Browse Books

Reviewer: Raphael M. Malyankar

Enabling computers and other machines to process and understand optical sensory input is a problem that has seen a great deal of activity in both research and industry for several decades; it is one of the more important and active research problems being explored even today. Davies' experience in the field of machine vision is long and varied, and this shows in the polish and breadth of the third edition of this textbook. This book consists of four main parts. The first (chapters 2 through 8) discusses the elements of vision systems and elementary operations, such as image filtering, thresholding, edge detection, centroid detection, and mathematical morphology. The second part (chapters 9 through 15) addresses intermediate-level vision problems and techniques, ranging from the detection of fundamental features, such as lines, planes, and circles, to elementary abstract pattern matching for object identification. Much of this section is concerned with discussions of the Hough transform and derivations thereof, and their use in intermediate-level vision techniques. Part 3 (chapters 16 through 21) covers problems arising from the three-dimensional nature of the real world. This part begins with a discussion of issues for machine vision arising from projection and perspective; continues with problems and techniques in dealing with motion, applications for recognizing and tracking moving objects, and robot navigation; and concludes with a chapter on theoretical concepts for camera calibration and image rectification. The fourth main section (chapters 22 through 28) addresses problems and techniques pertaining to a diverse collection of machine vision applications, including automated visual inspection in manufacturing and other commercial contexts, statistical pattern recognition, artificial neural networks, considerations pertaining to textures, and, finally, image acquisition and a theoretical perspective on hardware system structuring considerations for vision systems. The four main sections are followed by a short final section, consisting of one chapter, with an integrated view of machine vision systems in terms of the techniques described earlier, and general considerations for designing vision systems (such as requirements and tradeoffs). An appendix outlining theoretical issues in robust statistics follows. Exercises are provided after most chapters, and an extensive bibliography is included at the end of the book. (Some peripheral references are outdated; for example, the book by Clocksin and Mellish on Prolog, mentioned on page 435, is now in its fifth edition [1].) One significant and useful feature of this book is the inclusion of the author's evaluations, and other comments, on the theory, tradeoffs, and practicalities of machine vision algorithms. Almost every chapter concludes with a brief summary of the advantages, disadvantages, and applicability of the algorithms and techniques it describes, followed by bibliographical notes containing pointers to the source material from which the contents of the chapter are drawn. (These notes will be of special interest to the reader who is using the book for independent study.) Chapter 29 ("Machine Vision: Art or Science") rounds off the book nicely in this regard. Another very useful feature is the inclusion, in several chapters, of short case studies, to explain and illustrate problems with, and applications of, the techniques presented in the chapter. The two greatest strengths of this book are its broad coverage and its careful and lucid exposition of the subject matter. A large number of problems and techniques are covered, and verbal explanations accompanied by illustrative diagrams and photographs are provided for everything. This is not to say that the book lacks formality; formal presentations of the underlying mathematical concepts using mathematical notation are indeed included where warranted, but even in these situations the key ideas are explained in the accompanying text. The stress on verbal explanations might be somewhat disconcerting for the advanced student or researcher, who might want to see more detailed formal presentations, but the bibliographical notes should help such readers with pointers to the original literature, or at least with author names and keywords to use while searching for primary sources or current research. At over 900 pages, this book is unusually long for a textbook, but I actually found myself wishing for more explanatory material on some topics. The most important example of this is in chapters 9 and 11, which could do with more elucidation of the Hough transform and its derivations and uses: the basic ideas are elegantly simple, but might be difficult for the novice to understand merely from the explanations in this book. Less important examples are Prolog and search algorithms, outlined in chapter 15. (For one thing, breadth-first and depth-first search are easy to understand from diagrams, but are difficult to explain with words alone; further, Prolog is much easier to explain with an example of a Prolog program.) I would also have liked to see a diagrammatic "roadmap" in the introduction, giving an integrated view of linkages between the main concepts in selected application areas (perhaps something along the lines of an extended version of the matrix on the inner front cover). However, given the already considerable size of this book, it would certainly be difficult to add more material, whether to flesh out explanations or to make its coverage more comprehensive. The preface indicates that this book is intended both for university students at the final-year undergraduate and graduate levels, and for the researcher and practitioner. It will serve as an excellent primary text for university courses, though the instructor will find it necessary to add programming exercises, and to supplement a graduate course, in particular, with selections from the research literature. The independent reader will have a little more trouble, but should still learn the basics and pick up enough hints to identify primary sources and current research topics in the research literature. There is no compact disc enclosed, which I do not consider to be a problem, though some might find it desirable. More important, especially from the perspective of the independent reader, is the lack of mention of a readers' Web site, and of a survey of vision-related journals and Internet resources (the publisher does provide a Web site, but that is intended for instructors). These are, however, only minor deficiencies in a generally fine work. All in all, this book is an excellent learning and teaching resource for both university courses and the independent reader; it will be well worth the time expended on studying it. Online Computing Reviews Service