Customer Reviews

Introduction to Statistical Pattern Recognition, Second Edition (Computer Science & Scientific Computing)

5 star:		(1)
4 star:		(2)
3 star:		(1)
2 star:		(0)
1 star:		(0)

 

 

Fukunaga is a standard source for pattern recognition methods often cited in the engineering literature. Covers parametric (particularly linear and quadratic discriminant algorithms) and nonparametric methods (density estimation). It is designed for and popular with engineers. When I was working at Nichols Research Corporation Fukunaga's papers and this book (earlier edition) were often cited as sources to justify the algorithms we used for discrimination problems. In fact Fukunaga had been a consultant to the company (used primarily by the Boston branch of the company where the KENN algorithms were developed). It is a reputable source. I still like Duda and Hart (1973) for good explanations of the fundamental concepts. The second edition that was recent ly published with Stark as a third author is also highly recommended. For statisticians McLachlan's book is now far and away the best source.

Multivariate analysis is borrowed to name a NEW subject, Statistical Pattern Recognition (SPR). Many statisticians think it unfair or a shame. In spite of these, it is a good reference book of SPR. :-)

[1] Many contents of this book can be found in any graduate textbook of Multivariate Analysis, for instance, Fisher's linear disciminant, etc.
[2] The book is badly printed. Why not using LaTeX?
[3] Guassian distribution is assumed here and there.
[4] It may be good as a reference book, but definitely not as a textbook.

If you are writing a machine learning paper, and need to cite something to support an argument, you can almost always cite Fukunaga. This work is a standard reference in the field. The presentation of most material is very terse, but that is great if you already have a good feel for the material and need to look up some details about some algorithm or technique. There isn't much about neural networks here, but for the rest of the pattern recognition techniques, this is almost always the first place to start. Another strong point for this book is the use of realistic examples, which illustrate many of the statistical techniques.

I do not like to consult this book for the following, quite superficial reason. The book is sloppily produced and proofread
(and the fault is [probably] mainly the publisher's instead of the author's). This manifests itself, e.g., as follows

(1) the typography is flawed (the equations hurt at least my eyes);
(2) at its each appearance, the all-important >< -sign goes the wrong way.