The Facebook AI team have made a break through in modern face recognition technology with a 25% improvement over current state of the art systems.
Their DeepFace technology uses computers to match faces with almost human level performance. When tested on a random dataset DeepFace achieved a 97.25% match accuracy - humans achieve a 97.53% accuracy.
How does DeepFace work?
In modern face recognition the detection method consists of four stages:
The DeepFace team have improved both the alignment step and the representation step by using 3D face modeling to derive a face representation from a nine-layer deep neural network. This deep network involves more than 120 million parameters. You can read about it in detail via the Facebook website here.
How could DeepFace be used?
DeepFace could be used by Facebook for automatic face tagging in photos? (Although it did have to turn off previous automatic face recognition for all European users a couple of years ago by order of the Irish Data Protection Commission).
The benefits for security agencies and the police would be obvious. It could be used to automatically match a person's face from a CCTV image to their passport or driving license photo for example?
Fun uses could include online dating. You could use the technology to filter profiles based on appearance. For example, if you selected 5 people you found attractive the technology could be used to find people with a similar appearance.
What would you use it for?
Why not add your suggestions to the comment form below?...