4.

1 Applications of Neural Networks in Pattern Classification

learn to distinguish patterns of
environment,
ralern recognition is the study of how machines can observe the the patterns. The
about the categories of
nterest from theirbackground, and make sound and reasonable decisions
best pattern recognizers in most instances are humans.
at your disposal, the
more relevant patterns
Pattern recognition is critical in most human decision making tasks: The can surely
to proponents of artificialintelligence, since computers
better your decisions will be. This is hopeful news
be taught to recognize patterns.
land airplanes
e w of the pattern recognition example include, diagnose of
disease and programs that help pilots
depend in some way on pattern recognition.
4.1.1 Pattern Recognition
of patterns are important problems in a variety
of
Automatic recognition, description, classification, and grouping
medicine, marketing, computer vision, artificial
engineering and scientific disciplines such as biology, psychology,
a handwritten cursive word, a human face, or
intelligence, and sensing. A pattern could be a fingerprint image,
remote
a speech signal.
consist of one of the following two tasks: 1) supervised classification
Given a pattern, its recognition/classification may
the input pattern is identified as a member of a predefined class, 2) unsupervised
(e.g., discriminant analysis) in which
to a hitherto unknown class.
(e.g., clustering) in which the pattern assigned
is
classification
categorization task, where the classes are either
as a classification or
The recognition problem here is being posed
classification) or are learned based on the similarity of patterns (in
defined by the system designer (in supervised
tunsunervised classification). These applications include data mining (identifying "pattern", e.g., correlation, or an
a
outiier in millions of
multidimensional patterns), document classification (efficiently searching text documents),
retrieval of multimedia databases, and biometrics.
financialforecasting, organization and
available computing power, wnile enabling faster processing of huge data sets, has also
The rapidly growing and
and diverse methods Tor data analysis and classification. At the same time, demands
facilitated the use of elaborate
are rising enormousiy aue to the availability of large databases and
on automatic pattern recognition systems
and cost).
requirements (speed, accuracy,
stringent performance
The design of a pattern
recognition system essentially involves the following three asperts
ts:
Data acquisition
and preprocessing
1.
2. Data representation
3. Decision making.
 The probem domain dictates the choice of sensor(s),
preprocessing technique, representato
decision making model. t is generally agreed that a well-defined and sufficiently constrained recognition pro
(small intraclass variations and large interclass variations) will lead to a compact pattern representation and a >i

decision making strategy.

Learning Trom a set or examples (training set) is an important and desired attribute of most pattern recognitio
systems.
The four best known approaches for pattern recognition are:

1 Template matching
2. Statisticalclassification
3 Syntactic or structural matching
4 Neural networks.