Preface
The aim of t hi s book is to serve pedagogi c goals as a complement of the book Pattern Recognition,
4th Edition, by S. Theodoridis and K. Kout roumbas (Academic Press, 2009). It is t he offspring of
our experience in teaching pattern recognition for a number of years to different audiences such as
students with good enough mathematical background, students who are more practice-oriented, pro-
fessional engineers, and computer scientists attending short t raining courses. The book unravel s along
two directions.
The first is to develop a set of MATLAB-based examples so that students wil l be able t o experiment
with methods and algorithms met in the various stages of designing a pattern recognition system—that
is, classifier design, feature select i on and generation, and syst em evaluat i on. To this end, we have made
an ef fort to “design” examples that will help the reader grasp the basics behind each method as well as
the respective cons and pros. In pattern recognition, there are no magic recipes that dictate which method
or technique to use for a speci fic problem. Very often, old good and simple (in concept) t echniques can
compete, from an efficiency point of vi ew, with more modern and elaborate techniques. To t his end,
that is, selecting the most appropriat e techni que, it is unfortunate that these days more and more people
follow the so-called black-box approach: try di fferent techniques, using a related S/W package t o play
with a set of paramet ers, even if the real meaning of these parameters is not really understood.
Such an “unsci entific” approach, which real l y prevents thought and creation, also deprives the
“user” of the ability to understand, explain, and interpret the obtained results. For this reason, most of
the examples in thi s book use simulated data. Hence, one can experiment with different paramet ers and
study the behavior of the respective method/algorithm. Havi ng control of the data, readers will be able
to “study,” “investigate,” and get familiar with the pros and cons of a technique. One can create data that
can push a technique to its “limits”—that is, where it fails. In addition, most of the real-life problems are
solved in high-dimensional spaces, where visualization is impossible; yet, visualizing geometry is one
of the most powerful and effective pedagogic tools so that a newcomer t o the field will be able t o “see”
how various methods work. The 3-dimensioanal space is one of t he most primitive and deep-rooted
experiences in the human brain because everyone is acquain ted with and has a good feeling about and
understanding of i t.
The second direction is to provide a summary of the relat ed theory, without mat hematics. We
have made an effort, as much as possible, to present the theory using arguments based on physical
reasoning, as well as point out the role of the various parameters and how they can influence t he
performance of a method/algorithm. Nevertheless, for a more thorough understanding, the mathematical
formulation cannot be bypassed. It is “there” where t he real worthy secrets of a method are, where the
deep understanding has its undisputable roots and grounds, where science li es. Theory and practice
are interrelated— one cannot be developed without the other. This is the reason that we consider this
book a complement of the prev iously published one. We consider it another branch leaning toward the
pract i cal side, the other branch being t he more theoretical one. Both branches are necessary to form the
pattern-recognitiontree, which has its roots i n the work of hundreds of researchers who have effortlessl y
contributed, over a number of decades, both in theory and practice.
ix
