An efficient indexing scheme for iris biometric using K-d-b trees

Abstract

In this paper, an indexing approach is proposed for clustered SIFT keypoints using k-d-b tree. K-d-b tree combines the multidimensional capability of k-d tree and balancing efficiency of B tree. During indexing phase, each cluster center is used to traverse the region pages of k-d-b tree to reach an appropriate point page for insertion. For m cluster centers, m such trees are constructed. Insertion of a node into k-d-b tree is dynamic that generates balanced data structure and incorporates deduplication check as well. For retrieval, range search approach is used which finds the intersection of probe cluster center with each region page being traversed. The iris identifiers on the point page referenced by probe iris image are retrieved. Results are obtained on publicly available BATH and CASIA Iris Image Database Version 3.0. Empirically it is found that k-d-b tree is preferred over state-of-the-art biometric database indexing approaches. © 2013 Springer-Verlag.