Artificial neural networks for image understanding

This book, according to its preface, is aimed mainly at researchers in ireas such as remote sensing, medical image processing, robot \ ision. military reconnaissance, Lartography and geology who are Lonsidering the use of neural networks for their processing tasks. To this end, the book describes neural network architectures and algorithms as applied to basic image processing tasks, and gives examples where these techniques have been used. There are references to relevant research papers, so as to allow readers to discover the important details. Chapter I gives a brief introduction to neural networks which, as usual, gives the wrong impression that all networks are based around the perceptron model of a neuron. Chapters 2 4 introduce the basic tasks in image processing of preprocessing, feature extraction and texture analysis; standard image processing algorithms are briefly introduced and examples are given showing where neural networks have been used for these purposes. Some of these examples use network types which are not described until later chapters. Chapter 5, ‘Supervised classifiers’, gives examples of the use of two supervised neural network types for classification processes, the multi-layer perceptron and Hopfield networks. Chapter 6 does the same for unsupervised networks; Kohonen, ART and competitive learning networks being described. Chapter 7 describes associative memory networks, including the BAM, the selective reflex memory, the Hoptield network and counter propagation networks. Chapter 8, on three-dimensional structures and motion, introduces neural and traditional methods for such purposes. Chapter 9 describes specialist hardware produced, such as coprocessors for example, to implement neural networks: this chapter would probably be better as an appendix. Finally, Chapter IO describes some more applications of image processing. The main thing this book provides is a list of applications of neural networks to image processing. However, there is insufficient information for the reader to implement a neural network in any application, let alone for image processing; for that the reader must consult a good introduction to neural networks and the particular reference(s) on the application. Thus the book is only a review of the subject. It is, however. a poor review in that it only describes the applications rather than <criticize them. Also, for a book published in 1994 to list relatively few references dated after 1990 is poor. Also, as usual for an American book, it ignores the weightless form of neural network which has been used in many image processing applications. The book is. however, easy to understand. well written (though its organization is not ideal), and with plenty of illustrations. It is probably useful as an indicator of possible application areas for neural networks, and as such I would recommend it for a reference library, but not (as is claimed in the book) as essential reading for anyone working in the field.