Candidate List Reduction based on the Analysis of Fingerprint Indexing Scores

This paper proposes new candidate list reduction criteria for fingerprint indexing approaches. The basic idea is that, given a query fingerprint, the initial set of scores produced by an indexer could contain useful information to reduce the candidate list. Novel reduction criteria have been proposed, and extensive experiments have been carried out over five publicly available benchmarks, using two state-of-the-art fingerprint indexing techniques. Although quite simple, the proposed criteria achieved remarkable results, allowing a substantial reduction of the candidate list: for instance, at 1% error rate, the average penetration rate of a state-of-the-art minutiae-based indexer decreases from 27% to 3.9% on FVC2000 DB2. The new reduction criteria are applicable to any indexing approach, since they only require a list of scores as input.