10.3 Operating threshold specification

Before · SP 800-76-1

After · SP 800-76-2

Operating threshold specification

Before · SP 800-76-1

After · SP 800-76-2

The threshold applied to scores from the biometric comparison algorithms shall be set to achieve false match rates at or below the respective values in Table 16 . The threshold shall be calibrated in tests conformant to Annex A 27 . Agencies may require lower (more secure) FMR values; particularly some implementations can attain lower false match rates.

Table 16 – Maximum allowed false match rates by modality Modality Authentication False match rate Notes Fingerprint minutia matching Off-card 0.001 Applies to one comparison with one finger. See NOTE 1 and Section 10.5 Fingerprint minutia matching On-card 0.001 Iris image matching Off-card 0.0001 Applies to one comparison with one eye. See NOTE 1 and Section 10.5 . See NOTE 2 Face image matching Off-card 0.001 Applies to one comparison. See NOTE 1 and Section 10.5 NOTE 1 Transactional false accept rates will be higher than these values if the transaction allows presentation of more than one sample (e.g., a second finger) or biometrics (e.g., iris and face).

23 Those clauses, 4.5.3 and 4.5.4 for off-card fingerprint comparison, and 5.7.3.2 and 5.7.4 for on card comparison, qualify components requiring core minutiae-based interoperable accuracy. They do this in laboratory tests. The accuracy criteria were never intended to be adopted as operational verification criteria – particularly the FMR = 0.01 threshold was instituted to bar non- interoperable minutia detection algorithms and matchers – but is not appropriate as an Agency security policy. 24 This represents the case where a lost card is found by someone who casually attempts a biometric authentication. 25 For fingerprints and face, industry convention is for recognition algorithms to produce similarity scores, for which higher thresholds produce fewer false matches. For iris, the convention is to produce distance or dissimilarity scores, for which lower thresholds produce fewer false matches. 26 As required by ISO/IEC 19795 biometric testing standards, test reports almost universally show this tradeoff using detection-error tradeoff characteristics (DETs) or almost equivalently receiver operating characteristics (ROCs). 27 Threshold calibration information is available under NIST's MINEX and IREX programs. NOTE 2 The iris false match rate is considerably lower because iris is readily capable of achieving this lower more secure value without appreciable increases in false non-match rates [IREX-III]