Information Technology Reference
In-Depth Information
Neurocomputing-based practices are essentially characterized by imitating biolog-
ical functions, which exhibits natural candidate technology in biometric research.
Biometric recognition can be customarily dignified as typical nonlinear problem.
The neurocomputing-based techniques have appeared very successful in solving
these typical problem.
Since neurocomputing-based techniques have better ability to deal with nonideal
scenarios, therefore, they can provide robust and efficient solutions in the varied
environmental situations which include variations in illumination, noise, pose, and
partial occlusion.
The neurocomputing-based techniques are efficiently adaptive. This capacity
allows the development of real time biometric system capable of online learn-
ing and able to conveniently adjust if new subject is added in the biometric data
base.
Biometric recognition systems have shown excellent performance in the field of
secured access control, forensic science and surveillance using face, palm, finger-
print, iris, periocular and oculo-movement scan path signals and other biometric traits
for human recognition [
30
]. The fundamental requirement of any biometric recog-
nition system is a human trait having several desirable features like universality,
distinctiveness, permanence, collectable, performance, circumvention, and accept-
ability. However, a human characteristic possessing all these features has not yet been
identified. As a result, none of the single biometric trait can provide perfect recog-
nition. It is also difficult to achieve very high recognition rates using single trait due
to problems like noisy sensor data and nonuniversality or lack of distinctiveness of
the chosen biometric trait. Therefore, the performance of a biometric system may
be improved by utilizing a number of different biometric identifiers. The result of
combined effects will be more robust to noise; and minimize the problem of nonuni-
versality and lack of distinctiveness. The face recognition, one of the most important
biometric traits due to its non intrusive nature, has been considered in this topic for
illustration of computing capabilities of high-dimensional neurocomputing. Results
demonstrated in Chap.
7
on face biometric trait is very competitive with respect to
related existing techniques.
1.4 Scope and Organization of the Topic
The second generation neurocomputing may conceptually be defined as the era of
high-dimensional neural networks, higher-order neurons and fast supervised and
unsupervised learning algorithms. These advances have prompted active research
activities in neurocomputing where it was supposed to be saturation in the first gener-
ation. The artificial neural networks with high-dimensional parameters are developed
to explore natural processing of high-dimensional data, considering themeither in the
form of number or vector. They are characterized on the basis of high-dimensional
information, which flows through the network. They lead to adaptive systems which
Search WWH ::
Custom Search