Biometric authentication method based on cepstral characteristics of external ear echograms and biometrics-to-code neural converter

Abstract

Open biometric images (fingerprint, iris, face) are "in sight" and therefore compromised in the natural environment. In this work, it is proposed to use data on the internal structure of the outer ear obtained using echography as biometric images. The individual characteristics of the ear canal of subjects are hidden from direct observation and cannot be copied by photographing. The proposed authentication method is based on cepstral analysis of echograms of the ear canal using neural network biometrics to code converters, trained in accordance with GOST R 52633.5. The neural network biometrics-code converter allows you to associate a user's cryptographic key or password with his biometric image. This is a shallow neural network of one or two layers of neurons, which is configured to generate a key specified during training when an image of a known user arrives, and when an unknown image arrives at its inputs, generate a random code with high entropy. At the entrance to this network, cepstral signs of echograms were received. To apply the method in practice, you need a special device that combines a headphone with a sound-proof housing and a microphone. The results obtained can be called optimistic EER = 0.031 (FAR = 0.001 at FRR = 0.23). The use of neural network converters biometrics-code showed a relatively higher percentage of errors in comparison with multilayer neural networks and the naive Bayes classification scheme, however, neural network biometrics to code converters allows you to implement authentication in a protected mode. This means that the subject's biometric data will be protected from compromise at the stages of storage, execution and transmission via communication channels.