Applied Recognition achieves new benchmark for face recognition accuracy

Applied Recognition, Inc. has achieved the 0.2 percent threshold for Cross-Over Error Rate (CER) in testing against the FERET dataset maintained by the US National Institute of Standards and Technology (NIST) for facial recognition systems evaluation.

The CER is observed when a technology is configured in a manner that its ‘false positive rate’ is equivalent to its ‘false rejection rate’, providing an accurate measure of face recognition technologies’ efficacy.

“This latest advance in accuracy will further broaden the number of applications where face recognition is the dominant choice to achieve a high degree of security while enhancing customer experience,” said Ray Ganong, co-CEO of Applied Recognition. “This new level of accuracy supports a completely automated workflow for financial transactions, such as opening a bank account or applying for a loan. Stringent identity verification requirements can be met to facilitate transactions that previously had to take place ‘face to face’. When combined with other methods of identity verification, such as credit checks and cell phone number verifications, the false authentication rate can be reduced to 1 in 10 million and without the need to route many bona fide applications through exception processes.”

In order to achieve this new level of performance, Applied Recognition upgraded its core face recognition methods to incorporate advanced machine learning techniques.

More specifically, residual networks are used to generate face vectors and face-to-face comparisons are processed using advanced statistical techniques.

Applied Recognition’s SDKs and the line of “Ver-ID” applications will all benefit from this update. For example, Ver-ID Credentials establishes a person’s identity by verifying the validity of a government issued photo ID then comparing the photo on the ID to an in-session “selfie”.

The company’s liveness-detection technology ensures that the individual who is being authenticated against the photo ID is actually the same individual requesting verification.

“We now offer our customers ‘the best of both worlds,” said Don Waugh, co-CEO of Applied Recognition. “Looking back, developers seeking high-accuracy face recognition needed to rely on cloud-services with all the attendant problems: privacy concerns, latency, and reliance on a persistent Internet connection. Our technology’s accuracy now substantially exceeds that of popular cloud-services while running directly on users’ mobile devices or personal computers at sub-second speeds.”

In May, Applied Recognition was awarded a patent for face detection and recognition technology that applies to advertising, entertainment and social networks, enabling companies to create personalized marketing messages.

Article Topics

Applied Recognition  |  biometrics  |  facial recognition  |  machine learning  |  NIST