The new engine incorporates significant improvements in the speed and accuracy of its system, already available across its entire product portfolio.
The National Institute of Standards and Technology (NIST) in the US has released the latest report on the 1:N evaluation of biometric facial recognition vendors. This test consists of finding a person within a database of identities. To do this, a facial image is compared against a gallery of millions of images and it is determined whether that person is within that database.
When performing these searches, there are two essential elements to take into account: the speed or agility in obtaining a result and the accuracy and reliability of the result itself. In most cases, greater accuracy will require a longer search time. However, Veridas’ engine has managed to improve both parameters, finding a perfect balance between security and customer experience.
This latest version is capable of performing a search in 322 milliseconds within a gallery of 1.6 million faces. At the same time, the search error for galleries of 12 million identities has been reduced from 15.41% to 7.53%.
This improvement places Veridas in the 8th position out of 62 engines in facial identification for 12 million galleries, ahead of other providers such as Aware (25.31% error), Cogent (21.19% error), HIK (21.29% error), Incode (16.19% error), Microsoft (14.82% error), Cognitec (11.03% error), Yitu (10.57% error) o Visionlabs (9.20% error).
All these improvements are already in production environments for all Veridas products and services for digital identity verification and fraud detection.
Leading and certified technology
Veridas has recently become the only company to be included in NIST’s evaluations for facial recognition (both 1:1 and 1:N) and speaker recognition (1:1). In addition, Veridas has passed ISO 30.107 iBeta Level 1 PAD (Presentation Attack Detection) test for its Liveness Detection solution and also holds ISO 27.001 on the Security of Information Systems accredited by Bureau Veritas.
NIST, the institution of reference in the field of biometrics
Results shown from NIST do not constitute an endorsement of any particular system, product, service, or company by NIST.
What is the difference between verification (1:1) and identification (1:N)?
There are two ways to perform a biometric comparison:
- 1:1 or verification: A photo or audio is compared against another existing evidence (e.g. the photo or audio from the enrollment). Depending on the similarity score it is determined if it is the same person.
- 1:N or identification: A photo or audio is compared again, but unlike the previous case, it is done against a database containing many other photos/audios. That is, it is compared against each and every one of the pieces of evidence present in that database and it is determined if that person is within that group.