MODEL OF THE SYSTEM OF BIOMETRIC VERIFICATION OF INFORMATION SYSTEMS USERS

Abstract

A hybrid model of the system of text-independent dynamic verification of users of information systems,
which is based on the integrated use of artificial immune systems and artificial neural networks, is
proposed. The verifiable data of text-independent dynamic user biometrics are represented by two sequences
of information units of fixed-size feature vectors corresponding to the images of two classes –
‘friend’ and ‘stranger’. This representation is oriented towards the massively parallel decentralized data
processing adopted in artificial immune systems. The subsequent verification of the users of both classes is
realized by a probabilistic artificial neural network, which computes the probability densities of the concentration
of information units of both classes in the feature space. In addition to the probability density
characteristics of the information units, the allowable 1st and 2nd kind error prices for images of each
class are used. The final result of biometric verification of the working user is controlled based on the
current comparison of the aggregate statistical estimates of the probability density and the acceptable
price of errors of the images of each of the two classes. The proposed approach to verifying the identity of
a working user allows to propose a general scheme of this procedure for significantly different modalities
of dynamic biometrics: voice, handwriting, and keyboard typing. The implementation of such an approach
for specific modality biometrics will be slightly different, but the general verification scheme can be maintained.
The advantages of the proposed approach are: the possibility of text-independent analysis of dynamic
biometry of different modality, arbitrary volume, content and language; possibility of making a verification decision in continuous mode at the rate of user's work arrival; in the future to increase the
accuracy of the verification system by increasing the dimensionality of the neural network; the possibility
of using the history of analysis of verification results of real users for further more accurate tuning of the
system. A relative disadvantage of the work is the necessity of program realization of a neural network of
large dimensionality. However, in the future, this disadvantage will be quickly leveled with the increase of
computing performance.