FMR and FNMR
Definition
False Match Rate (probability that a non-matching pair is accepted as a match) and False Non-Match Rate (probability that a genuine matching pair is rejected); the two primary performance metrics for biometric and AI forensic matching systems.
- FMR
- False Match Rate: non-matching pair wrongly accepted as a match
- FNMR
- False Non-Match Rate: matching pair wrongly rejected
- Use
- Primary performance metrics for biometric and AI matching systems
- Relationship
- Adjusting the decision threshold trades one rate against the other
Common questions
Why can't a biometric system simply be tuned to minimise both FMR and FNMR at once?+
The two rates move in opposite directions as the matching threshold changes: making the system stricter to cut false matches raises false non-matches, and loosening it to reduce false non-matches raises false matches. A system is tuned to a specific operating point balancing both, chosen for the intended application.
Which error rate matters more in a forensic identification context versus an access-control context?+
In forensic identification, where wrongly linking a suspect to evidence carries severe consequences, a low FMR is usually prioritised even at the cost of more false non-matches requiring manual review. In consumer access control, where inconvenience matters more, a lower FNMR may be favoured instead.
Does a published low FMR for a matching algorithm mean it performs equally well on any evidence sample?+
No. Published error rates are typically measured on a specific test dataset and quality of input, and can be substantially worse on degraded, partial or low-quality forensic samples, so the reported rate should not be assumed to transfer directly to real casework conditions.
Related terms
- Convolutional Neural Network (CNN)
- A deep learning architecture that learns hierarchical spatial features from image data by passing the input through successive layers of learnable filters;...
- ENFSI AI Position Paper
- The 2023 European Network of Forensic Science Institutes policy document specifying that AI tools used in forensic casework must have published blind-test...
- EU AI Act
- The European Union's comprehensive AI regulation, fully effective from 2026. Classifies deep-fake generators used for impersonation as high-risk; requires that AI-generated content...
- Haemoglobin Species
- The forms of haemoglobin relevant to bloodstain ageing: oxyhaemoglobin (HbO2, fresh), deoxyhaemoglobin (Hb), methaemoglobin (Met-Hb), haemichrome, and haematin; each has a distinct...
- Hyperspectral Imaging
- An imaging technique that captures a full reflectance spectrum (typically hundreds of narrow wavelength bands from UV to near-IR) at every pixel...
- NAFIS
- National Automated Fingerprint Identification System: India's centralised fingerprint database operated by the NCRB, launched 2022, designed to aggregate state police and CFSL...
- Nitrogen-Vacancy (NV) Centre
- A point defect in diamond where a nitrogen atom is adjacent to a lattice vacancy; the electron spin of the NV centre...
- OSAC
- Organization of Scientific Area Committees for Forensic Science: established by NIST and DOJ in 2014, it develops consensus forensic standards through structured...
- SQUID
- Superconducting Quantum Interference Device; an ultra-sensitive magnetometer exploiting quantum flux quantisation in a superconducting loop; requires cryogenic cooling but can detect magnetic...
- Terahertz (THz) Radiation
- Electromagnetic radiation in the frequency range approximately 0.1 to 10 THz (wavelengths 30 micrometres to 3 mm), situated between microwave and mid-infrared;...
- THz Time-Domain Spectroscopy (THz-TDS)
- A technique using ultrashort laser pulses to generate and detect broadband THz pulses; provides both amplitude and phase spectra, enabling complex refractive-index...