It’s about time! The fields of digital image processing and computer vision have been around for about 40 years now. One mark of a mature field is a dictionary of terms. To my knowledge there has not been a book devoted to this purpose before. Some texts have glossaries, but they have neither been as extensive nor comprehensive as to cover the whole field. Perhaps because of this, I have sometimes noticed imprecise use of terms, re-invention of terms, and failure to call something by a term that has already been established (but perhaps not become widely known). This 342-page handbook should remedy these problems very well.
The book has over 2500 definitions. Many of the definitions are accompanied by pictures or diagrams. Want to know what a “region of interest” is? Look it up, and you’ll find an 18-line definition and a figure. The definition contains not only what it is, but when it might be used (to reduce competition or to focus processing such as to reduce algorithmic distraction or distortion). Furthermore, there is a textual example of use (for tracking a target through an image sequence). The figure shows a picture of a man’s face with a box around the eye, and the image portion within that box is extracted.
I looked up Lambert’s Law. This is the rule that the observed shading on an ideal diffuse reflector is independent of the observer’s position, but dependent upon the angle, q, of the source. How is this angle measured? A diagram is right there to show this.
Here’s a term that I could have used the dictionary for in the past, “profile”. A profile is a histogram of the number of on-valued pixels in rows and columns. There is a horizontal and vertical profile – which I’m sure I’ve mixed up in the past. I have always been unsure whether the horizontal profile is along the horizontal axis, or accumulates horizontal rows, in which case it’s along the vertical axis. The dictionary tells me that the horizontal profile is an accumulation of pixels in each row and vertical profile is an accumulation for columns. Now, I won’t make that mistake again.
The definitions are well-written, clear, and concise. The accompanying figures are a marvel of high-definition printing. Although they are downright miniscule, averaging 2 by 5 cm (¾ inch by 2 inches), they are usually very readable and instructive. Look up low-pass filtering and the accompanying figure shows a well-focused photograph of a piece of complex machinery and a blurred—low-pass filtered—picture beside it. Look up “watermarking” and you’ll be pointed to “digital watermarking” where there is a photo of a face, an image containing the word “watermark”, and a final image where these are combined and you can make out the watermark within the face image. Look up “semantic net” and the accompanying figure shows a simple example of a line drawing of an arch upon supports, and the semantic net, which shows how these components relate to one another.
Although not in excess, the authors are not afraid to use equations when that is the best way to complement a textual definition. The cosine transform, cross-correlation, and multi-variate normal distribution—to name a few—all appear with equations.
The cross-referencing is excellent, as indicated by underlined words and phrases. For instance, the entry for “halftoning” has “See dithering.” Under “image warping”, the definition contains an example of its use containing, “to correct some image distortion”.
I have a minor quibble with the “References”—or bibliography—section, which lists 13 text books in the field. As explained in the Preface, this is included to aid readers who want to delve more deeply into concepts than a dictionary can provide. After each book listing, there is a one or two sentence description to give readers an idea how well the book might meet the reader’s purposes. However, by inclusion in the dictionary and by being located up front between the Preface and the definitions, there is an implication that the list is comprehensive. But many good books are not on this list, and some books that do appear, are admitted to be “dated.” Two classic, popular, up-to-date, and excellent books that have been mentioned recently in issues of this Newsletter, but do not appear in this list, are Machine Vision (3rd ed) by Davies, and Digital Image Processing (2nd ed.) by Gonzalez and Woods.
My small quibble aside, I believe this dictionary will make an excellent complement to the library of any computer vision and image processing student, researcher, or practitioner. I know that the next time I’m writing about profiles, I won’t delay with my confusion between horizontal and vertical. I’ll just reach for my copy of this dictionary.
Computer Vision and Image Processing
By R.B. Fisher, K. Dawson-Howe, A. Fitzgibbon, C. Robertson, E. Trucco
John Wiley & Sons, August 12, 2005
Reviewed by: Larry O’Gorman
Book Reviews Published in
the IAPR Newsletter
Kernel Methods for Pattern Analysis by Shawe-Taylor and Cristianini, Oct. 05
Machine Vision Books, Jul. ‘05
CVonline: an overview, Apr. ‘05
The Guide to Biometrics by Bolle, et al Jan. ‘05
Pattern Recognition Books, Jul. ‘04