The concept of using cameras to recognize people is not new, the technology, through the use of CCTV has been around for decades. However, the rise in sophisticated facial recognition platforms, which make use of artificial intelligence, has triggered new concerns. The technology is sufficiently advanced that, in theory at least, people can be individually identified wherever they go. As our lives have become increasingly digital and online, humanity has been leaving data trails everywhere they click or swipe, particularly via social media and through the use of smartphones.
Here every time you upload a photograph or video to a social media platform, the in-built facial recognition systems learn more about you. As these proprietary algorithms ingest data about you are, plus the place you were in together with he people you know, the more powerful they become.
To partially address the growing concern about privacy, University of Toronto Faculty of Applied Science & Engineering researchers have created an algorithm that can dynamically disrupt facial recognition systems, helping to stop a social media platform from identifying the individual(s) in an image.
According to lead researcher Professor Parham Aarabi: “”Personal privacy is a real issue as facial recognition becomes better and better. This is one way in which beneficial anti-facial-recognition systems can combat that ability.”
The new algorithm relies on a deep learning technique termed adversarial training. This process places two artificial intelligence algorithms up against one other. In tests the researchers have shown what happens when two neural networks compete. Here one was set to identify faces and the second acted to disrupt the facial recognition task of the first. The counter algorithm was shown to work, and it creates an Instagram-like filter that can be applied to photographs to protect individual privacy. The system is powerful enough to disrupts image-based search, plus feature identification, emotion and ethnicity estimation, and any other face-based attributes that can be extracted by artificial intelligence.
The research will be presented at the IEEE International Workshop on Multimedia Signal Processing, in July 2018.
