Procedure for Adapting a Neural Network to Eye Iris Recognition

The article is devoted to the problem of adapting the structural parameters of a convolutional neural network to iris recognition during biometric authentication and iridology. It is shown that the most effective type of neural network model is the convolutional neural network. There was determined a list of convolutional neural network parameters, which should be adapted to the conditions of the iris recognition problem. There are proposed a number of adaptation principles, based on an analogy with iris recognition by a human-expert. On the basis of the proposed principles, there was developed an original procedure for adapting the neural network model to the conditions of the iris recognition problem. In contrast to the known solutions, the development involves the use of the proposed adaptation principles, which allow determining the main parameters of the convolution and sub-sample layers. It was experimentally proved that the use of the proposed procedure allowed developing a neural network model whose accuracy at the level of 0,95 corresponds to the best modern solutions of a similar purpose. There is shown the expediency of further researches in the development of the neural network analysis Iris method using CNN.