Authors
Zhumazhanova S.
|
Degree
|
Post-Graduate Student, Junior Researcher, Omsk State Technical University |
|---|---|
|
E-mail
|
samal_shumashanova@mail.ru |
|
Location
|
Omsk |
|
Articles
|
Perspectives of the use of wide neural networks for human changed state detection by the thermograms of the face and neck regionsBiometrics is a field of computer science that studies the way of creating computer models of physiological
(hand geometry, fingerprints, iris, etc.) and behavioral (signature dynamics, handwriting, gait,
etc.) human characteristics for identification of one or several subjects, as well as their psychophysiological
state. Thermal imaging methods are perspective from this point of view for monitoring of
subjects’ state. The result of recognition of the subject or their state depends on the effectiveness of
each stage of the identification system. This article provides an overview of the results of research on
the automatic recognition of subjects’ states using thermal facial images, focusing on the used identification
features and the most promising decision-making algorithms. Nowadays, «deep» neural networks
are becoming popular. Results of research show that this network architecture is not optimal.
«Wide» neural networks, built on the basis of various functionals, have advantages over «deep» neural
networks. Some proximity measures work more efficiently with features that have a high crosscorrelation,
while other functionals focus on its absence. Experimental data confirm the effectiveness
of «wide hybrid» neural networks consisting of subnets of various functionals and capable of training
from a small number of examples of biometric images. The task that needs to be solved is to test such
networks on the identification features of thermal images of the face and neck regions.
Read more...
|
