Twenty years ago in the field of images, there were only image processing books.  Then came the first book devoted mainly to higher-level image methods, Ballard and Brown’s, Computer Vision.  Since then, there have been a number of books on image processing and analysis at its various levels.  Two books have recently been published, both titled Machine Vision.  I was interested in seeing how the texts in this field have evolved and at which audiences they are aimed.

 

E.R. Davies’ Machine Vision Theory, Algorithms, Practicalities

At the risk of sounding trite, the first word that comes to mind when first seeing E. R. Davies book Machine Vision Theory, Algorithms, Practicalities is big.  It is 934 pages, 29 chapters. The second word, after examining the book, is understanding – and this explains the length.  Davies makes a conscientious effort not just to teach the various methods of computer vision but to convey an understanding of these.  This is done with excellent explicative text and figures as well as detailing the finer points of applicability, pros and cons, alternatives, and practicalities.  This is not a book that is heavy in rigorous definitions or mathematics; instead it is heavy in clear and intuitive explanation.  The popularity and usefulness of this style is evident as this book, originally published in 1990, is now in its 3^rd edition (2005).  The book has well-deserved longevity.

 

One of the ways that understanding is conveyed here is by application case studies.  These include: monitoring vehicle traffic, animal tracking, vehicle guidance, food quality inspection, and noise suppression in neural networks.  A handy table in the front flap identifies each application and some of the methods used in each.  For those who learn best by example and for those practitioners with objectives similar to those of the described applications, this feature is very useful.

 

The book is organized by parts and chapters.  Part 1, “Low Level Vision”, includes topics such as filtering, thresholding, edge detection, and morphology.  Part 2, “Intermediate Level Vision”, includes line, circle, ellipse, and hole detection.  Part 3, “3-D Vision and Motion” includes perspective, motion, and camera calibration.  Part 4, “Toward Real-time Pattern Recognition Systems” includes visual inspection systems (with plenty of examples), statistical pattern recognition, neural networks, and real-time hardware systems.  Finally, Part 5, “Perspectives on Vision”, ascrnds one level above the methods and examines the approach in specifying and designing vision systems.  It is here, in a chapter called, “Machine Vision: Art or Science” that Davies adds to and sums up some of his over-riding advice on designing vision systems.

 

The wide breadth of coverage includes some topics that (to my experience) are not widely employed now (such as SIMD systems). There is one missing topic:  there is no formal treatment of scale-space (multi-resolution, or Gabor filtering) techniques.

 

If a person is to buy one book on machine vision and its related topics (such as statistical pattern recognition and vision architectures), this is a good choice.  As a student text, this is appropriate for the undergraduate level or as a first course at the graduate level. Practical problems are included with each chapter.  As an advanced text, it should be complemented by papers from current literature.  For practitioners, especially those from other fields, this book offers a single source to easily find a topic, learn it, and understand it by example.  The absence of a software CD with this book means that student or practitioner will have to obtain programs elsewhere.

 

Snyder and Qi’s Machine Vision

A new book, Machine Vision by Snyder and Qi, is squarely aimed at students taking a first course in computer vision.  The book is much shorter than Davies book (433 pages), and does not include the treatment of several applications discussed by Davies.  The book does include more formal mathematical explanations of methods as befits a student text.  More pointedly for a student text, this book contains several assignments, located throughout each chapter with associated topics.  The assignments are well-designed to convey understanding through solution and practice.  Also included with this book is a CD, containing image software, assignment material, images and documentation. 

 

The authors state in the introduction that they have endeavored to write conversationally and entertainingly for the student.  I imagine that students would enjoy some of the jokes and quotes.  My favorite quote is from Julius Caesar for the Segmentation chapter, “Galia est omnes divisa in partes tres.” (All of Gaul is divided in three parts.)

 

Professors are tasked with beginning a course with review material to make sure that all the students start on the same page. This book actually provides this as Chapter 2.  In Chapter 3, some programming basics are discussed and the Image File System (IFS) software package is introduced.  After this introductory material, the  book describes linear operators and kernels, using a classic digital signal processing progression.  Edge detection and scale space are handled in this chapter.  The next chapters are: image relaxation, mathematical morphology, segmentation, shape labeling, and parametric transforms (mainly the Hough transform).  After these chapters, the book steps up one level to intermediate level vision (the book is not organized in parts as is Davies’) to describe graphs, matching, and statistical pattern recognition, clustering, and syntactic pattern recognition.  Finally, a few applications are briefly discussed (in 4 pages) in one chapter and the final chapter is devoted to automatic target recognition.

 

I found that this book has sufficient material for a student text.  I liked the writing style, the many figures and diagrams, and many images included on the CD.  This is not a cookbook meant for a practitioner, although I imagine after taking a course using this as the text, that the book and CD would be very useful for that student who ended up working in machine vision.

 

I liked both of these books and believe either would make an excellent choice for their intended audience.

BOOKSBOOKSBOOKS

Machine Vision Books

 

By:  Larry O’Gorman