Machine Recognition in Complex Domain - High Dimensional Neurocomputing: Growth, Appraisal and Applications - page 146

RW − t is IC representation of a image. Thus, statisti-
cally independent feature vectors of images have been obtained in B. R PCA as
preprocessor also ensures that C ICA algorithm does not magnify the effects of
noise.

=

9. Each row of matrix B

7.3 Human Recognition Systems

The machine recognition is both important and challenging technique. This chapter is
aimed at developing an intelligent machine recognition system that is noise invariant
and can work well in occluded and blurred environment. This also aims at reducing
the computational cost and provides a faster recognition systemwith complex-valued
neurons presented in Chap. 4 . This chapter explores the abilities of real and complex
ICA for feature extraction and compare them with real and complex PCA. In order
to verify the utility of neurons-based classifiers in complex domain, the empirical
studies are on two standard face data sets are presented. This chapter demonstrate the
comprehensive performance of various neuron models along with feature extraction
techniques in terms of training epochs for learning, number of learning parameters
(weights) for storage and accuracy of recognition system in varying conditions.
Recent developments in machine intelligence and call for better security appli-
cations have brought biometrics into focus. It is known that signature, handwriting,
voice, and fingerprint have a long history. More recently retinal scan, iris scan, peri-
ocular, occulomotion and facial information are considered for biometrics. When
deploying a biometrics based system, we consider its accuracy, cost, ease of use,
whether it allows integration with other systems as well as the ethical consequences
of its use. Biometrics can be defined as the automated use of physiological and
behavioral characteristics for verifying or recognizing the identity of living person.
The example involved in the measurement of biometric features are classified as per
their characteristics:
1. Physiological features: face, fingerprints, palm print, DNA, iris, periocular, hand
geometry etc.
2. Behavioral features: signature and typing rhythms, voice, gait, etc.
Physiological features are more stable than behavioral. The reason is that phys-
iological features are often nonalterable except in the case of severe injury, while
behavioral features may fluctuate due to stress, fatigue, or illness. The interest and
research activities in security systems have been increased over the past few years.
This growth is largely driven by a growing demand for identification and access
control at very important places. For the purpose of better security applications, the
variety of identification attributes can be classified into three broad categories:
•

Lowest level security: something we have, such as photo ID.

• Middle level security: something we know, such as password or PIN.

• Highest level security: something we do or something we are, which comprises

physiological and behavioral biometrics.