Multi-Instance Iris Recognition

Human irises are very rich in texture and have a high degree of uniqueness. This texture information can be used for identification of a human being. Information extracted from the human irises can be used to build a multi-instance biometric identification system. In this paper, a new method of feature vector extraction based on the Webber Local Descriptor algorithm is proposed for creating an iris-based multi-instance biometric system. The performance of an individual channel and a multi-instance system are evaluated, and the effectiveness of the Webber Local Descriptor for biometric recognition is investigated. Evaluations using the Webber Local Descriptor as a feature vector resulted in 88.39% Equal Error Rate (EER) for TAR-TRR when a multi-instance approach was adopted, which is a 6.44% improvement compared to the single instance approach.