Indexing Iris Database Using Multi-Dimensional R-Trees

Abstract

Iris is one of the most widely used biometric modality for recognition due to its reliability, non-invasive characteristic, speed and performance. The patterns remain stable throughout the lifetime of an individual. Attributable to these advantages, the application of iris biometric is increasingly encouraged by various commercial as well as government agencies. Indexing is done to identify and retrieve a small subset of candidate data from the database of iris data of individuals in order to determine a possible match. Since the database is extremely large, it is necessary to find fast and efficient indexing methods. In this thesis, an efficient local feature based indexing approach is proposed using clustered scale invariant feature transform (SIFT) keypoints, that achieves invariance to similarity transformations, illumination and occlusion. These cluster centers are used to construct R-trees for indexing. This thesis proposes an application of R-trees for iris database indexing. The system is tested using publicly available BATH and CASIA-IrisV4 databases.