The algorithm of identification of the persons designed in the company «vokord» was the best in the test MegaFace
The Russian company «vokord» or VOCORD was founded in 1999 and specializing in the development of intelligent video surveillance and audio registration, created the most efficient algorithm for the identification of persons in the world. This is evidenced by the results of testing using a database MegaFace.
Test base MegaFace includes a million photographs, more than 690 thousand different people taken under natural conditions in arbitrary positions, with the free face, under different lighting and exposure settings. The competing algorithms are evaluated according to ability of face recognition with the increasing number of «distracting officials» (from 10 to 1 million).
Algorithm recognition VOCORD implemented by means of deep neural networks, have shown the best results for the identification of individuals with rank 1. Test conditions prescribe the application to choose the most similar to the standard image of a human face out of a million test photos. The algorithm correctly recognized 75,127% which is a record value. For comparison: development of a Google — FaceNet v8 shows the indicator 70,496%.
The developers note that the algorithm is already used in the production of VOCORD company. In addition to high precision, its advantages include high productivity, allowing to perform per second to three million comparisons.
Sources: MegaFace, «VOCORD»