Iris biometric indexing

Abstract

Indexing/retrieving sets of iris biometric signatures has been a topic of increasing popularity, mostly due to the deployment of iris recognition systems in nationwide scale scenarios. In these conditions, for each identification attempt, there might exist hundreds of millions of enrolled identities and is unrealistic to match the probe against all gallery elements in a reasonable amount of time. Hence, the idea of indexing/retrieval is - upon receiving one sample - to find in a quick way a subset of elements in the database that most probably contains the identity of interest, i.e., the one corresponding to the probe. Most of the state-of-the-art strategies to index iris biometric signatures were devised to decision environments with a clear separation between genuine and impostor matching scores. However, if iris recognition systems work in low quality data, the resulting decision environments are poorly separable, with a significant overlap between the distributions of both matching scores. This chapter summarizes the state-of-the-art in terms of iris biometric indexing/retrieval and focuses in an indexing/retrieval method for such low quality data and operates at the code level, i.e., after the signature encoding process. Gallery codes are decomposed at multiple scales, and using the most reliable components of each scale, their position in a n-ary tree is determined. During retrieval, the probe is decomposed similarly, and the distances to multi-scale centroids are used to penalize paths in the tree. At the end, only a subset of branches is traversed up to the last level.