10.5 Agency consideration of false rejection performance

Before · SP 800-76-1

After · SP 800-76-2

Agency consideration of false rejection performance

Before · SP 800-76-1

After · SP 800-76-2

An authentication transaction may involve several core comparisons each of which will be expected to have failure rates given by false match rates (FMR, for impostors) and false non-match rates (FNMR, for genuine comparisons). These are matching error rates defined over outcomes of sample comparisons . Operational authentication performance is quantified in terms of both the false reject rate (FRR) and the false accept rate (FAR) which are defined over outcomes of transactions 28 : In PIV, FRR is the proportion of legitimate cardholders incorrectly denied access; FAR would be the proportion of impostors incorrectly allowed access. The error rates depend on a number of factors including: the environment, the number of attempts (e.g., finger placements on the sensor), the sensor itself, the quality of the PIV Card templates' parent images, the number of fingerprints or irises invoked, and the familiarity

28 A transaction might include several comparisons from repeated presentations of multiple fingers or irises. of users with the process. The use of two fingers or irises in all authentication transactions offers substantially improved performance over single-instance authentication. This document does not establish false rejection performance criteria – how often genuine users are unable to successfully authenticate – because it does not represent a direct security objective. Agencies are cautioned that false rejection performance is operationally vital in access control applications and is achieved by using high performance cameras and algorithms, by ensuring good quality enrollment, by correct control of the environment, by adherence to enrollment specifications, by subject and operator instruction, and by subject habituation. Agencies are therefore strongly encouraged to consider: ― Establishing a policy on how many times a subject can attempt to authenticate; ― Establishing false rejection accuracy criteria against which tests and qualification procedures can be conducted; ― Referring to false rejection performance measures reported for algorithms evaluated using the IREX test and calibration procedure; ― Referring to false rejection performance measures reported for algorithms conforming to the MINEX test and calibration procedure; ― Training staff to recognize poor quality images at time of enrollment; ― Requiring the use of multiple samples (e.g., two fingers); ― Conditions under which an alternative modality for authentication (e.g., iris instead of fingerprint) might be used; ― Conditions under which an additional modality for authentication (e.g., iris and fingerprint) might be used; ― Conducting their own supplementary tests. These might be performance tests of single products or interoperability tests, and might be used to estimate application-specific performance. The execution of tests conforming to one or more parts of the ISO/IEC 19795 standard is strongly recommended because biometric testing is a specialized discipline. Particularly a number of subtleties and difficulties exist that can potentially fatally undermine a test. This specification does not: ― Preclude agencies from establishing more stringent false match criteria. The false match criteria can always be met by setting a high (i.e., stringent) comparison threshold. However, more stringent thresholds imply elevated false rejection errors because of the error-rate tradeoff. One mitigation is to use two fingers or two eyes.