METHOD FOR AUTOMATIC FORMATION OF A BIOMETRIC IMAGE BASE "FRIEND" FOR TRAINING AND TESTING OF HIGH-RELIABILITY BIOMETRIC-NEURAL NETWORK AUTHENTICATION OF AN INDIVIDUAL Russian patent published in 2020 - IPC G06K9/62 G06F21/32 G06N3/02 

FIELD: personal identification systems.

SUBSTANCE: invention relates to biometric identification and personal authentication and can be used in information security systems and access control systems. In the disclosed method, examples of the natural hand-written biometric image "Friend" are entered, their conversion into an electronic form, standardization, Fourier transform, classification based on given criteria, selection of pairs of example parents, calculation for each pair of distance between their biometric parameters, selection of number of descendant samples depending on probability of occurrence of distance between parameters of pair of example parents, generation of synthetic examples of vectors of biometric parameters of the image "Friend" by morphing by linear interpolation of biometric parameters of example-descendants so as to ensure preservation of natural correlation links of biometric parameters of example parents, formation of an embryo of a biometric "Friend" image base based on natural and synthetic examples of "Friend" biometric parameters vectors, division of a base embryo into a training and test parts, neuron network conversion of a vector of fuzzy biometric "Friend" parameters into a clear unambiguous code of an access key for personal authentication.

EFFECT: technical result consists in reducing the time for input of natural handwritten images for training, further training and testing of the base of biometric images "Friend".

1 cl, 1 dwg