Abstract
In biometric authentication systems, the process of prov-ing and verifying the authenticity of a user-claimed name through the user's presentation of his biometric image is performed and by converting this image in accordance with a predefined authentication protocol. An important issue remains the transformation of biometric data into code. The paper discusses the two most well-known conversion technology of biometrics in the code, the scheme of con-version of biometric parameters in the code key. It is shown that one of the main reasons for the difficulties of biometric authentication is the high dimensionality of the problem. To solve this problem, there are artificial neural networks or "fuzzy extractors". Of the many existing learn-ing algorithms of neural networks selected algorithm for automatic training of large artificial neural networks. Shows the use of entropy device to reduce the dimension of the problem of converting the biometrics-code. To reduce the volume of calculations made the transition to the Ham-ming distances.