Image processing is concerned with enhancing, drawing features from, and understanding real-world data represented as a matrix of spatially organized values. Optimization problems are those concerned with maximizing (or minimizing) some objective function by varying one or more parameters, often under a set of constraints. This book examines the application of optimization to common problems in image processing: segmentation, template matching, and shape detection.

The book begins with a very short introduction to machine learning in general, moving to a discussion of general optimization problems. A specific optimization method, electromagnetism-like optimization (EMO), is presented in some detail; it is used as the optimization method that is applied to image processing problems throughout the book. EMO is similar to evolutionary computation, except that there is no mutation and the population of particles is transformed in solution space through an EM-like response of each particle to all the rest. A formula similar to Coulomb attraction is applied, with the particle’s goodness of fit interpreted as the “charge.” As a very simple explanation, this has the effect of drawing poor solutions toward better ones.

The EMO method is applied to multilevel segmentation, through selecting solutions of both Otsu’s class separation and Kapur’s entropy method. Results were satisfactory, though the surprisingly low number of iterations required by the EMO raises the question of whether this is an ideal demonstration of the method. The authors conclude that “electromagnetism systems can be effectively considered as an attractive alternative for this purpose.”

The application of EMO to template matching is possibly the most impressive and actionable section in the book. EMO optimization is used instead of exhaustive search for template locations to be examined. On a variety of images, the optimal match point is found in less than half the number of template evaluations when compared with other algorithms such as imperialist competitive advantage. It may be that the properties of the match surface when real images are processed is well suited to an EMO-like optimization method.

A third application, detection of circular shapes, is investigated. The method is not based on Hough methods, but rather on circle fitting to points in the Canny edge image. Response of the method to variations such as multiple circles and ellipses is demonstrated. The circle detection method using EMO is applied to the detection of blood cells in imagery; a full chapter describes this application.

Readers interested in improving the accuracy or performance of image processing systems would do well to read this book. Though the text could be clearer, the material is not difficult to understand. While the theme of the book is that EMO is an ideal optimization method for image processing, the larger message is that the application of optimization methods to core imaging algorithms is well worth the effort.