Statistical Pattern Recognition


ISSN: 0368-492X

Article publication date: 1 April 2000




Andrew, A.M. (2000), "Statistical Pattern Recognition", Kybernetes, Vol. 29 No. 3, pp. 392-398.



Emerald Group Publishing Limited

This gives an extremely comprehensive treatment of a range of topics under the general heading, treated in a style that is mathematically rigorous but at the same time allows the author’s personality to show through in informal comments about the power and applicability of the methods. It is written in textbook style, with exercises for the reader, as well as brief descriptions of real‐world applications, and five appendices covering aspects of the underlying statistical mathematics. It is a valuable and up‐to‐date reference work, with something in the region of 800 references and the useful feature of an author as well as a subject index. There are also references to a surprisingly large number of Websites at which relevant material is available, sometimes offering downloadable software.

The main concern is with supervised classification, or the assignment of an incoming data vector to one of a known set of categories. Attention is also given to unsupervised classification or clustering, in which the categories are determined by the data. One of the ten chapters is explicitly devoted to clustering. Much of the treatment in the book is also relevant to regression analysis, which is a related topic though not in itself classificatory. Both parametric and nonparametric statistical methods are considered. It is acknowledged that some of the literature quoted, especially the earlier items, follow a cybernetic approach in discussing biological cognition and learning. These aspects are not explicitly treated in the present book, but this does not reduce its usefulness as a reference for workers in these areas.

Despite the dissociation from biological studies as such, perceptions and other neural nets are introduced as means of implementing the methods, under the essentially‐statistical general headings of linear discriminant analysis and nonlinear discriminant analysis. The self‐organising feature maps due to Kohonen are also treated, this time under the general heading of clustering.

Most chapters have a list of applications of the techniques they introduce, as well as a section on “recommendations” suggesting how the reader should go about selecting and implementing a method in practice. Of the exercises for the reader, some refer to practical schemes and others are essentially mathematical. Many of them assume availability of computing facilities, for example to generate batches of data according to given criteria and then to apply methods of analysis to them.

The nature of the applications ranges far and wide and includes recognition of hand‐written characters, speech, and faces, and detection of patterns in medical, economic and geographical data, and in radar echoes. On page 77 a specific robotics application is mentioned, namely a robotic harvester for fruit, with sensory capability to classify images as belonging to one of the three classes of fruit, leaves and sky.

This is an admirable textbook and reference work, covering a field that is relevant to many cybernetic studies and projects. It is well presented, up‐to‐date and authoritative and will be a valuable addition to many bookshelves.

Related articles