Data-knowledge driven: a new learning strategy for iris recognition

This article focuses on the issues of poor interpretability and low universality of traditional iris recognition models in unsteady states. It proposes a new learning strategy for iris recognition: data-knowledge driven strategy, whose core idea is that the iris category knowledge is extracted from the clustering range of the iris feature data, and the knowledge is integrated into the recognition decision-making process to promote the recognition. The process of knowledge cluster analysis enables users to clearly understand the process of obtaining decision basis, and improves the interpretability of the process of recognition model design. The iris feature knowledge is set according to the consistent fact reflected in the data distribution of a large number of iris samples in various scenarios under the same process, which enhances the universality of the iris recognition model in the unsteady state. In addition, the data-knowledge-driven mode decreases the impact of the semantic gap between iris feature data and iris physiological form on the iris recognition model, thus effectively reducing the dependence of the iris recognition model training on data. An iris recognition model aiming at the process of feature expression and recognition is tested in different iris libraries. The experiment results show that the application of data-knowledge driven strategy to iris recognition is feasible and rationality, and it can make the recognition model complete the unlimited iris category recognition which can be expanded at any time.