Generating weights for biometric tokens in probabilistic matching systems

摘要

Generating weights for biometric tokens in probabilistic matching systems is disclosed, where these weights are generated from computations performed on matched sets and unmatched sets of a reference data set. In an embodiment, scores from a similarity scoring function are distributed among bins, and a weight is computed for each bin as the log of (the matched set ratio/the unmatched set ratio), where the ratios are computed as the number of scores in a particular bin as compared to the total size of the set. The weights may then be used subsequently with scores computed by the scoring function to assess confidence of a computed similarity score, and are directed toward making the output of the probabilistic matching system more data-driven and more accurate.

主权项

1. A method for generating weights for biometric tokens in probabilistic matching systems, comprising:
analyzing biometric tokens of a reference data set, the reference data set comprising a plurality of biometric tokens for each of a plurality of distinct entities, the reference set further comprising a matched set of the tokens and an unmatched set of the tokens, by performing a pair-wise comparison of the tokens in the matched set and of the tokens in the unmatched set using a similarity scoring function; determining a plurality of scoring bins, based on similarity scores computed by the analyzing, wherein an upper and a lower boundary of each of the scoring bins is selected for separating the similarity scores; computing, for each of the scoring bins, a weight for the scoring bin, the weight for each bin computed in view of how many of the similarity scores from the matched set fall into the bin and how many of the similarity scores from the unmatched set fall into the bin; and using the weights for assessing subsequently-computed similarity scores from the similarity scoring function.