produces a vast and unneces- sary oversampling. At low magnifications, or for viewing fine detail in large scenes (such as aerial and satellite imagery), high pixel counts make sense. When the limitation on resolu- tion lies with the optics, it may not. Focusing Regardless of what type of camera is employed to acquire images, it is important to focus the optics correctly to capture the fine details in the image. Often the human eye is used to per- form this task manually. In some situations, such as automated microscopy of pathology slides or surveillance tracking of vehicles, automatic focusing is required. This brings computer processing into the initial step of image capture. Sometimes, in the interests of speed, the processing is performed in dedicated hardware circuits attached to the camera. But in many cases the algorithms are the same as might be applied in the computer (described in Chapter
- 44. 21 Acquiring Images © 2011 by Taylor Francis Group, LLC 5), and the focusing is accomplished in software by stepping the optics through a range of settings and choosing the one that gives the “best” picture. Several different approaches to automatic focus are used. Cameras used for macroscopic scenes may employ methods that use some distance measuring technology, e.g., using high frequency sound or infrared light, to determine the distance to the subject so that the lens position can be adjusted. In microscopy applications this is impractical, and the variation with focus adjustment captured in the image itself must be used. Various algorithms are used to detect the quality of image sharpness, and all are successful for the majority of images in which there is good contrast and fine detail present. Each approach selects some implementa- tion of a high-pass filter output which can be realized in various ways, using either hardware or software, but must take into account the effect of high frequency noise in the image and the optical transfer function of the optics (Green et al., 1985; Firestone et al., 1991; Boddeke et al., 1994; Sun et al., 2004; Buena-Ibarra, 2005; Bueno et al., 2005; Brazdilova Kozubek, 2009; Shim et al., 2010.) Electronics and bandwidth limitations Video cameras of either the solid-state chip or tube type produce analog voltage signals cor- responding to the brightness at different points in the image. In the standard definition RS-170 signal convention, the voltage varies over a 0.7-volt range from minimum to maximum bright- ness, as shown above in Figure 1.8. The scan is nominally 525 lines per full frame, with two interlaced 1/60th-second fields combining to make an entire image. Only about 480 of the scan lines are usable, with the remainder lost during vertical retrace. In a typical broadcast television picture, more of these lines are lost due to overscanning, leaving about 400 lines in the actual viewed area. The time duration of each scan line is 62.5 µs, part of which is used for horizontal retrace. This leaves 52 µs for the image data, which must be subdivided into the horizontal spacing of discernible pixels. For PAL (European) television, these values are slightly different based on a 1/25th-second frame time and more scan lines, and the resulting resolution is slightly higher. Until recently in the United States, broadcast television stations were given only a 4-MHz bandwidth for their signals, which must carry color and sound information as well as the brightness signal. This narrow bandwidth limits the number of separate voltage values that can be distinguished along each scan line to a maximum of 330, as mentioned above, and this value is reduced if the signal is degraded by the electronics or by recording using standard videotape recorders. Consumer-quality videotape recorders reduce the effective resolution substantially; in “freeze frame” playback, they display only one of the two interlaced fields, so that only about 200 lines are resolved vertically. Using such equipment as part of an image analysis system makes choices of cameras or digitizer cards on the basis of resolution (actually the number of sampled pixels) irrelevant. There is a major difference between the interlace scan used in standard definition television and a non-interlaced or “progressive” scan. The latter gives better quality because there are no line-to-line alignment or shift problems. Most high definition television (HDTV) modes use progressive scan. The format requires a higher rate of repetition of frames to fool the human eye into seeing continuous motion without flicker, but it has many other advantages. These include simpler logic to read data from the camera (which may be incor- porated directly on the chip), more opportunity for data compression because of redun- dancies between successive lines, and simpler display and storage devices. Practically all
- 45. 22 The Image Processing Handbook, Sixth Edition © 2011 by Taylor Francis Group, LLC scientific imaging systems such as digital cameras, direct-scan microscopes (the scanning electron microscope or SEM, scanning tunneling microscope or STM, the atomic force microscope or AFM, etc.), flat-bed scanners, film or slide digitizers, and similar devices use progressive scan. HDTV modes include many more differences from conventional television than the use of progressive scan. The pixel density is much higher, with a wider aspect ratio of 16:9 (instead of the 4:3 used in NTSC television) and the pixels are square. A typical HDTV mode presents 1920 × 1080 pixel images at the rate of 60 full scans per second, for a total data rate exceeding 2 gigabits per second, several hundred times as much data as analog broadcast television. One consequence of this high data rate is the use of data compression techniques, which are discussed in Chapter 3, and the use of digital transmission tech- niques using cable or optical fiber instead of broadcast channels. Over-the-air, satellite, and cable transmission of HDTV signals all involve compression, often with a significant loss of image quality. Regardless of the effects on consumer television, the development of HDTV hardware is likely to produce spin-off effects for computer imaging, such as high pixel density cameras with pro- gressive scan output, high bandwidth recording devices, and superior CRT or LCD displays. For example, color cameras being designed for HDTV applications output digital rather than analog information by performing the analog-to-digital conversion within the camera, with at least 10 bits each for red, green, and blue. Even the best system can be degraded in performance by such simple things as cables, con- nectors, or incorrect termination impedance. Another practical caution in the use of standard cameras is to avoid automatic gain or brightness compensation circuits. These can change the image contrast or linearity in response to bright or dark regions that do not even lie within the digitized portion of the image, make comparison between images difficult, and increase the gain and noise for a dim signal. Figure 1.21 shows a micrograph with its brightness histogram. This is an important tool for image analysis, which plots the number of pixels as a function of their brightness values. It is used extensively in subsequent chapters. The histogram shown is well spread out over the available 256 brightness levels, with peaks corresponding to each of the structures in the metal sample. If a bright light falls on a portion of the detector in the solid-state camera that is Figure 1.21 A gray scale image digitized from a metallographic microscope and its brightness histo- gram, which plots the number of pixels with each possible brightness value.
- 46. 23 Acquiring Images © 2011 by Taylor Francis Group, LLC not part of the image area of interest (e.g., due to internal reflections in the optics), automatic gain circuits in the camera may alter the brightness-voltage relationship so that the image changes. This same effect occurs when a white or dark mask is used to surround images placed under a camera on a copy stand. The relationship between structure and brightness is changed, making subsequent analysis more difficult. Issues involving color correction and calibration are dealt with below, but obtaining absolute color information from video cameras is not possible because of the broad range of wave- lengths passed through each filter, the variation in illumination color (even with slight voltage changes on an incandescent bulb), and the way the color information is encoded. Matching colors so that the human impression of color is consistent requires calibration, which is dis- cussed in Chapter 4. The color temperature of the illumination used is critical to matching colors in images. Figure 1.22 shows an image recorded using filtered sunlight, with an effective color tem- perature (described more fully in Chapter 3) of approximately 5000K, using a white card and prior exposure to allow the camera to perform a color balance adjustment. Opening the raw image file with different assumed color temperatures produces substantial changes in the visual perception of the colors. Digitization of the analog voltage signal from the detector may be done either in the camera or in a separate external circuit (such as a “frame grabber” board placed inside the computer). The analog signal is usually digitized with a “flash” ADC (analog-to-digital converter). This is a chip using successive approximation techniques (described above) to rapidly sample and measure the voltage. For video-rate imaging this must be done in less than 100 ns, producing a number value from 0 to 255 that represents the brightness. Slower readout allows for more than 8 bit conversion, and many digital still cameras have 12 or even 14 bit ADCs, although the dynamic range and noise level in the detector may not be that good. The brightness number is stored in memory and another reading made, so that a series of brightness values is obtained along each scan line. Figure 1.23 illustrates the digitization of a signal into equal steps in both time and value. Additional circuitry is needed to trigger each series of readings so that positions along successive lines are consistent. Digitizing several hundred or thousand points along each scan line, repeating the process for each line, and transmitting or storing the val- ues into memory produces a digitized image for further processing or analysis. Figure 1.22 An image taken with filtered sunlight and an effective color temperature of 5000K, but stored as a raw file and opened using different assumed color temperatures. From left to right, 3500, 4500, 5500, 6500K.
- 47. 24 The Image Processing Handbook, Sixth Edition © 2011 by Taylor Francis Group, LLC Pixels It is most desirable to have the spacing of the pixel values be the same in the horizontal and vertical directions (i.e., square pixels), as this simplifies many processing and measurement operations. There are some theoretical advantages to having pixels arranged as a hexagonal grid, but because of the way that all acquisition hardware actually functions, and to simplify the addressing of pixels in computer memory, this is almost never done. Accomplishing the goal of square pixels with an analog video camera requires a well-adjusted clock to control the acquisition. Since the standard-definition video image is not square, but has a width-to-height ratio of 4:3, the digitized image may represent only a portion of the entire field of view. Digitizing boards (frame grabbers) were first designed to record 512 × 512 arrays of values, since the power-of-two dimension simplified design and memory addressing. Later generations acquired a 640 wide by 480 high array, which matched the image propor- tions and the size of standard VGA display monitors while keeping the pixels square. Because of the variation in clocks between cameras and digitizers, it was common to find distortions of several percent in pixel squareness. This can be measured and compensated for after acquisi- tion by resampling the pixels in the image, as Chapter 4 describes. Most digital still cameras acquire images that have a width-to-height ratio of 4:3 (the aspect ratio of conventional video) or 3:2 (the aspect ratio of 35mm film) and have square pixels. Since pixels have a finite area, those which straddle a boundary in the scene effectively aver- age the brightness levels of two regions and have an intermediate brightness that depends on how the pixels lie with respect to the boundary. This means that a high lateral pixel resolution and a large number of distinguishable gray levels are needed to accurately locate boundaries. Figure 1.24 shows several examples of an image with varying numbers of pixels across its width, and Figure 1.25 shows the same image with varying numbers of gray levels. For the most common types of image acquisition devices, such as cameras, the pixels rep- resent an averaging of the signal across a finite area of the scene or specimen. However, there are other situations in which this is not so. At low magnification, for example, the scanning electron microscope beam samples a volume of the specimen much smaller than the dimension of a pixel in the image. So does the probe tip in a scanned probe Figure 1.23 Digitization of an analog voltage signal along one line in an image (blue) produces a series of values that correspond to a series of steps (red) equal in time and rounded to integral multiples of the smallest measurable increment.
- 48. 25 Acquiring Images © 2011 by Taylor Francis Group, LLC microscope. Range imaging of the moon from the Clementine orbiter determined the elevation of points about 10 cm in diameter using a laser rangefinder, but at points spaced apart by 100 meters or more. In these cases, the interpretation of the relationship between adjacent pixels is slightly dif- ferent. Instead of averaging across boundaries, the pixels sample points that are discrete and well separated. Cases of intermediate or gradually varying values from pixel to pixel are rare, and the problem instead becomes how to locate a boundary between two sampled points on either side. If there are many points along both sides of the boundary, and the boundary can be assumed to have some geometric shape (such as a locally straight line), fitting methods can Figure 1.24 Four representations of the same image, showing a variation in the number of pixels used. From the upper left: 256 × 256; 128 × 128; 64 × 64; 32 × 32. In all cases, a full 256 gray values are retained. Each step in coarsening of the image is accomplished by averaging the brightness of the region covered by the larger pixels.
- 49. 26 The Image Processing Handbook, Sixth Edition © 2011 by Taylor Francis Group, LLC be used to locate it to a fraction of the pixel spacing. These methods are discussed further in Chapter 10 on image measurements. Gray scale resolution In addition to defining the number of sampled points along each scan line, and hence the resolution of the image, the design of the ADC also controls the precision of each measure- ment. High speed flash analog-to-digital converters usually measure each voltage reading to produce an 8-bit number from 0 to 255. This full range may not be used for an actual image, which may not vary from full black to white. Also, the quality of most analog video cameras Figure 1.25 Four representations of the same image, with variation in the number of gray levels used. From the upper left: 32; 16; 8; 4. In all cases, a full 256 × 256 array of pixels is retained. Each step in the coarsening of the image is accomplished by rounding the brightness of the original pixel value.
- 50. 27 Acquiring Images © 2011 by Taylor Francis Group, LLC and other associated electronics rarely produces voltages that are free enough from electronic noise to justify full 8-bit digitization anyway. A typical “good” camera specification of 49 dB signal-to-noise ratio implies that only 7 bits of real information are available, and the eighth bit is random noise. But 8 bits corresponds nicely to the most common organization of com- puter memory into bytes, so that 1 byte of storage can hold the brightness value from 1 pixel in the image. High end digital still cameras and most scanners produce more than 256 distinguishable brightness values, and for these it is common to store the data in 2 bytes or 16 bits, giving a possible range of 65536:1, which exceeds the capability of any current imaging device (but not some other sources of data that may be displayed as images, such as surface elevation mea- sured with a scanned probe, a topic in Chapter 15). For a camera with a 10 or 12 bit output, the values are shifted over to the most significant bits and the low order bits are either zero or random values. For display and printing purposes 8 bits is enough, but the additional depth can be very important for processing and measurement, as discussed in subsequent chapters. In many systems the histogram of values is still expressed as 0.255 for compatibility with the more common 8-bit range, but instead of being restricted to integers the brightness consists of floating point values. That is the convention used in this book. When the stored image is displayed from computer memory, the numbers are used in a dig- ital-to-analog converter to produce voltages that control the brightness of a display monitor, often a cathode ray tube (CRT) or liquid crystal display (LCD). This process is comparatively noise-free and high resolution, since computer display technology has been developed to a high level for other purposes. These displays typically have 256 steps of brightness for the red, green, and blue signals, and when equal values are supplied to all three the result is perceived as a neutral gray value. The human eye cannot distinguish all 256 different levels of brightness in this type of display, nor can they be successfully recorded or printed using ink-jet or laser printers (discussed in Chapter 3). About 20–40 brightness levels can be visually distinguished on a CRT, LCD, or photographic print, suggesting that the performance of the digitizers in this regard is more than adequate, at least for those applications where the performance of the eye is enough to begin with, or the purpose of the imaging is to produce prints. A somewhat different situation that results in another limitation arises with images that cover a very large dynamic range. Real-world scenes often include brightly lit areas and deep shade. Scientific images such as SEM pictures have very bright regions corresponding to edges and protrusions and very dark ones such as the interiors of depressions. Astronomical pictures range from the very bright light of stars to the very dark levels of dust clouds or interstellar space. If only 256 brightness levels are stretched to cover this entire range, there is not enough sensitivity to small variations to reveal detail in either bright or dark areas. Capturing images with higher bit depth, for instance 12 bits (4096 brightness levels, which is approximately the capability of a film camera), can record the data, but it cannot be viewed successfully on a display screen or in a print. Processing methods that can deal with such high dynamic range images to facilitate visual interpretation are shown in Chapter 5. Images acquired in very dim light, or some other imaging modalities such as X-ray mapping in the scanning electron microscope (SEM), impose another limitation of the gray scale depth of the image. When the number of photons (or other particles) collected for each image pixel is low, statistical fluctuations and random noise become important. Figure 1.26 shows the effect of high ASA settings (high amplifier gain) on random pixel variations in an image. The two
- 51. 28 The Image Processing Handbook, Sixth Edition © 2011 by Taylor Francis Group, LLC images were recorded with the same camera and identical illumination and aperture settings; changing the ASA setting on the camera resulted in different exposure times. Figure 1.27 shows a fluorescence microscope image in which a single video frame illustrates substantial statistical noise, which prevents distinguishing or measuring the structures pres- ent. Averaging together multiple frames collects more signal and results in an improvement in the signal- to noise-ratio, and hence in the visibility of detail. Noise Images in which the pixel values vary within regions that are ideally uniform in the original scene can arise either because of limited counting statistics for the photons or other signals, losses intro- duced in the shifting of electrons within the chip, or due to electronic noise in the amplifiers or cabling. In any case, the variation is generally referred to as noise, and the ratio of the contrast which is due to differences present in the scene represented by the image to the noise level is the signal-to-noise ratio. When this is low, the features present may be invisible to the observer. Figure 1.28 shows an example in which several features of different size and shape are super- imposed on a noisy background with different signal-to-noise ratios. The ability to discern the presence of the features is generally proportional to their area and independent of shape. In the figure, a smoothing operation is performed on the image with the poorest signal-to- noise ratio, which somewhat improves the visibility of the features. The methods available for improving noisy images by image processing are discussed in the chapters on spatial and frequency domain methods. However, the best approach to noisy images, when it is available, is to collect more signal and improve the statistics. Increasing the exposure (either by increasing the exposure time, the lens aperture, or the illumination) reduces noise due to statistical effects, as shown in Figures 1.26 and 1.27. The improvement in quality is proportional to the square root of the amount of light (or other Figure 1.26 The same image recorded with an ASA setting of 1600 (top) and 100 (bottom), showing the increase in random pixel noise produced by higher gain in the camera.
- 52. 29 Acquiring Images © 2011 by Taylor Francis Group, LLC signal) collected. It is necessary to use a detector with a sufficient well size to hold the elec- trons and to use a sufficiently high bit depth in the image to preserve the contrast details. Cooling the detector and associated electronics chips can reduce electronic noise during long acquisitions of many minutes. Most uncooled camera chips begin to show unacceptable pixel noise due to dark current with integration times of more than a few seconds. Acquiring images at video rates of 25–30 frames per second is sometimes referred to as “real time” imaging, but of course this term should properly be reserved for any imaging rate that is adequate to reveal temporal changes in a particular application. For some situations, time- lapse photography may only require one frame to be taken at periods of many minutes, hours, or even days. For others, very short exposures and high rates are needed. Special cameras that do not use video frame rates or bandwidths can achieve rates up to ten times that of a standard video camera for full frames and even higher for small image dimensions. These cameras typically use a single line of detectors and optical deflection (e.g., a rotating mirror or prism) to cover the image area. For many applications, the repetition rate does not need to be that high. Either stroboscopic imaging or a fast shutter speed may be enough to stop the important motion to provide a sharp image. Electronic shutters can be used to control solid state imaging devices, instead Figure 1.27 Averaging of a noisy (low photon intensity) image (light microscope image of bone mar- row). From the upper left: one frame; averaging of 4; 16; 256 frames.
- 53. 30 The Image Processing Handbook, Sixth Edition © 2011 by Taylor Francis Group, LLC of a mechanical shutter. Exposure times under 1/10,000th of a second can easily be achieved, but of course this short exposure requires high illumination intensity. High depth images Other devices that produce data sets that are often treated as images for viewing and measure- ment produce data with a much greater range than a camera. For instance, a scanned stylus instrument that measures the elevation of points on a surface may have a vertical resolution of a few nanometers with a maximum vertical travel of hundreds of micrometers, for a range-to- resolution value of 105. This requires storing data in a format that preserves the full resolution values, and such instruments typically use 4 bytes per pixel. An elevation map of the earth’s surface encoded with 8 bits (256 values) spread over the range from sea level to the top of Mt. Everest corresponds to about 100 feet per bit and would not not show most of Florida as dis- tinct from sea level. With 2 bytes per pixel (65,536 values) each bit represents about 6 inches and the map can distinguish the curbs along most streets. With 4 bytes per pixel (4 billion values) each bit corresponds to less than 200 µm, and the roughness of sand on a beach can be recorded. In some cases with cameras having a large brightness range, the entire 12- or 14-bit depth of each pixel is stored. However, since this depth exceeds the capabilities of most CRTs to display, or of the user to see or print, reduction may be appropriate. If the actual bright- ness range of the image does not cover the entire possible range, scaling (either manual or (a) (b) (c) (d) Figure 1.28 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; (d) image (c) after spatial smoothing.
- 54. 31 Acquiring Images © 2011 by Taylor Francis Group, LLC automatic) to select just the range actually used can significantly reduce storage require- ments. Many computer programs (e.g., Adobe Photoshop®) offer routines to read the raw linear format from the camera and convert it with adjustments for lighting, vignetting, and contrast. In other cases, especially when performing densitometry, a conversion table is used. For den- sitometry, the desired density value varies as the logarithm of the brightness; this is discussed in detail in Chapter 10. A range of 256 brightness steps is not adequate to cover a typical range from 0 to greater than 3.0 in optical density (i.e., one part in 103 of the incident illumina- tion is transmitted) with useful precision, because at the dark end of the range, 1 step in 256 represents a very large step in optical density. Using a digitization with 12 bits (1 step in 4096) solves this problem, but it may be efficient to convert the resulting value with a logarithmic lookup table to store an 8-bit value (occupying a single computer byte) that represents the optical density. Lookup tables (LUTs) can be implemented either in hardware or software. They use the origi- nal value as an index into a stored or precalculated table, which then provides the derived value. This process is fast enough that acquisition is not affected. Many digital still cameras use LUTs to convert the linear output from the detector and ADC to a value that mimics the behavior of film. The LUTs discussed here are used for image acquisition, converting a 10-, 12-, or 14-bit digitized value with a nonlinear table to an 8-bit value that can be stored. LUTs are also used for displaying stored images, particularly to substitute colors for gray scale values to create pseudo-color displays, but also to apply correction curves to output devices (displays and printers) in order to match colors. This topic is discussed later in this chapter and in Chapter 3. Many images do not have a brightness range that covers the full dynamic range of the digi- tizer. The result is an image whose histogram covers only a portion of the available values for storage or for display. Figure 1.29 shows a histogram of such an image. The flat (empty) regions of the plot indicate brightness values at both the light and dark ends that are not used by any of the pixels in the image. Expanding the brightness scale by spreading the his- togram out to the full available range, as shown in the figure, may improve the visibility of features and the perceived contrast in local structures. The same number of brightness values are missing from the image, as shown by the gaps in the histogram, but now they are spread uniformly throughout the range. Other ways to stretch the histogram nonlinearly are shown in Chapter 5. Figure 1.29 Linear expansion of a histogram to cover the full range of storage or display.
- 55. Exploring the Variety of Random Documents with Different Content
- 56. Men who, 'mid noise and dirt, and play and prate, Could calmly mend the pen, and wash the slate. Punishments were rare; indeed, flogging was absolutely prohibited; and the setting an imposition would have been equally against the genius loci, had lesson-books existed out of which to hear it afterwards. A short imprisonment in an unfurnished room—a not very formidable black-hole—with the loss of a goutte, now and then, and at very long intervals, formed the mild summary of the penal code Pestalozzi. It was Saturday, and a half holiday, when we arrived at Yverdun, and oh the confusion of tongues which there prevailed! All Bedlam and Parnassus let loose to rave together, could not have come up to that diapason of discords with which the high corridors were ringing, as, passing through the throng, we were conducted to the venerable head of the establishment in his private apartments beyond. In this gallery of mixed portraits might be seen long-haired, highborn, and high-cheek-boned Germans; a scantling of French gamins much better dressed; some dark-eyed Italians; Greeks in most foreign attire; here and there a fair ingenuous Russian face; several swart sinister-looking Spaniards, models only for their own Carravagio; some dirty specimens of the universal Pole; one or two unmistakeable English, ready to shake hands with a compatriot; and Swiss from every canton of the Helvetic confederacy. To this promiscuous multitude we were shortly introduced, the kind old man himself taking us by the hand, and acting as master of the ceremonies. When the whole school had crowded round to stare at the new importation, Here, said he, are four English boys come from their distant home, to be naturalised in this establishment, and made members of our family. Boys, receive them kindly, and remember they are henceforth your brothers. A shout from the crowd proclaiming its ready assent and cordial participation in the adoption, nothing remained but to shake hands à l'Anglaise, and to fraternise without loss of time. The next day being Sunday, our skulls were craniologically studied by Herr Schmidt, the head usher;
- 57. and whatever various bumps or depressions phrenology might have discovered thereon were all duly registered in a large book. After this examination was concluded, a week's furlough was allowed, in order that Herr Schmidt might have an opportunity afforded him of seeing how far our real character squared with phrenological observation and measurement, entering this also into the same ledger as a note. What a contrast were we unavoidably drawing all this time between Yverdun and Westminster, and how enjoyable was the change to us! The reader will please to imagine as well as he can, the sensations of a lately pent up chrysalis, on first finding himself a butterfly, or the not less agreeable surprise of some newly metamorphosed tadpole, when, leaving his associates in the mud and green slime, he floats at liberty on the surface of the pool, endowed with lungs and a voice,—if he would at all enter into the exultation of our feelings on changing the penitential air of Millbank for the fresh mountain breezes of the Pays de Vaud. It seemed as if we had—nay, we had actually entered upon a new existence, so thoroughly had all the elements of the old been altered and improved. If we looked back, and compared past and present experiences, there, at the wrong end of the mental telescope, stood that small dingy house, in that little mis-yclept Great Smith Street, with its tiny cocoon of a bedroom, whilom our close and airless prison; here, at the other end, and in immediate contact with the eye, a noble chateau, full of roomy rooms, enough and to spare. Another retrospective peep, and there was Tothill Fields, and its seedy cricket ground; and here, again, a level equally perfect, but carpeted with fine turf, and extending to the margin of a broad living lake, instead of terminating in a nauseous duck-pond; while the cold clammy cloisters adjoining Dean's Yard were not less favourably replaced by a large open airy play-ground, intersected by two clear trout-streams—and a sky as unlike that above Bird-Cage Walk as the interposed atmosphere was different; whilst, in place of the startling, discordant Keleusmata of bargees, joined to the creaking, stunning noise of commerce in a great city, few out-of-door sounds to meet our ear, and these few, with the exception of our own, all quiet, pastoral, and soothing, such as, later in life, make
- 58. Silence in the heart For thought to do her part, and which are not without their charm even to him who whistles as he goes for want of thought. No wonder, then, if Yverdun seemed Paradisaical in its landscapes. Nor was this all. If the views outside were charming, our domestic and social relations within doors were not less pleasing. At first, the unwelcome vision of the late head- master would sometimes haunt us, clad in his flowing black D.D. robes—tristis severitas in vultu, atque in verbis fides, looking as if he intended to flog, and his words never belying his looks. That terrible Olympian arm, raised and ready to strike, was again shadowed forth to view; while we could almost fancy ourselves once more at that judicial table, one of twenty boys who were to draw lots for a hander. How soothingly, then, came the pleasing consciousness, breaking our reverie, that a very different person was now our head-master—a most indulgent old man whom we should meet ere long, with hands uplifted, indeed, but only for the purpose of clutching us tight while he inflicted a salute on both cheeks, and pronounced his affectionate guten morgen, liebes kind, as he hastened on to bestow the like fatherly greeting upon every pupil in turn. THE DORMITORY. The sleeping apartments at the chateau occupied three of the four sides of its inner quadrangle, and consisted of as many long rooms, each with a double row of windows; whereof one looked into the aforesaid quadrangle, while the opposite rows commanded, severally, views of the garden, the open country, and the Grande Place of the town. They were accommodated with sixty uncurtained stump bedsteads, fifty-nine of which afforded gîte to a like number of boys; and one, in no respect superior to the rest, was destined to receive the athletic form of Herr Gottlieb, son-in-law to Vater Pestalozzi, to whose particular charge we were consigned during the
- 59. hours of the night. These bedrooms, being as lofty as they were long, broad, and over-furnished with windows, were always ventilated; but the in-draught of air, which was sufficient to keep them cool during the hottest day in summer, rendered them cold, and sometimes very cold, in the winter. In that season, accordingly, especially when the bise blew, and hail and sleet were pattering against the casements, the compulsory rising to class by candlelight was an ungenial and unwelcome process; for which, however, there being no remedy, the next best thing was to take it as coolly, we were going to say—that of course—but, as patiently as might be. The disagreeable anticipation of the réveil was frequently enough to scare away sleep from our eyes a full hour before the command to jump out of bed was actually issued. On such occasions we would lie awake, and, as the time approached, begin to draw in our own breath, furtively listening, not without trepidation, to the loud nose of a distant comrade, lest its fitful stertor should startle another pair of nostrils, on whose repose that of the whole dormitory depended. Let Æolus and his crew make what tumult they liked inside or outside the castle—they disturbed nobody's dreams—they never murdered sleep. Let them pipe and whistle through every keyhole and crevice of the vast enceinte of the building—sigh and moan as they would in their various imprisonments of attic or corridor; howl wildly round the great tower, or even threaten a forcible entry at the windows, nobody's ears were scared into unwelcome consciousness by sounds so familiar to them all. It was the expectation of a blast louder even than theirs that would keep our eyes open—a blast about to issue from the bed of Herr Gottlieb, and thundering enough, when it issued, to startle the very god of winds himself! Often, as the dreaded six A.M. drew nigh, when the third quarter past five had, ten minutes since, come with a sough and a rattle against the casements, and still Gottlieb slept on, we would take courage, and begin to dream with our eyes open, that his slumbers might be prolonged a little; his face, turned upwards, looked so calm, the eyes so resolutely closed—every feature so perfectly at rest. It could not be more than five minutes to six—might not he who had slept so long, for once oversleep himself? Never! However
- 60. placid those slumbers might be, they invariably forsook our unwearied one just as the clock was on the point of striking six. To judge by the rapid twitchings—they almost seemed galvanic—first of the muscles round the mouth, then of the nose and eyes, it appeared as though some ill-omened dream, at that very nick of time, was sent periodically, on purpose to awaken him; and, if so, it certainly never returned απρακτος. Gottlieb would instantly set to rubbing his eyes, and as the hour struck, spring up wide awake in his shirt sleeves—thus destroying every lingering, and, as it always turned out, ill-founded hope of a longer snooze. Presently we beheld him jump into his small-clothes, and, when sufficiently attired to be seen, unlimber his tongue, and pour forth a rattling broadside—Auf, kinder! schwind!—with such precision of delivery, too, that few sleepers could turn a deaf ear to it. But, lest any one should still lurk under his warm coverlet out of earshot, at the further end of the room, another and a shriller summons to the same effect once more shakes the walls and windows of the dormitory. Then every boy knew right well that the last moment for repose was past, and that he must at once turn out shivering from his bed, and dress as fast as possible; and it was really surprising to witness how rapidly all could huddle on their clothes under certain conditions of the atmosphere! In less than five minutes the whole school was dressed, and Gottlieb, in his sounding shoes, having urged the dilatory with another admonitory schwind, schwind! has departed, key and candle in hand, to arouse the remaining sleepers, by ringing the Great Tom of the chateau. So cold and cheerless was this matutinal summons, that occasional attempts were made to evade it by simulated headach, or, without being quite so specific, on the plea of general indisposition, though it was well known beforehand what the result would be. Herr Gottlieb, in such a case, would presently appear at the bedside of the delinquent patient, with very little compassion in his countenance, and, in a business tone, proceed to inquire from him, Why not up?—and on receiving for reply, in a melancholy voice, that the would-be invalid was sehr krank, would instantly pass the word for the doctor to be summoned. That doctor
- 61. —we knew him well, and every truant knew—was a quondam French army surgeon—a sworn disciple of the Broussais school, whose heroic remedies at the chateau resolved themselves into one of two —i. e., a starve or a vomit, alternately administered, according as the idiosyncracy of the patient, or as this or that symptom turned the scale, now in favour of storming the stomach, now of starving it into capitulation. Just as the welcome hot mess of bread and milk was about to be served to the rest, this dapper little Sangrado would make his appearance, feel the pulse, inspect the tongue, ask a few questions, and finding, generally, indications of what he would term une légère gastrite, recommend diète absolue; then prescribing a mawkish tisane, composed of any garden herbs at hand, and pocketing lancets and stethoscope, would leave the patient to recover sans calomel—a mode of treatment to which, he would tell us, we should certainly have been subjected in our own country. Meanwhile, the superiority of his plan of treatment was unquestionable. On the very next morning, when he called to visit his cher petit malade, an empty bed said quite plainly, Very well, I thank you, sir, and in class. But these feignings were comparatively of rare occurrence; in general, all rose, dressed, and descended together, just as the alarum-bell had ceased to sound; and in less than two minutes more all were assembled in their respective class- rooms. The rats and mice, which had had the run of these during the night, would be still in occupation when we entered; and such was the audacity of these vermin that none cared alone to be the first to plant a candle on his desk. But, by entering en masse, we easily routed the Rodentia, whose forces were driven to seek shelter behind the wainscot, where they would scuffle, and gnaw, and scratch, before they finally withdrew, and left us with blue fingers and chattering teeth to study to make the best of it. Uncomfortable enough was the effort for the first ten minutes of the session; but by degrees the hopes of a possible warming of hands upon the surface of the Dutch stoves after class, if they should have been lighted in time, and at any rate the certainty of a hot breakfast, were entertained, and brought their consolation; besides which, the being up in time to welcome in the dawn of the dullest day, while health
- 62. and liberty are ours, is a pleasure in itself. There was no exception to it here; for when the darkness, becoming every moment less and less dark, had at length given way, and melted into a gray gloaming, we would rejoice, even before it appeared, at the approach of a new day. That approach was soon further heralded by the fitful notes of small day-birds chirping under the leaves, and anon by their sudden dashings against the windows, in the direction of the lights not yet extinguished in the class-rooms. Presently the pigs were heard rejoicing and contending over their fresh wash; then the old horse and the shaggy little donkey in the stable adjoining the styes, knowing by this stir that their feed was coming, snorted and brayed at the pleasant prospect. The cocks had by this time roused their sleepy sultanas, who came creeping from under the barn-door to meet their lords on the dunghill. Our peacock, to satisfy himself that he had not taken cold during the night, would scream to the utmost pitch of a most discordant voice; then the prescient goats would bleat from the cabins, and plaintively remind us that, till their door is unpadlocked, they can get no prog; then the punctual magpie, and his friend the jay, having hopped all down the corridor, would be heard screaming for broken victuals at the school-room door, till our dismissal bell, finding so many other tongues loosened, at length wags its own, and then for the next hour and a half all are free to follow their own devices. Breakfast shortly follows; but, alas! another cold ceremony must be undergone first. A preliminary visit to pump court, and a thorough ablution of face and hands, is indispensable to those who would become successful candidates for that long- anticipated meal. This bleaching process, at an icy temperature, was never agreeable; but when the pipes happened to be frozen—a contingency by no means unfrequent—and the snow in the yard must be substituted for the water which was not in the pump, it proved a difficult and sometimes a painful business; especially as there was always some uncertainty afterwards, whether the chilblained paws would pass muster before the inspector-general commissioned to examine them—who, utterly reckless as to how the boys might be off for soap, and incredulous of what they would fain attribute to the adust complexion of their skin, would require to
- 63. have that assertion tested by a further experiment at the pump head. THE REFECTORY. Forbear to scoff at woes you cannot feel, Nor mock the misery of a stinted meal.—Crabbe. The dietary tables at the chateau, conspicuous alike for the paucity and simplicity of the articles registered therein, are easily recalled to mind. The fare they exhibited was certainly coarse—though, by a euphemism, it might have been termed merely plain—and spare withal. The breakfast would consist of milk and water—the first aqueous enough without dilution, being the produce of certain ill- favoured, and, as we afterwards tasted their flesh, we may add ill- flavoured kine, whose impoverished lacteals could furnish out of their sorry fodder no better supplies. It was London sky-blue, in short, but not of the Alderney dairy, which was made to serve our turn at Yverdun. This milk, at seven in summer, and at half-past seven in winter, was transferred boiling, and as yet unadulterated, into earthenware mixers, which had been previously half-filled with hot water from a neighbouring kettle. In this half-and-half state it was baled out for the assembled school into a series of pewter platters, ranged along the sides of three bare deal boards, some thirty feet long by two wide, and mounted on tressels, which served us for tables. The ministering damsels were two great German Fraus, rejoicing severally in the pleasing names of Gretchen and Bessie. When Frau Gretchen, standing behind each boy, had dropt her allowance of milk over his right shoulder—during which process there was generally a mighty clatter for full measure and fair play— the other Frau was slicing off her slices of bread from a brown loaf a yard long, which she carried under her arm, and slashed clean through with wonderful precision and address. It was now for all those who had saved pocket-money for menus-plaisirs to produce their cornets of cinnamon or sugar, sprinkle a little into the milk, and
- 64. then fall to sipping and munching with increased zest and satisfaction. So dry and chaffy was our pain de ménage that none ventured to soak it entire, or at once, but would cut it into frustrums, and retain liquid enough to wash down the boluses separately. In a few minutes every plate was completely cleaned out and polished; and the cats, that generally entered the room as we left it, seldom found a drop with which they might moisten their tongues, or remove from cheeks and whiskers the red stains of murdered mice on which they had been breaking their fast in the great tower. So much for the earliest meal of the day, which was to carry us through five hours, if not of laborious mental study, at least of the incarceration of our bodies in class, which was equally irksome to them as if our minds had been hard at work. These five hours terminated, slates were once more insalivated and put by clean, and the hungry garrison began to look forward to the pleasures of the noon-day repast. The same bell that had been calling so often to class would now give premonitory notice of dinner, but in a greatly changed tone. In place of the shrill snappish key in which it had all the morning jerked out each short unwelcome summons from lesson to lesson, as if fearful of ringing one note beyond the prescribed minute, it now would take time, vibrate far and wide in its cage, give full scope to its tongue, and appear, from the loud increasing swell of its prolonged oyez, to announce the message of good cheer like a herald conscious and proud of his commission. Ding-dong!—come along! Dinner's dishing!—ding-dong! Da capo and encore! Then, starting up from every school-room form throughout the chateau, the noisy boys rushed pell-mell, opened all the doors, and, like emergent bees in quest of honey, began coursing up and down right busily between the salle-à-manger and the kitchen—snuffing the various aromas as they escaped from the latter into the passage, and inferring from the amount of exhaled fragrance the actual progress of the preparations for eating. Occasionally some sly Tom would peep into the kitchen, while the Fraus were too busy to notice him, and watch the great cauldron that had been milked dry of its stores in the morning, now discharging its aqueous contents of a much-attenuated bouillon—the
- 65. surface covered with lumps of swimming bread, thickened throughout with a hydrate of potatoes, and coloured with coarse insipid carrots, which certainly gave it a savoury appearance. It was not good broth—far from it, for it was both sub-greasy and super- salted; but then it was hot, it was thick, and there was an abundant supply. It used to gush, as we have said, from the great stop-cock of the cauldron, steaming and sputtering, into eight enormous tureens. The shreds of beef, together with whatever other solids remained behind after the fluid had been drawn off, were next fished up from the abyss with long ladles, and plumped into the decanted liquor. The young gastronome who might have beheld these proceedings would wait till the lid was taken off the sauerkraut; and then, the odour becoming overpoweringly appetising, he would run, as by irresistible instinct, into the dining-room, where most of the boys were already assembled, each with a ration of brown bread in his hand, and ready for the Fraus, who were speedily about to enter. The dinner was noisy and ungenteel in the extreme—how could it be otherwise? ventre affamé n'a point d'oreilles. Hardly was the German grace concluded, and the covers removed, when that bone of contention, the marrow bone, was caught up by some big boy near the top of the table, and became the signal for a general row. All in his neighbourhood would call out second, third, fourth, fifth, c., for said bone; and thus it would travel from plate to plate, yielding its contents freely to the two or three first applicants, but wholly inadequate—unless it could have resolved itself altogether into marrow—to meet all the demands made upon its stores. Then arose angry words of contention, which waxed hot as the marrow waxed cold, every candidate being equally vociferous in maintaining the priority of his particular claim. Earnest appeals in German, French, Spanish, English, c., were bandied from one to the other in consequence, as to who had really said après toi first! At last the dry bone was found undeserving of further contention; and, ceasing to drop any more fatness upon any boy's bread, the competition for it was dropt too. When now we had half-filled our stomachs with a soup which few physicians would have withheld from their fever patients on the score of its strength, we threw in a
- 66. sufficiency of bread and sauerkraut to absorb it; and, after the post- prandial German grace had been pronounced, the boys left the table, generally with a saved crust in their pockets, to repair to the garden and filch—if it was filching—an alliaceous dessert from the beds, which they washed in the clear stream, and added, without fear of indigestion, to the meal just concluded within the chateau. Most of us throve upon this Spartan diet; but some delicate boys, unendowed with the ostrich power of assimilation usual at that period—for boys, like ostriches, can digest almost anything—became deranged in their chylopoietics, and continued to feel its ill effects in mesenteric and other chronic ailments for years afterwards. An hour was given for stomachs to do their work, before we reassembled to ours in the class-room. At half-past four precisely, a gouté, was served out, which consisted of a whacking slice of bread, and either a repetition of the morning's milk and water, or café au lait, (without sugar bien entendu,) or twenty-five walnuts, or a couple of ounces of strong-tasted gruyère, or a plateful of schnitz (cuttings of dried apples, pears, and plums). We might choose any one of these several dainties we liked, but not more. Some dangerous characters —not to be imitated—would occasionally, while young Frau Schmidt stood doling out the supplies from her cupboard among the assembled throng, make the disingenuous attempt to obtain cheese with one hand and schnitz with the other. But the artifice, we are happy to say, seldom succeeded; for that vigilant lady, quick-eyed and active, and who, of all things, hated to be imposed upon, would turn round upon the false claimant, and bid him hold up both his hands at once—which he, ambidexter as he was, durst not do, and thus he was exposed to the laughter and jeers of the rest. At nine, the bell sounded a feeble call to a soi-disant supper; but few of us cared for a basin of tisane under the name of lentil soup—or a pappy potato, salted in the boiling—and soon after we all repaired to our bedrooms—made a noise for a short time, then undressed, and were speedily asleep under our duvets, and as sound, if not as musical, as tops.
- 67. Our common fare, as the reader has now seen, was sorry enough; but we had our Carnival and gala days as well as our Lent. Vater Pestalozzi's birthday, in summer, and the first day of the new year, were the most conspicuous. On each of these occasions we enjoyed a whole week's holiday; and as these were also the periods for slaughtering the pigs, we fed (twice a-year for a whole week!) upon black puddings and pork à discretion, qualified with a sauce of beetroot and vinegar, and washed down with a fluid really like small- beer. CLASSES. The school-rooms, which lay immediately under the dormitories on the ground-floor, consisted of a number of detached chambers, each of which issued upon a corridor. They were airy—there was plenty of air at Yverdun—and lofty as became so venerable a building; but they were unswept, unscrubbed, peeled of their paint, and, owing to the little light that could find its way through two very small windows punched out of the fortress walls, presented, save at mid-day, or as the declining sun illumined momentarily the dark recess, as comfortless a set of interiors as you could well see. It required, indeed, all the elasticity of youth to bear many hours' daily incarceration in such black-holes, without participating in the pervading gloom. Such dismal domiciles were only fit resorts for the myoptic bat, who would occasionally visit them from the old tower; for the twilight horde of cockroaches, which swarmed along the floor, or the eight-eyed spiders who colonised the ceiling. The tender sight, too, of a patient just recovering from ophthalmia would here have required no factitious or deeper shade—but merits like these only rendered them as ungenial as possible to the physiology and feelings of their youthful occupants. If these apartments looked gloomy in their dilapidations and want of sun, the sombre effect was much heightened by the absence of the ordinary tables and chairs, and whatever else is necessary to give a room a habitable appearance. Had an appraiser been commissioned to make out a
- 68. complete list of the furniture and the fixtures together, a mere glance had sufficed for the inventory. In vain would his practised eye have wandered in quest of themes for golden sentences, printed in such uncial characters that all who run may read; in vain for the high-hung well-backed chart, or for any pleasing pictorial souvenirs of Æsop or the Ark—neither these nor the long coloured Stream of Time, nor formal but useful views in perspective, adorned our sorry walls. No old mahogany case clicked in a corner, beating time for the class, and the hour up-striking loud that it should not be defrauded of its dues. No glazed globe, gliding round on easy axis, spun under its brassy equator to the antipodes on its sides being touched. No bright zodiac was there to exhibit its cabalistic figures in pleasing arabesques. In place of these and other well-known objects, here stood a line of dirty, much-inked desks, with an equally dirty row of attendant forms subjacent alongside. There was a scantling—it seldom exceeded a leash—of rickety rush-bottom chairs distributed at long intervals along the walls; a coal-black slate, pegged high on its wooden horse; a keyless cupboard, containing the various implements of learning, a dirty duster, a pewter plate with cretaceous deposits, a slop-basin and a ragged sponge;—and then, unless he had included the cobwebs of the ceiling, (not usually reckoned up in the furniture of a room,) no other movables remained. One conspicuous fixture, however, there was, a gigantic Dutch stove. This lumbering parallelogram, faggot-fed from the corridor behind, projected several feet into the room, and shone bright in the glaze of earthenware emblazonments. Around it we would sometimes congregate in the intervals of class: in winter to toast our hands and hind quarters, as we pressed against the heated tiles, with more or less vigour according to the fervency of the central fire; and in summer either to tell stories, or to con over the pictorial History of the Bible, which adorned its frontispiece and sides. We cannot say that every square exactly squared with even our schoolboy notions of propriety in its mode of teaching religious subjects; there was a Dutch quaintness in the illustrations, which would sometimes force a smile from its simplicity, at others shock, from its apparent want of decorum and reverence. Preeminent of
- 69. course among the gems from Genesis, Adam and Eve, safe in innocency and naked truth, here walked unscathed amidst a menagerie of wild beasts—there, dressed in the costume of their fall, they quitted Eden, and left it in possession of tigers, bears, and crocodiles. Hard by on a smaller tile, that brawny knave of clubs, Cain, battered down his brother at the altar; then followed a long picture-gallery of the acts of the patriarchs, and another equally long of the acts of the apostles. But, queer as many of these misconceptions might seem, they were nothing to the strange attempts made at dramatising the parables of the New Testament— e. g. a stout man, staggering under the weight of an enormous beam which grows out of one eye, employs his fingers, assisted by the other, to pick out a black speck from the cornea of his neighbour. Here, an unclean spirit, as black as any sweep, issues from the mouth of his victim, with wings and a tail! Here again, the good Samaritan, turbaned like a Turk, is bent over the waylaid traveller, and pours wine and oil into his wounds from the mouths of two Florence flasks; there, the grain of mustard-seed, become a tree, sheltering already a large aviary in its boughs; the woman, dancing a hornpipe with the Dutch broom, has swept her house, and lo! the piece of silver that was lost in her hand; a servant, who is digging a hole in order to hide his lord's talent under a tree, is overlooked by a magpie and two crows, who are attentive witnesses of the deposit:— and many others too numerous to mention. So much for the empty school-room, but what's a hive without bees, or a school-room without boys? The reader who has peeped into it untenanted, shall now, if he pleases, be introduced, dum fervet opus full and alive. Should he not be able to trace out very clearly the system at work, he will at least be no worse off than the bee-fancier, who hears indeed the buzzing, and sees a flux and reflux current of his winged confectioners entering in and passing out, but cannot investigate the detail of their labours any farther. In the Yverdun, as in the hymenopterus apiary, we swarmed, we buzzed, dispersed, reassembled at the sound of the bell, flocked in and flocked out, all the day long; exhibited much restlessness and activity, evincing that something was going on, but what, it would have been hard to
- 70. determine. Here the comparison must drop. Bees buzz to some purpose; they know what they are about; they help one another; they work orderly and to one end,— How skilfully they build the cell, How neat they spread the wax, And labour hard to store it well With the sweet food, c. c. In none of these particulars did we resemble the busy bee. This being admitted, our object in offering a few words upon the course of study pursued at the chateau is not with any idea of enlightening the reader as to anything really acquired during the long ten hours' session of each day; but rather to show how ten hours' imprisonment may be inflicted upon the body for the supposed advantage of the mind, and yet be consumed in profitless labour, and diligence which maketh not rich; to prove, by an exhibition of their opposites, that method and discipline are indispensable in tuition, and (if he will accept our pathemata for his mathemata and guides in the bringing up of his sons) to convince him that education, like scripture, admits not of private interpretation. Those who refuse to adopt the Catholic views of the age, and the general sense of the society in which they live, must blame themselves if they find the experiment of foreign schools a failure, and that they have sent their children farther to fare worse. And now to proceed to the geography class, which was the first after breakfast, and began at half-past eight. As the summons-bell sounded, the boys came rushing and tumbling in, and ere a minute had elapsed were swarming over, and settling upon, the high reading-desks: the master, already at his work, was chalking out the business of the hour; and as this took some little time to accomplish, the youngsters, not to sit unemployed, would be assiduously engaged in impressing sundry animal forms—among which the donkey was a favourite—cut out in cloth, and well powdered, upon one another's backs. When Herr G—— had finished his chalkings,
- 71. and was gone to the corner of the room for his show-perch, a skeleton map of Europe might be seen, by those who chose to look that way, covering the slate: this, however, was what the majority of the assembly never dreamt of, or only dreamt they were doing. The class generally—though ready when called upon to give the efficient support of their tongues—kept their eyes to gape elsewhere, and, like Solomon's fool, had them where they had no business to be. The map, too often repeated to attract from its novelty, had no claim to respect on other grounds. It was one of a class accurately designated by that careful geographer, old Homer, as μαπς ου Κατα Κοσμον. Coarse and clumsy, however, as it necessarily would be, it might still have proved of service had the boys been the draughtsmen. As it was, the following mechanically Herr G——'s wand to join in the general chorus of the last census of a city, the perpendicular altitude of a mountain, or the length and breadth of a lake, could obviously convey no useful instruction to any one. But, useful or otherwise, such was our regime,—to set one of from fifty to sixty lads, day after day, week after week, repeating facts and figures notorious to every little reader of penny guides to science, till all had the last statistical returns at their tongue's tip; and knew, when all was done, as much of what geography really meant as on the day of their first matriculation. Small wonder, then, if some should later have foresworn this study, and been revolted at the bare sight of a map! All our recollections of map, unlike those of personal travel, are sufficiently distasteful. Often have we yawned wearily over them at Yverdun, when our eyes were demanded to follow the titubations of Herr G——'s magic wand, which, in its uncertain route, would skip from Europe to Africa and back again— qui modo Thebas modo me ponit Athenis; and our dislike to them since has increased amazingly. Does the reader care to be told the reason of this? Let him—in order to obtain the pragmatic sanction of some stiff-necked examiner—have to get up all the anastomosing routes of St Paul's several journeyings; have to follow those rebellious Israelites in all their wanderings through the desert; to draw the line round them when in Palestine; going from Dan to Beersheba, and meting out the valley of Succoth; or, finally, have
- 72. to cover a large sheet of foolscap with a progressive survey of the spread of Christianity during the three first centuries—and he will easily enter into our feelings. To return to the class-room: The geographical lesson, though of daily infliction, was accurately circumscribed in its duration. Old Time kept a sharp look-out over his blooming daughters, and never suffered one hour to tread upon the heels or trench upon the province of a sister hour. Sixty minutes to all, and not an extra minute to any, was the old gentleman's impartial rule; and he took care to see it was strictly adhered to. As the clock struck ten, geography was shoved aside by the muse of mathematics. A sea of dirty water had washed out in a twinkling all traces of the continent of Europe, and the palimpsest slate presented a clean face for whatever figures might next be traced upon it. The hour for Euclidising was arrived, and anon the black parallelogram was intersected with numerous triangles of the Isosceles and Scalene pattern; but, notwithstanding this promising début, we did not make much quicker progress here than in the previous lesson. How should we, who had not only the difficulties inseparable from the subject to cope with, but a much more formidable difficulty—viz. the obstruction which we opposed to each other's advance, by the plan, so unwisely adopted, of making all the class do the same thing, that they might keep pace together. It is a polite piece of folly enough for a whole party to be kept waiting dinner by a lounging guest, who chooses to ride in the park when he ought to be at his toilet; but we were the victims of a much greater absurdity, who lost what might have proved an hour of profitable work, out of tenderness to some incorrigibly idle or Bœotian boy, who could not get over the Pons Asinorum, (every proposition was a pons to some asinus or other,) and so made those who were over stand still, or come back to help him across. Neither was this, though a very considerable drawback, our only hindrance—the guides were not always safe. Sometimes he who acted in that capacity would shout Eureka too soon; and having undertaken to lead the van, lead it astray till just about, as he supposed, to come
- 73. down upon the proof itself, and to come down with a Q. E. D.: the master would stop him short, and bid him—as Coleridge told the ingenious author of Guesses at Truth—to guess again. But suppose the guess fortunate, or that a boy had even succeeded, by his own industry or reflection, in mastering a proposition, did it follow that he would be a clear expositor of what he knew? It was far otherwise. Our young Archimedes—unacquainted with the terms of the science, and being also (as we have hinted) lamentably defective in his knowledge of the power of words—would mix up such a farrago of irrelevancies and repetitions with the proof, as, in fact, to render it to the majority no proof at all. Euclid should be taught in his own words,—just enough and none to spare: the employment of less must engender obscurity; and of more, a want of neatness and perspicacity. The best geometrician amongst us would have cut but a bad figure by the side of a lad of very average ability brought up to know Euclid by book. Another twitch of the bell announced that the hour for playing at triangles had expired. In five minutes the slate was covered with bars of minims and crotchets, and the music lesson begun. This, in the general tone of its delivery, bore a striking resemblance to the geographical one of two hours before; the only difference being that ut, re, me had succeeded to names of certain cities, and fa, so, la to the number of their inhabitants. It would be as vain an attempt to describe all the noise we made as to show its rationale or motive. It was loud enough to have cowed a lion, stopped a donkey in mid-bray—to have excited the envy of the vocal Lablache, or to have sent any prima donna into hysterics. When this third hour had been bellowed away, and the bell had rung unheard the advent of a fourth—presto—in came Mons. D——, to relieve the meek man who had acted as coryphæus to the music class; and after a little tugging, had soon produced from his pocket that without which you never catch a Frenchman—a thème. The theme being announced, we proceeded (not quite tant bien que mal) to scribble it down at his dictation, and to amend its orthography afterwards from a corrected copy on the slate. Once more the indefatigable bell obtruded its
- 74. tinkle, to proclaim that Herr Roth was coming with a Fable of Gellert, or a chapter from Vater Pestalozzi's serious novel, Gumal und Lina, to read, and expound, and catechise upon. This last lesson before dinner was always accompanied by frequent yawns and other unrepressed symptoms of fatigue; and at its conclusion we all rose with a shout, and rushed into the corridors. On resuming work in the afternoon, there was even less attention and method observed than before. The classes were then broken up, and private lessons were given in accomplishments, or in some of the useful arts. Drawing dogs and cows, with a master to look after the trees and the hedges; whistling and spitting through a flute; playing on the patience of a violin; turning at a lathe; or fencing with a powerful maître d'armes;—such were the general occupations. It was then, however, that we English withdrew to our Greek and Latin; and, under a kind master, Dr M——, acquired (with the exception of a love for natural history, and a very unambitious turn of mind) all that really could deserve the name of education. We have now described the sedentary life at the chateau. In the next paper the reader shall be carried to the gymnasium; the drill ground behind the lake; to our small menageries of kids, guinea pigs, and rabbits; be present at our annual ball and skating bouts in winter, and at our bathings, fishings, frog-spearings, and rambles over the Jura in summer. FOOTNOTES: [14] Cicero, De Fin., ii. 1.
- 75. THE CROWNING OF THE COLUMN, AND CRUSHING OF THE PEDESTAL. It was said in the debate on the Navigation Laws, in the best speech made on the Liberal side, by one of the ablest of the Liberal party, that the repeal of the Navigation Laws was the crowning of the column of free trade. There is no doubt it was so; but it was something more. It was not only the carrying out of a principle, but the overthrow of a system; it was not merely the crowning of the column, but the crushing of the pedestal. And what was the system which was thus completely overthrown, for the time at least, by this great triumph of Liberal doctrines? It was the system under which England had become free, and great, and powerful; under which, in her alone of all modern states, liberty had been found to coexist with law, and progress with order; under which wealth had increased without producing divisions, and power grown up without inducing corruption; the system which had withstood the shocks of two centuries, and created an empire unsurpassed since the beginning of the world in extent and magnificence. It was a system which had been followed out with persevering energy by the greatest men, and the most commanding intellects, which modern Europe had ever produced; which was begun by the republican patriotism of Cromwell, and consummated by the conservative wisdom of Pitt; which had been embraced alike by Somers and Bolingbroke, by Walpole and Chatham, by Fox and Castlereagh; which, during two centuries, had produced an unbroken growth of national strength, a ceaseless extension of national power, and at length reared up a dominion which embraced the earth in its grasp, and exceeded anything ever achieved by the legions of Cæsar, or the phalanx of Alexander. No vicissitudes of time, no shock of adverse fortune, had been able permanently to
- 76. arrest its progress. It had risen superior alike to the ambition of Louis XIV. and the genius of Napoleon; the rude severance of the North American colonies had thrown only a passing shade over its fortunes; the power of Hindostan had been subdued by its force, the sceptre of the ocean won by its prowess. It had planted its colonies in every quarter of the globe, and at once peopled with its descendants a new hemisphere, and, for the first time since the creation, rolled back to the old the tide of civilisation. Perish when it may, the old English system has achieved mighty things; it has indelibly affixed its impress on the tablets of history. The children of its creation, the Anglo-Saxon race, will fill alike the solitudes of the Far West, and the isles of the East; they will be found equally on the shores of the Missouri, and on the savannahs of Australia; and the period can already be anticipated, even by the least imaginative, when their descendants will people half the globe. It was not only the column of free trade which has been crowned in this memorable year. Another column, more firm in its structure, more lasting in its duration, more conspicuous amidst the wonders of creation, has, in the same season, been crowned by British hands. While the sacrilegious efforts of those whom it had sheltered were tearing down the temple of protection in the West, the last stone was put to the august structure which it had reared in the East. The victory of Goojerat on the Indus was contemporary with the repeal of the Navigation Laws on the Thames. The completion of the conquest of India occurred exactly at the moment when the system which had created that empire was repudiated. Protection placed the sceptre of India in our hands, when free trade was surrendering the trident of the ocean in the heart of our power. With truth did Lord Gough say, in his noble proclamation to the army of the Punjaub, on the termination of hostilities, that what Alexander had attempted they had done. Supported by the energy of England, guided by the principles of protection, restrained by the dictates of justice, backed by the navy which the Navigation Laws had created, the British arms had achieved the most wonderful triumph recorded in the annals of mankind. They had subjugated a hundred and forty millions of men
- 77. in the Continent of Hindostan, at the distance of ten thousand miles from the parent state; they had made themselves felt alike, and at the same moment, at Nankin, the ancient capital of the Celestial Empire, and at Cabool, the cradle of Mahommedan power. Conquering all who resisted, blessing all who submitted, securing the allegiance of the subjects by the justice and experienced advantages of their government, they had realised the boasted maxim of Roman administration— Parcere subjectis et debellare superbos, and steadily advanced through a hundred years of effort and glory, not unmixed with disaster, from the banks of the Hoogley to the shores of the Indus—from the black hole of Calcutta to the throne of Aurengzebe. Nulla magna civitas, said Hannibal, diu quiescere potest—si foris hostem non habet, domi invenit: ut praevalida corpora ab externis causis tuta videntur, suis ipsis viribus conficiuntur.[15] When the Carthaginian hero made this mournful reflection on the infatuated spirit which had seized his own countrymen, and threatened to destroy their once powerful dominion, he little thought what a marvellous confirmation of it a future empire of far greater extent and celebrity was to afford. That the system of free trade—that is, the universal preference of foreigners, for the sake of the smallest reduction of price, to your own subjects—must, if persisted in, lead to the dismemberment and overthrow of the British empire, cannot admit of a moment's doubt, and will be amply proved to every unbiassed reader in the sequel of this paper. Yet the moment chosen for carrying this principle into effect was precisely that, when the good effects of the opposite system had been most decisively demonstrated, and an empire unprecedented in magnitude and magnificence had reached its acme under its shadow. It would be impossible to explain so strange an anomaly, if we did not recollect how wayward and irreconcilable are the changes of the human mind: that action and reaction is the law not less of the moral than
- 78. of the material world; that nations become tired of hearing a policy called wise, not less than an individual called the just; and that if a magnanimous and truly national course of government has been pursued by one party long in possession of power, this is quite sufficient to make its opponents embrace the opposite set of tenets, and exert all their influence to carry them into effect when they succeed to the direction of affairs, without the slightest regard to the ruin they may bring on the national fortunes. The secret of the long duration and unexampled success of the British national policy is to be found in the protection which it afforded to all the national interests. But for this, it must long since have been overthrown, and with it the empire which was growing up under its shadow. No institutions or frames of government can long exist which are not held together by that firmest of bonds, experienced benefits. What made the Roman power steadily advance during seven centuries, and endure in all a thousand years? The protection which the arms of the legions afforded to the industry of mankind, the international wars which they prevented, the general peace they secured, the magnanimous policy which admitted the conquered states to the privileges of Roman citizens, and caused the Imperial government to be felt through the wide circuit of its power, only by the vast market it opened to the industry of its multifarious subjects, and the munificence with which local undertakings were everywhere aided by the Imperial treasury. Free trade in grain at length ruined it: the harvests of Libya and Egypt came to supersede those of Greece and Italy,—and thence its fall. To the same cause which occasioned the rise of Rome, is to be ascribed the similar unbroken progress of the Russian territorial dominion, and that of the British colonial empire in modern times. What, on the other hand, caused the conquests of Timour and Charlemagne, Alexander the Great and Napoleon, to be so speedily obliterated, and their vast empires to fall to pieces the moment the powerful hand which had created them was laid in the dust? The want of protection to general interests, the absence of the strong bond of experienced benefits; the oppressive nature of the conquering
- 79. Welcome to our website – the perfect destination for book lovers and knowledge seekers. We believe that every book holds a new world, offering opportunities for learning, discovery, and personal growth. That’s why we are dedicated to bringing you a diverse collection of books, ranging from classic literature and specialized publications to self-development guides and children's books. More than just a book-buying platform, we strive to be a bridge connecting you with timeless cultural and intellectual values. With an elegant, user-friendly interface and a smart search system, you can quickly find the books that best suit your interests. Additionally, our special promotions and home delivery services help you save time and fully enjoy the joy of reading. Join us on a journey of knowledge exploration, passion nurturing, and personal growth every day! ebookbell.com