An Efficient Iris Recognition Method Based on Restricted Boltzmann Machine

As one of major methods for iris feature extraction, 2D-Gabor filter is capable of texture features in different directions and scales. Restricted Boltzmann Machine (RBM) is quite favored because of its simple structure and fastness in classification. However, due to conflicts between the feature vector dimension and the complexity of the network structure, rarely few combine them for iris recognition. This paper proposes a multi-class iris recognition method which combines 2D-Gabor feature extraction and classification model of RBM together. Firstly, 2D-Gabor filter is employed to extract energy-orientation feature of iris texture, whose dimension will not increase with the increasing number of filters. In this case, as the number of nodes in hidden layers of RBM network is determined to a definite value, complexity of the whole RBM design is simplified. Experiments show that this method displays high recognition accuracy on sample sets.