Face Engine Performance

Ayonix engineers are working on performance improvements both in speed and accuracy.
As far as we consider the accuracy, face engines of other companies on small databases such as 1000 people enrollment makes no much difference in indoor environments. Difference occurs when increasing the face database size. For example, enrolling 1million records makes algorithm difficult to separate people since each feature becomes near to each other.
Being aware of these facts, Ayonix continues to develop its technology which results in the release of more advanced algorithms.

Performance Measurement

1Comparison with Ayonix’s new-developed algorithm and its current algorithm
2Comparison with Ayonix’s new-developed algorithm and other vendor technologies
3Conducting above tests with different benchmark databases
For benchmark databases, Ayonix uses FERET Face database, FRGC Database, Yale Face database, Arface database and Ayonix’s database.

Here is a Database content comparison between Feret and Ayonix DB

Feret DB has almost frontal faces and images are taken in controlled environments.

Ayonix DB has almost non frontal images collected from Internet randomly.

Ayonix DB contains sunglass images, face expression images, various makeup images, different resolutions, different lights, shadows, noises and all problems of outdoor and indoor environments.

Here is the performance difference between Ayonix and other vendors


Ayonix has participated to NIST 2014(National Institute of Standards and Technologies) face recognition tests.

Here is the performance difference between Ayonix and other vendors


In NIST 2014 tests, Ayonix has been the second fastest company by recognizing 1.7 million faces in one second.

Right now, Ayonix has improved further by doubling the speed two times. Now, Ayonix recognizes 3.4million faces in one second.