V.N. Gridin, V.I. Solodovnikov, I.A. Evdokimova
Data is a valuable resource, with access which is necessary to be strictly monitored and regulated. One of the most popular approaches, used to prevent the possible threat of unauthorized access, is the cryptographic protection or encryption. There is a continuous search for new methods and algorithms to ensure reliable data protection.
The article investigates the possibility of using neural network methods for text encryption. That’s caused by their ability to recover distorted signals and recognize objects that differ from the reference characteristics. The proposed approach is based on the LVQ-net (Learning Vector Quantilization) paradigm. Algorithms of key formation, encryption and decoding are considered. The basic idea is to find a distorted code that can be recognized or restored by the network. At the preliminary stage, the random generation of different clusters is taking place, considering the appearance frequency of characters or their groups from the original alphabet. Further, the encryption process occurs within the boundaries of the obtained clusters. The encryption key includes information about the selected neural network template and its structural characteristics. The encryption key dependences on number of used clusters and the dimension of encryption space are considered