Intelligent Image Processing review

Retroactively posted April 15, 2005

I obtained a copy of Steve Mann's textbook, "Intelligent Image Processing", via interlibrary loan from UMass, so I thought I'd send out a brief review for anyone who might be interested in what it was like.

Executive summary: An all-in-one compilation of virtually everything Steve has done, with a strange mix of too much detail and too little detail. Interesting and possibly inspiring, but not overly useful for the average wearable computing enthusiast / researcher: good to have on your bookshelf, but unlikely to get well-worn from constant use.

Longer version: The book covers several different subfields of research that Mann has been involved in over the years, mixed in with a heavy dose of idiosyncratic jargon and social commentary. Much of it is adapted from his previously published (and unpublished) corpus of work, so the result is a very personal "textbook". Nobody but Mann himself could really use this as the basis for a course, for instance.

Introduction. The start of the book is mostly introductory and general: why wearable computers are interesting, what are the characteristics they should have, what trends in society can be extended or subverted by the use of such technology, augmented versus mediated reality, etc. There is also a lot of historical background, albeit filtered through one man's impression of what the important contributions have been.

EyeTap. This was the section I was most intrigued by, and while it is a good introduction to the field -- what sort of problems do mediated reality systems run into and what solutions are proposed -- I really wanted an answer to the question of "What the hell does Steve wear and how can I build one?" and on that area I was disappointed. Still, it was useful, and to some extent it was even successful as a conversion: i.e. I'm more convinced that his approach is a good idea as opposed to a snow-job of jargon, and what remains is a mere question of practical implementation.

Image processing. These sections are fairly strong, by virtue of both being generally useful and tending towards mathematical rigor. Mann goes through a discussion of the response functions of cameras and display devices and how they influence image-processing algorithms. There are lengthy sections on the use of multiple exposures to create composite high-dynamic-range images and on stitching together multiple images with proper perspective mapping to create panoramas and perform head-tracking via video. The algorithm details sometimes get vague, but source code is available too, and there is enough mathematical background for the interested student to put the pieces together. For readers with a background in photography, there is additional discussion of Mann's previous work on the use of custom light sources to illuminate scenes (via blending multiple images) in artistic and revealing ways.

Appendices. There are some interesting extra sections, on designing custom key-input devices (I particularly love the section on using your hand shape to mold heated plastic: you may need to repeat the procedure if you exceed your pain threshold!). The "how to build a covert computer imaging system into ordinary looking sunglasses" appendix is completely worthless, however, as the advice boils down to (a) get a Kopin display, (b) remember to label the wires when you solder, (c) mount in your sunglasses, (d) ???, (e) profit! There was no discussion or diagrams of the needed optical system whatsoever, for instance.

So, overall, I'm happy to have found a copy to flip through, but glad I didn't gamble on buying it for myself. The main advantage to the text is having all or most of Mann's work in one place, but there is little extra information beyond what has been long available from his technical reports and various write-ups on the web. The definitive answers as to what he is up to are illuminated, but not really resolved here -- this is partially because the text is not current: although published in 2002, the serious references in his bibliography stop around 1998, so the last two "generations" or so of research by him and by others in the field are not reflected at all.