Pattern recognition in images has many applications, from scanning bar codes to identifying human faces. Feature-based recognition methods examine images piecewise, and make decisions based on comparing features and their relationships. Correlation pattern recognition methods, on the other hand, process entire images at once. If u(x,y) represents a pixel value at a point (x,y) in an unknown image, and v(x,y) the value in a reference image, then the correlation of two images is defined as c(m,n) = &Sgr;&Sgr;u(x,y)v*(x+m,y+n), where the summation is over values of x and y for which v(x+n,y+m) is defined. The “recognition” of the image u(x,y) is then based on the maximum value attained by ¦c(m,n)¦. For images represented continuously, the summations are, as usual, replaced by integrals. The reference image or function v(x,y) is often called a filter, since it may be a composite of several images. In fact, in optical correlators, the unknown image is projected through an optical filter representing the reference images. If the apparatus is set up correctly, the resulting image will be the correlation of the two images. As the authors of this book say, the correlation is computed “at the speed of light.”

In their introduction, the authors describe an ambitious scope for their book. Their audience includes graduate students in electrical or optical engineering, as well as “seasoned workers” who need “a useful current overview of the discipline.” Presumably to satisfy this wide range, nearly half of the text is taken up with “basic concepts in related fields,” such as matrix analysis, probability, Fourier analysis, and signal processing. There is even a short section on basic optical physics.

The meat of the book, however, is in chapters 5 and 6, on correlation filters, and chapters 8 and 9, on limited-modulation filters. Although the authors try to present approaches for both optical and digital filters, they clearly find optical image processing more interesting. In fact, both areas are interesting and challenging, and there should be more detailed explanations of the techniques for analysis and design of both types of correlators. For example, the algorithm in chapter 8 for the design of optical correlators solves a mixed discrete and continuous optimization problem. The problem and the algorithm need a more detailed description. To make room for this in a future edition, the material on basic concepts could be replaced by references to some of the excellent texts in linear algebra, probability, and so on that are listed in the bibliography.