Facial recognition tech is already being used by big tech companies and the government. But Facebook is working on an experimental algorithm that doesn't even need a good shot of your face to pick you out in photos.
Human eyes can often easily pick out a person they know based on things other than their faces like their hairstyle, clothing, body type and pose, but machines are still struggling with that problem.
"There are a lot of cues we use. People have characteristic aspects, even if you look at them from the back," Yann LeCun, the head of Facebook's artificial intelligence lab told New Scientist. "For example, you can recognize Mark Zuckerberg very easily, because he always wears a gray T-shirt."
Facebook's research is trying to teach computers how to pick up on those same characteristics to identify people in pictures when they don't have a clear shot of a person's face.
The researchers took over 37,000 public images from Flickr -- some with faces visible, others without -- and tried to tell apart the nearly 2,400 people in them, according to a research paper released in January and presented at a computer vision conference in Boston earlier this month.
The researchers had a 47 percent success rate with Facebook's advanced facial recognition system, DeepFace, on a sample of 6,500 of those images. But when trying out the experimental algorithm designed to consider other factors on the same group, they were able to recognize individual identities 83 percent of the time.
The technology might someday be a powerful addition to the social network's image tagging feature, or be added in to the company's new mobile app, Moments, which scans through images on a user's camera to help identify people in them for more private sharing. But Facebook spokesperson Ari Entin did not directly address if the company is considering integrating the technology into its current products in response to a Post query on the subject. "This is basic, long-term research," he said over e-mail.
Privacy advocates say this type of photo identification technology could raise similar concerns to facial recognition technology. "Automated systems can increasingly identify you at a distance without you being involved with them in any way," said Alvaro Bedoya, the executive director of Georgetown Law's Center on Privacy & Technology. "My suspicion is that what you're going to start seeing is that faces are just the tip of the iceberg -- people are going to start IDing you from a distance by irises, the shape of your ears, or the way you walk."
Bedoya is one of a group of privacy advocates that recently withdrew from a government-sponsored process seeking to come up with a voluntary code of conduct for private companies using facial recognition technology over concerns that it would not result in rules that gave enough protection to consumers. The technology currently lacks specific regulation at the federal level.