TripleA: Accelerated accuracy-preserving alignment for iris-codes

摘要

The discriminative power of the iris enables reliable biometric recognition on large-scale databases where a rapid comparison of biometric reference data is essential to limit response times. In case of national-sized databases a one-to-many comparison might still represent a bottleneck of a biometric identification system, in particular if numerous relative tilt angles have to be considered in the comparisons stage. While a compensation of head tilts improves the robustness of an iris recognition system, extensive feature alignment increases the probability of a false match as well as comparison time. In this paper we present a novel method to accelerate iris biometric comparators in an accuracy-preserving way. Emphasis is put on the alignment of iris biometric reference data, i.e. iris-codes. Based on an analysis of the nature of iris-codes and comparison scores between them we propose an efficient two-step alignment process referred to as TripleA. This scheme, which can be operated in various modes, significantly reduces the amount of relative tilt angles to be considered during iris-code comparisons. Hence, comparison time as well as the probability of a false match are reduced at the same time. In an experimental evaluation on the Casia v4-Interval iris database we achieve a more than fourfold speed-up in the comparison stage maintaining biometric performance using different feature extraction techniques.