- Email: [email protected]
Face ID tech can see through your disguise
In this section n Hottest place on Earth’s surface, page 8 n Rats replay scary memories in their sleep, page 9 n How clean energy will make China a superpower, page 20
DITCH the hat and scarf – face recognition software can now identify you despite such disguise. Amarjot Singh at the University of Cambridge and his team trained a machine learning algorithm to locate the 14 key points on the face that our brains pay most attention to when we look at someone. It only needs to see a fraction of these points to guess where the others are likely to be. The researchers then showed the system 2000 photos of people wearing hats, glasses, scarves and fake beards, hand labelling them to indicate the location of those key points, even if they couldn’t be seen. Finally, the algorithm was given a subset of the images to learn how disguised faces corresponded to the same faces without any disguise. It was able to accurately identify people wearing basic disguises like a cap and scarf 69 per cent of the time (arxiv.org/abs/1708.09317). This isn’t as good as systems that recognise undisguised faces, but the algorithm is better at seeing through disguises, says Singh. “In effect, it is able to see through your mask.” You can also probably say goodbye to CV Dazzle, the camouflage make-up mooted as a way to stay anonymous in a world of face recognition. “This will work very well for this type of camouflage,” Singh says. The team will present its findings next month at the International Conference on Computer Vision in Venice, Italy. The system could help identify criminals who are trying to hide their identities, says Singh. But he admits that authoritarian governments could also use it to identify protesters. “There’s always a trade-off between security and privacy,” says Anil Jain at Michigan State University. But he says that people in public spaces are already under constant surveillance by security cameras, so they shouldn’t be too worried about every improvement in the technology. Matt Reynolds n
JENTZPHOTO /ALAMY STOCK PHOTO
Face ID tech can see through your disguise
–Mind the gap–
NYC subway runs best with quantum maths WITH its antiquated trains, rusty arrive more or less randomly. rails and straphangers who keep “If you were waiting at a stop the doors from closing, the New for 5 minutes, waiting for the York City subway system could next 5 minutes does you no hardly be described as efficient. good,” says Trogdon. In a more And yet, some trains arrive with functional transit system, you’d a reliable regularity, following a expect that after waiting for a neat statistical model similar to while, the probability of a train that seen in quantum systems. arriving soon would be quite high. Aukosh Jagannath at the The Poisson distribution does University of Toronto, Canada, not guarantee this. and Tom Trogdon at the “I think the data is confirming University of California, Irvine, people’s intuition about the two used the subway system’s real“The southbound 1 train on time data feeds to analyse gaps between arrival times on two lines. the west side of Manhattan follows more efficient They found that the southbound 1 line that runs down random matrix patterns” the west side of Manhattan shows what are called random matrix lines,” says Trogdon. The 1 line patterns, which are “a sign of is one of the three local subway greater efficiency”, says lines serving the west side of Jagannath, who is now at Harvard Manhattan, so it’s far less crowded University. These trains run at than the 6, which at the time of more regular intervals (Physical the study was the only local line Review E, doi.org/cczj). on the east side. In contrast, the 6 line that runs The efficiency analysis was up the east side of Manhattan is inspired by a landmark 1990 inefficient. Its trains follow the study in Cuernavaca, Mexico. At Poisson distribution, a statistical that time, buses in Cuernavaca model describing particles that operated with no central
controlling agency, and each bus belonged to the driver. To maximise the number of passengers they could transport – and therefore profit – the drivers set up a series of checkpoints to avoid clustering. Upon arriving at a checkpoint, the driver would learn when the previous bus had stopped, and would slow down or speed up to optimise gaps between vehicles. Analysing the records of when buses came and went, researchers found that the buses in Cuernavaca obeyed random matrix patterns. The parallel isn’t exact for the New York City subway system, however. The random matrix patterns break down at the last 10 stations of the southbound 1 line. What’s more, the northbound 1 line does not follow those patterns. “The analysis of the New York system is less clear [than for Cuernavaca],” says Ariel Amir at Harvard University. Still, he says this kind of analysis is the first step towards optimising the subway system. For the commuters who take more than 1.7 billion rides on New York’s subterranean rails a year, that’s always going to be a plus. Mark Kim n 16 September 2017 | NewScientist | 7
DITCH the hat and scarf – face recognition software can now identify you despite such disguise. Amarjot Singh at the University of Cambridge and his team trained a machine learning algorithm to locate the 14 key points on the face that our brains pay most attention to when we look at someone. It only needs to see a fraction of these points to guess where the others are likely to be. The researchers then showed the system 2000 photos of people wearing hats, glasses, scarves and fake beards, hand labelling them to indicate the location of those key points, even if they couldn’t be seen. Finally, the algorithm was given a subset of the images to learn how disguised faces corresponded to the same faces without any disguise. It was able to accurately identify people wearing basic disguises like a cap and scarf 69 per cent of the time (arxiv.org/abs/1708.09317). This isn’t as good as systems that recognise undisguised faces, but the algorithm is better at seeing through disguises, says Singh. “In effect, it is able to see through your mask.” You can also probably say goodbye to CV Dazzle, the camouflage make-up mooted as a way to stay anonymous in a world of face recognition. “This will work very well for this type of camouflage,” Singh says. The team will present its findings next month at the International Conference on Computer Vision in Venice, Italy. The system could help identify criminals who are trying to hide their identities, says Singh. But he admits that authoritarian governments could also use it to identify protesters. “There’s always a trade-off between security and privacy,” says Anil Jain at Michigan State University. But he says that people in public spaces are already under constant surveillance by security cameras, so they shouldn’t be too worried about every improvement in the technology. Matt Reynolds n
JENTZPHOTO /ALAMY STOCK PHOTO
Face ID tech can see through your disguise
–Mind the gap–
NYC subway runs best with quantum maths WITH its antiquated trains, rusty arrive more or less randomly. rails and straphangers who keep “If you were waiting at a stop the doors from closing, the New for 5 minutes, waiting for the York City subway system could next 5 minutes does you no hardly be described as efficient. good,” says Trogdon. In a more And yet, some trains arrive with functional transit system, you’d a reliable regularity, following a expect that after waiting for a neat statistical model similar to while, the probability of a train that seen in quantum systems. arriving soon would be quite high. Aukosh Jagannath at the The Poisson distribution does University of Toronto, Canada, not guarantee this. and Tom Trogdon at the “I think the data is confirming University of California, Irvine, people’s intuition about the two used the subway system’s real“The southbound 1 train on time data feeds to analyse gaps between arrival times on two lines. the west side of Manhattan follows more efficient They found that the southbound 1 line that runs down random matrix patterns” the west side of Manhattan shows what are called random matrix lines,” says Trogdon. The 1 line patterns, which are “a sign of is one of the three local subway greater efficiency”, says lines serving the west side of Jagannath, who is now at Harvard Manhattan, so it’s far less crowded University. These trains run at than the 6, which at the time of more regular intervals (Physical the study was the only local line Review E, doi.org/cczj). on the east side. In contrast, the 6 line that runs The efficiency analysis was up the east side of Manhattan is inspired by a landmark 1990 inefficient. Its trains follow the study in Cuernavaca, Mexico. At Poisson distribution, a statistical that time, buses in Cuernavaca model describing particles that operated with no central
controlling agency, and each bus belonged to the driver. To maximise the number of passengers they could transport – and therefore profit – the drivers set up a series of checkpoints to avoid clustering. Upon arriving at a checkpoint, the driver would learn when the previous bus had stopped, and would slow down or speed up to optimise gaps between vehicles. Analysing the records of when buses came and went, researchers found that the buses in Cuernavaca obeyed random matrix patterns. The parallel isn’t exact for the New York City subway system, however. The random matrix patterns break down at the last 10 stations of the southbound 1 line. What’s more, the northbound 1 line does not follow those patterns. “The analysis of the New York system is less clear [than for Cuernavaca],” says Ariel Amir at Harvard University. Still, he says this kind of analysis is the first step towards optimising the subway system. For the commuters who take more than 1.7 billion rides on New York’s subterranean rails a year, that’s always going to be a plus. Mark Kim n 16 September 2017 | NewScientist | 7