Review of five specific books on pattern recognition

$85.50 is the price for the 1,847 pages of reading in pattern recognition contained in the books under re;iew. Except for collections of papers and a book of solicited chapters edited by Mendel and Fu, these are the first texts in statistical and geometric pattein recognition since Nilsson's book Learning Machines, published in 1965. So it is with some ihterest and concern that we examine this multiple offering. Work on pattern recognition properly got its start in the late 1950's, primarily due to the work of Rosenblatt and Block on the Perceptron at Cornell University and Selfridge and col!eagues at M.I.T. There followed a book by Sebestyen and an influential paper by Highley-man. The first book on the general subject of pattern recognition appeared in 1965. This was Nilsson's book, Learning Muchines. The primary emphasis in this book was on discriminant functions and the Perceptron, Adaline, linear threshold device approach. This simple but comprehensive book made clear some of the fundamental ideas that had arisen in pattern recognition in the preceding years. While it cannot be said that any of these techniques allowed the solution of the pattern recognition problem, it was clear that the general problem of recognizing patterns had inspired new approaches to some old mathematical problems in hypothesis testing and statistics. In 1969, the book Perceptrons by Minsky and Papert appeared. The book is by far the deepest and most entertaining entry in the field of pattern recognition to date. However, it deals specifically with linear threshold devices and what the authors call the emerging field of computational geometry. This book should be part of the cultural background of any serious researcher in pattern recognition, but I seriously doubt whether it will be of any direct utility. What Minsky and Papert succeeded in emphasizing was that given a device, say the Perceptron, one should find the natural family of problems for which it is suited. Somehow in the previous years, it had been assumed that ad hoc devices were somehow universally good, and no attempt was made to delineate the natural family of problems for which they were suited. Thus the change in philosophy was extremely important. In any case, except for the book by Minsky and Papert, the book by Nilsson, an excellent text by Rosenfeld on pictorial pattern recognition, there has been little except for volumes of collected papers published in pattern recognition. Now we …