Abstract: Firstly, LogGabor filter was used to extract the iris amplitude features, according to the type of iris library the filter parameters were optimized by the improved genetic particle swarm optimization algorithm. Secondly, principal component analysis was used to reduce dimensions, thereby reducing noise and redundancy. Thirdly, a dynamic radial basis function (RBF) neural network was constructed and iris recognition was performed by the Euclidean distance between iris amplitude features. Finally, we compared a variety of small iris libraries with other iris recognition algorithms. The experimental results show that the algorithm has higher accuracy in onetoone iris recognition, the ROC curve is closer to the coordinate axis, and filter is more versatile, which improves the recognition rate of the small iris library. It solves the problem of slow learning convergence speed and poor structural universality of traditional algorithms.

Key words: iris recognition, LogGabor filter, genetic particle swarm optimization algorithm, dynamic radial basis function (RBF) neural network, Euclidean distance