UK academics introduce improved neural network for highly precise facial recognition
Researchers from the University of Surrey’s Centre for Vision, Speech, and Signal Processing have been working on a more precise deep neural network to increase the accuracy of facial recognition algorithms, so they deliver less errors, writes media outlet Verdict. Dubbed OSNet, the technology was presented at the International Conference on Computer Vision in Seoul, South Korea.
The researchers claim OSNet has “outperformed many popular identification systems already in use,” and only needs 2.2 million parameters, less than traditional deep neural network models or the ResNet50 infrastructure that requires 24 million parameters, currently integrated in many facial recognition systems.
OSNet can identify other small details such as logos and types of clothes worn, to perform re-identification of people in different lighting and environments, a potentially critical capability for facial recognition systems.
“With OSNet, we set out to develop a tool that can overcome many of the person re-identification issues that other set-ups face – but the results far exceeded our expectations. The re-identification (ReID) accuracy achieved by OSNet has clearly surpassed that of human operators,” says Tao Xiang, Distinguished Professor of Computer Vision and Machine Learning at CVSSP.
“OSNet not only shows that it’s capable of outperforming its counterparts on many re-identification problems, but the results are such that we believe it could be used as a stand-alone visual recognition technology in its own right.”
Some facial recognition systems have been less accurate in matching people with darker skin, and improving the accuracy of facial recognition algorithms generally requires improving the training data set. A Google contractor was recently accused of misleading subjects in a biometric data collection project that specifically targeted dark-skinned homeless people in Atlanta to improve the detection algorithm and compatibility. Google’s name recently popped up in another scandal related to the use of a biometric facial recognition database that included photos of minors to train facial recognition algorithms for improved accuracy.
The U.S. National Institute of Standards and Technology claims, however, that the failure rate declined from 4 percent in 2014 to 0.2 percent in 2018.
In 2018, psychologists from the UK and Australia said facial recognition accuracy using images from poor quality CCTV footage can be improved by combining them with images from other footage in a computer-enhanced “face averaging” composite.