NIST Results Show Facial Recognition Accuracy Race Is Tightening

The latest results from the US National Institute of Standards and Technology’s Face Recognition Technology Evaluation (FRTE) 1:N (one-to-many) programme show that the gap between leading facial recognition algorithms is narrowing when measured against traditional accuracy benchmarks.

The FRTE programme assesses how effectively facial recognition systems can identify an individual by comparing a probe image against large databases of enrolled identities. Its results are widely used by governments, technology suppliers and system integrators to compare algorithm performance and understand how systems behave under different operational conditions.

As the strongest performing algorithms increasingly converge when their performance is measured against headline accuracy tests, differentiation is shifting towards performance in more challenging conditions. These include poor-quality or uncontrolled images, ageing, demographic variation and searches involving very large databases. Such factors are particularly important for border control and national identity systems, where operational images can differ significantly from controlled enrolment photographs.