Biometric Authentication of Computer Users Using Facial Images in Protected Execution Mode

Thegoalofthepresent workis toincrease the reliability of facial biometric-based key generation, which is used for remote authentication with the protection of biometric templates. A neuro-extractor model was developed to associate a feature vector with a cryptographic key or user password. The model is a shallow neural network based on partially connected trigonometric neurons. The model is trained to generate a cryptographic key or password at the output when a feature vector of a legitimate user is received at the input and to generate a random code when a feature vector of an unknown subject is received at the input. The proposed neuron model is based on the use of trigonometric measures of proximity. This approach can significantly improve the accuracy of biometric image classification and the length of cryptographic keys associated with a biometric image, while ensuring the confidentiality of biometric templates. An algorithm for the synthesis and automatic learning of neuro-extractor was proposed. The experimental evaluation showed next results: EER ≈ 0.6% and EER ≈ 1.2% on the open datasets Faces94 and LFW. The trigonometric neuron network is capable of producing a 2048-bit key, which is a higher reliability indicator compared to previously achieved results. The model can be used not only in facial biometric applications, but also in authentication by other modalities, as well as in building a secure mode for executing artificial intelligence (AI) algorithms. The secure mode is an important component of the trusted AI concept. This mode complicates the analysis of operations performed by AI, unauthorized control of AI, and the extraction of its knowledge by unauthorized persons. The trigonometric neuron network has increased resistance to such destructive effects.