The company, Oz Forensics, a developer of remote biometric identification systems and resident of the Skolkovo Foundation Information Technologies Cluster, took second place in the world for the accuracy of its facial recognition technology.
The Oz Forensics team researched how the scaling parameter (sphere radius) of the loss function affects the results of iterative training for the facial recognition system. One of the most famous publicly accessible tests in this field is the accuracy check model on the LFW (Labeled Faces in the Wild) dataset for verifying faces.
The test was created by the University of Massachusetts at Amherst and includes images of obviously poor quality - different lighting, no full-face, partially obscuration, a big age difference, etc. The full name of the test is “Unrestricted, Labeled Outside Data.” Aside from standard procedures for teaching computer vision systems, an important process is accurate facial detection and subsequent correct alignment.
Among these tested solutions, including commercial solutions, is the Oz Biometry technology from Oz Forensics showing results that are second in the world with an accuracy of 99.87%.
Artem Gerasimov, CEO of Oz Forensics:
“We use different tools to develop and perfect the technology. The successful combination of all the latest developments allows for great results in accuracy in the world’s independent tests. The current measure recognizes our biometric technology, Oz Biometry, as a quality product. From the business point of view, having of a solid grade from an independent testing center will have a positive effect on a client’s decision in favour of one vendor over another.”
Mikhail Styugin, head of the department of Information Security of the Skolkovo Foundation Information Technologies Cluster:
“The problem of remote biometric identification of users is more relevant now than ever before taking into account customers’ transition online and the internal business processes of companies. Oz Forensics previously showed brilliant results in detecting artificially modified documents and now in recognizing biometric “deep fake” attacks using deep learning neural networks. The expansion of the tech stack allows for the creation of new product solutions that are so important when taking into account the changes in the threat landscape in 2020.”