The Image Processing Handbook

 The Image Processing Handbook

Third Edition by John C. Russ - CRC Press

 Table of Contents

 Introduction

 Acknowledgments

Chapter 1—Acquiring Images

 Human reliance on images for information

 Using video cameras to acquire images

 Electronics and bandwidth limitations

 High resolution imaging

 Color imaging

 Digital cameras

 Color spaces

 Color displays Image types

 Range imaging

 Multiple images

 Stereoscopy

 Imaging requirements

Chapter 2—Printing and Storage

 Printing Dots on paper

 Color printing Printing hardware

 Film recorders

 File storage

 Optical storage media

 Magnetic recording

 Databases for images

 Browsing and thumbnails

 Lossless coding

 Color palettes

 Lossy compression

 Other compression methods

 Digital movies

Chapter 3—Correcting Imaging Defects

 Noisy images

 Neighborhood averaging

 Neighborhood ranking

 Other neighborhood noise reduction methods

 Maximum entropy

 Contrast expansion

 Nonuniform illumination

 Fitting a background function

 Rank leveling

 Color shading

 Nonplanar views

 Computer graphics

 Geometrical distortion

 Alignment

 Morphing

Chapter 4—Image Enhancement

 Contrast manipulation

 Histogram equalization

 Laplacian Derivatives

 The Sobel and Kirsch operators

 Rank operations

 Texture Fractal analysis

 Implementation notes

 Image math

 Subtracting images

 Multiplication and division

Chapter 5—Processing Images in Frequency Space

 Some necessary mathematical preliminaries

 What frequency space is all about

 The Fourier transform Fourier transforms of real functions

 Frequencies and orientations

 Measuring images in the frequency domain

 Orientation and spacing

 Preferred orientation

 Texture and fractals Filtering images Isolating periodic noise

 Masks and filters

 Selection of periodic information

 Convolution and correlation

 Fundamentals of convolution Imaging system characteristics

 Removing motion blur and other defects

 Template matching and correlation

 Autocorrelation

 Conclusion

Chapter 6—Segmentation and Thresholding Thresholding

 Multiband images

 Two- Dimensional thresholds

 Multiband thresholding

 Thresholding from texture

 Multiple thresholding criteria

 Textural orientation

 Accuracy and reproducibility

 Including position information

 Selective histograms

 Boundary lines

 Contours Image representation

 Other segmentation methods

 The general classification problem

Chapter 7—Processing Binary Images

 Boolean operations

 Combining Boolean operations

 Masks From pixels to features

 Boolean logic with features

 Selecting features by location

 Double thresholding

 Erosion and dilation

 Opening and closing Isotropy

 Measurements using erosion and dilation

 Extension to grey scale images

 Coefficient and depth parameters

 Examples of use The custer

 Skeletonization Boundary lines and thickening

 Euclidean distance map

 Watershed segmentation

 Ultimate eroded points

 Fractal dimension measurement

 Medial axis transform

 Cluster analysis

Chapter 8—Image Measurements

 Brightness measurements

 Determining location

 Orientation Neighbor relationships

 Alignment Counting features

 Special counting procedures

 Feature size

 Caliper dimensions

 Perimeter Ellipse fitting Describing shape

 Fractal dimension

 Harmonic analysis

 Topology Feature identification

 Three- Dimensional measurements

Chapter 9—3D Image Acquisition

 Volume imaging vs. sections

 Basics of reconstruction

 Algebraic reconstruction methods

 Maximum entropy

 Defects in reconstructed images

 Imaging geometries

 Three dimensional tomography

 High resolution tomography

Chapter 10—3D Image Visualization

 Sources of 3D data

 Serial sections Optical sectioning

 Sequential removal

 Stereo 3D data sets

 Slicing the data set

 Arbitrary section planes

 The use of color

 Volumetric display

 Stereo Viewing Special display hardware

 Ray tracing

 Reflection Surfaces Multiply connected surfaces

 Image processing in 3D Measurements on 3D images

 Conclusion

Chapter 11—Imaging Surfaces

 Producing surfaces

 Devices that image surfaces by physical contact

 Noncontacting measurements

 Microscopy of surfaces

 Surface composition imaging

 Processing of range images

 Processing of composition maps

 Data presentation and visualization

 Render- Ing and visualization

 Analysis of surface data

 Profile measurements

 The Birmingham measurement suite

 New approaches topographic analysis and fractal dimensions

 References

 Index

 Introduction

 Image processing is used for two somewhat differ- Ent purposes:

 a) Improving the visual appearance of images to a human viewer, and

 b) Preparing images for measurement of the features and structures present.

 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 purposes. To do the best possible job, it helps to know about the uses to which the processed 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.

 It also is useful to know about the printing process, since many images are processed in the context of reproduction or transmission. The measurement of images generally requires that features be well defined, either by edges or unique (and uniform) Brightness or color, texture, or some combination of these factors. The types of measurements that will be performed on entire scenes or individual features are important in determining the appropriate processing steps. 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 may be more appealing to the senses, and some may convey more meaning, but these two criteria may not be identical nor use identical methods. This handbook presents an extensiive collection of image processing tools, so that the user of computerbased systems can both understand those methods provided in packaged software, and program those additions which may be needed for particular applications.

 Comparisons are presented of different algorithms that may be used for similar purposes, using a selection of representative pictures from light and electron microscopes, as well as macroscopic, satellite and astronomical images. In revising the book for this new edition, I have 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 edition, and now there is a major new section on the important subject of surface images which includes both processing and measurement.

 The sections on the everadvancing hardware for image capture and printing have been expanded and information added on the newest technologies. More examples have been added in every chapter, and the reference list expanded and brought up to date. However, I have resisted suggestions to put “more of the math” into the book. There are excellent texts on image processing, compression, mathematical morphology, 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. Few people learn the principles of image processing from the equations. Just as we use images to “do science,” so most of us use images to learn about many things, including imaging itself. The hope is that by seeing 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.

 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 determine 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 particularly requires that you become a very analyt- Ical observer of images.

 If you can learn to see what the computer sees, you will become a better viewer and obtain the best possible images, suitable for further processing. To facilitate this hands- On learning process, I have collaborated with my son, Chris Russ, to produce a CD- ROM that can be used as a companion to this book. The Image Processing Tool Kit contains more than 200 images, many of them the examples from this book, plus over 100 Photoshop- Compatible plugins that implement many of the algorithms discussed here. These can be used with Adobe Photoshop® or any of the numerous programs (some of them free) That implement the Photoshop plugin interface, on either Macintosh or Windows computers.

 Many of the images were acquired directly from various microscopes and other sources using color or monochrome video cameras and digitized directly into the computer. Others were digitized using a digital camera (Polaroid DMC), and some were obtained using a 24- Bit color scanner, often from images supplied by many coworkers and researchers.

 These are acknowledged wherever the origin of an image could be determined. A few examples, taken from the literature, are individually referenced. The book was produced by directly making colorseparated films with an imagesetter without intermediate hard copy, negatives or prints of the images, etc.

 Amongst other things, this means that the author must bear full responsibility for any errors, since no traditional typesetting was involved. (It has also forced me to learn more than I ever hoped to know about some aspects of this 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.

 Special thanks are due to Chris Russ (Reindeer Games, Inc. , Asheville, NC) Who has helped to program many of these algorithms and contributed invaluable com- Ments, and to Helen Adams, who has proofread many pages, endured many discussions, and provided the moral support that makes writing projects such as this possible.

Chapter 1 Acquiring Images

 Human reliance on images for information Humans are primarily visual creatures. Not all animals depend on their eyes, as we do, for 99% or more of the information received about their surroundings. Bats use high frequency sound, cats have poor vision but a rich sense of smell, snakes locate prey by heat emission, and fish have organs that sense (and in some cases generate) Electrical fields. Even birds, which are highly visual, do not have our eye configuration.

 Their eyes are on opposite sides of their heads, providing nearly 360- Degree coverage but little in the way of stereopsis, and they have four or five different color receptors (we have three, loosely described as red, green, and blue). It is difficult to imagine what the world “looks like” to such animals.

 Even the word “imagine” contains within it our bias towards images, as does much of our language. People with vision defects wear glasses because of their dependence on this sense. We tolerate considerable hearing loss before resorting to a hearing aid and there are, practically speaking, no prosthetic devices for the other senses. This bias in everyday life extends to how we pursue more technical goals as well. Scientific instruments commonly produce images to communicate results to the operator, rather than generating an audible tone or emitting a smell.

 Space missions to other planets and Comet Halley always include cameras as major components, and we judge the success of those missions by the quality of the images returned.

 This suggests a few of the ways in which we have extended the range of our natural vision. Simple 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 wave- Lengths between about 400 and 700 nanometers) We now have sensors capable of detecting infrared and ultraviolet light, X- Rays, and radio waves. Figure 1 shows an example, an image presenting radio telescope data in the form of an image in which grey scale brightness represents radio intensity.

 Such devices and presentations are used to further extend our imaging capability.

 Figure 1 Radio astronomy produces images such as this view of NGC1265. These are offen displayed with false colors to emphasize subtle variations in signal brightness. Signals other than electromagnetic radiation can be used to produce images as well. 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 2 shows an acoustic microscope image of a composite material. Figure 2 Scanning acoustic microscope image (with superimposed signal profile along one scan line) Of a polished cross section through a composite. The central white feature is a fiber intersecting the surface at an angle. The arcs on either side are interference patterns which can be used to measure the fiber angle.

 Figure 3 An electron diffraction pattern from a thin foil of gold. The ring diameters correspond to diffraction angles that identify the spacings of planes of atoms in the crystal structure. Figure 4 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.

 Figure 5 Typical graphics used to communicate news information (Source: USA Today). Some images such as holograms or electron diffraction patterns are recorded in terms of brightness as a function of position, but are unfamiliar to the observer. Figures 3 and 4 show electron diffraction patterns from a transmission electron microscope, in which the atomic structure of the samples is revealed (but only to those who know how to interpret the image). Other kinds of data including weather maps with isotherms, elaborate graphs of business profit and expenses, and charts with axes representing time, family income, cholesterol level, or even more obscure parameters, have become part of our daily lives.

 Figure 5 shows a few examples. The latest developments in computer interfaces and dis- Plays make extensive use of graphics, again to take advantage of the large bandwidth of the human visual pathway. There are some important differences between human vision, the kind of information it yields from images, and the ways in which it seems to do so, compared to the use of imaging devices with computers for technical purposes. Human vision is primarily qualitative and comparative, rather than quantitative. We judge the relative size and shape of objects by mentally rotating them to the same orientation, overlapping them in ourminds, and performing a direct comparison.

 This has been shown by tests in which the time required to recognize features as being the same or different is proportional to the degree of misorientation or intervening distance.

 Figure 6 shows an example. Figure 6 Several views of a complex three- Dimensional figure. Which of the representations is/ Are identical and which are mirror images? The time needed to decide is proportional to the misalignment, indicating that we literally “turn the objects over” in our minds to compare them….

Download  The Image Processing Handbook