Improving identification accuracy on low resolution and poor quality iris images using an artificial neural network–based matching metric

摘要

The iris is currently believed to be one of the most accu-nrate biometrics for human identification. The majority of fielded irisnidentification systems use fractional Hamming distance to comparena new feature template to a stored database. Fractional Hammingndistance is extremely fast, but mathematically weights all regions ofnthe iris equally. Research has shown that different regions of the irisncontain varying levels of discriminatory information when using cir-ncular boundary assumptions. This research evaluates four statisticalnmetrics for accuracy improvements on low resolution and poor qual-nity images. Each metric statistically weights iris regions in an attemptnto use the iris information in a more intelligent manner. A similaritynmetric extracted from the output stage of an artificial neural net-nwork demonstrated the most promise. Experiments were performednusing occluded, subsampled, and motion blurred images from thenCASIA, University of Bath, and ICE 2005 databases. The neuralnnetwork-based metric improved accuracy at nearly every operatingnpoin