Exploring Quality Scores for Workload Reduction in Biometric Identification

Nowadays, the deployment of biometric identification systems operating on large-scale databases is considerably increasing. Hence, biometric systems need efficient processing, i.e. indexing queries, that allow reducing the computational workload (number of comparisons) for a single biometric identification transaction. In this context, approaches for computational work-load reduction can take advantage of the inherently more stable properties derived from the biometric character. Such character can refer to attributes (e.g. scars, droppy eyelids) that impact the biometric sample quality indirectly, and cannot be controlled during the biometric acquisition process. Therefore, samples exhibiting different characters lead to a variation in terms of sample quality.In this work, we investigate whether the quality scores resulting from different sample quality assessment methods can be suitable for computational workload reduction (indexing). We propose a nearest quality score-based intelligent search for indexing in biometric identification. The experimental evaluation is conducted on databases of three biometric characteristics, i.e. face, iris, and fingerprint. Experimental results report a significant computational workload reduction with respect to the exhaustive search scenario – down to 38%, 29%, and 31% over face, fingerprint, and iris, respectively.