Exploring the Methodology and Utility of Standardized Latent Fingerprint Matcher Scoring

ii Acknowledgements The authors would like to thank Michael Garris and Elham Tabassi for their many valuable suggestions that went into this report. Abstract Automated searches of fingerprints against a repository/database are important tools of the forensic community. Systems performing these searches are referred to as Automated Fingerprint Identification Systems (AFISs). The output of an AFIS is a fairly small set of prospective candidates with attendant matching scores (allowing for comparison of candidates, and frequently referred to as comparison scores). These scores provide an indication of how likely a particular candidate is a true mate of the search fingerprint (i.e., originates from the same individual). One difficulty in interpreting matching scores in usage is there is no accepted standard for its range and exact meaning (other than " bigger is better "). Experts need to become very familiar with the scoring of a specific system to make optimal use of the results. This report proposes that we standardize the scoring system. The standardized score becomes a number between 0 and 100 and carries two decimal places, for a total of four significant figures. Seven alternative algorithms for computing matching scores are outlined in this report, ranging from very simple to quite complex. A mid-complexity algorithm is then selected for detailed illustration/development. This report also provides a detailed analysis of scoring produced by the matchers previously tested by NIST.